Article: www.gamersnexus.net/guides/3251-raven-ridge-soc-voltage-guidelines-how-to-kill-cpu-with-safe-voltage How to Kill Your Motherboard: ruclips.net/video/vLMd-5yxTAc/видео.html What is LLC? ruclips.net/video/NMIh8dTdJwI/видео.html
So high voltage can be dangerous, and as you name only voltages(and i can see that's pretty universal in pc hardware) i assume what vrms driving soc outputs constant current? If no, why do we care about voltage but not about current?
@Gamers Nexus So my question was partially answered with the LLC video with buildzoid, but I'm still unsure. The short version is simply, since voltage reporting is mostly highly accurate "estimations", how possible is it that the reported cpuz voltage is higher than the actual voltage, the opposite case of which you're showing here. Reason I ask is I have a R7 1700 under Arctic's 240mm aio, and was playing around with the OC seeing how far I could push it. At one point I had cpuz reporting 1.51v and yet prime95 only got the cpu to 75c (this wasn't an everyday clock, I turned it way down since, calm down comment section). Since this performance is much higher than I've expected compared to multiple other reported overclocks, I'm curious if that 1.51v figure reported by cpuz is accurate or not, and by how much.
This is a below standard quality from what I've been used to from GN, all I've seen in this video is wild speculation that you may be putting more voltage into your CPU than you actually are without knowing what sensors (and where they're placed and how they work) are reporting what voltages to what software and then comparing that with non-representative measurements taken from the socket (because socket voltages aren't necessarily the same as what actually goes into the CPU due to series resistance and even though one of the probes was actually on a molex and not in the socket. If you had 6v going in and 5v going out of the cpu the cpu would have 1v, but your molex-socket measurement would report 6v, I'm not saying the measurement is necessarily incorrect of gives necessarily incorrect results but it does introduce variables that could be easily removed or controlled for.).
I dare say that the MB manufacturers ARE responsible, they should do everything in their power to tune the boards so they report correct numbers. At the very least make sure that you don't underreport it, that's a recipe for disaster. Which incidentally is why I never recommend Gigabyte to novice builders... Hey Gigabyte, go buy your engineers some multimeters ok? (Well the other guys as well of course...) Edit: I was intending to add something but I can't remember what it was. Oh yeah, the state of bios support for the APU's are still quite rough, that the MSI board got non functioning GPU voltage control right now is probably just that. Let's hope they improve on it on current boards and don't gloss over it to push sales of the new chipset boards that are coming...
Fartonaut resistance varys from board to chip though even. So getting it exactly accurate or llc perfectly on the nose would add significant cost due to the fact it would take a hardware monitor similar to volt metter except it's inlined with Internal wiring which ya also would have to either run it to a standout point and hope a user would properly ground it to the standoff when they install it in a case or have two wires coming off it. One to be taped on where ya wanna get x voltage reading and one for ground via molex or some other way. This will not work and adds extra cables. It's best to just use a volt metter and take off the side panel of your case. Easy peasy, what they should do is clarify such things as Steve has mentioned since overclocking seems to be championed (I'm unsure if that's too dramatic or not?)
That's simply absurd. You're basically asking that every MoBo have a high accuracy voltmeter built in. Go price such. Would you be willing to pay that premium on every MoBo? They use reasonably accurate techniques that are sufficient for 99% of end users needs. If you need more accurate readings then get a good voltmeter and take direct readings.
Except... the motherboard is reporting the correct voltages... the CPU isn't. You can see it in the video if you're familiar with the sensors on HWinfo64. The motherboard VCORE SOC sensor, which is cut off to read VCORE... is showing the correct voltages.
That is not the right way to measure the core voltage !!! You are not only measuring the core voltage but also adding in all the voltage drop on the cables from the motherboard to the PSU. Might not sound like much but there is easily 20A flowing back trough the ground cables (12V 5V 3.3V rails combined) and there can be easily 0.01 Ohm of resistance in the cables this can cause an extra 0.2V to appear (Amps*Ohms=Volts) The correct way to do it is to touch the other multimeter lead to a grounded pin ON the actual motherboard, ideally touch it to the other side of the capacitor you are measuring. If you repeat your measurements like that you will see a difference.
Agreed. And also, the recommendations are surely based on the board's own measured values, not externally measured values on a multimeter. I'd have to see long term failure rates to determine if these risks are accurate.
Absolutely right, thanks Asus for the probe points that do it right! In regard to the comment by Casey Goode, the voltages quoted are ABSOLUTE voltages, not "read by software or MB", absolute. The meaning is not ambiguous, the chip manufacturer cannot account for the different MB constructors voltages, they can only quote what their own equipment will take. If you have a board that delivers 200 mV over target, well, take it up with the MB manufacturer. What is FAR more important... over/under shoot. How far over and how far under will do serious damage over time.
berni8k You’re right. With those low voltages every little bit added counts. Wondering if he has a BSEE. Nothing against him if he doesn’t. Just want to see what passes for engineering these days.
CPU are made to last for a couple of decades when running within specs, even with mild voltage degradation their useful lifespan won't be affected. Most of the time the motherboard is more likely to die before the CPU does.
This is why I like top ROG boards. Voltage contacts on the front of the board and cleaner/more consistent power delivery. It would be interesting to see a board like the Apex and the most basic Z370 board. It would be interesting to add this metric to mobo reviews. Although, this doesn't really matter for discrete GPUs since all modern ones are voltage capped and don't perform better with higher voltage since clocks depend more on temps.
paskowitz when I built my Ryzen build around cyber Monday deals last year the reasons you stated about Asus ROG boards is why I went with the Asus ROG X370 Hero board. Before that build all the ROG components I had or used in builds for friends and family were absolute beasts and were parts I could rely on!!
I must say, the Z270 Maximus IX Apex is the best. So excessive for sure, but also for what it was built for, I don't think it can be beat. I also have the z270e strix board, and to be fair, if you want to OC, the heat sinks are decent on the VRM, and the VRM works great, with most of the options that other ROG boards have.
Yeah I love my ROG RVE10 and I'll always buy high end ROG boards for all my builds, the only thing to be careful of is that you make sure you go in an manually set ALL voltages as on the auto settings ESPECIALLY with x.m.p enabled, these boards do push a lot of voltage and there are lots of reports of these auto setttings killing cpus.
I've been taking CPU vcore measurements out of the socket since Ryzen launched (Prime X370 Pro) While they vary compared to HWinfo some say that there could be a drop from the back of the socket to the actual die. So things have to be taken with a grain of salt. I generally recommend not to go over 1.425v for a Ryzen chip. SOC never over 1.25 and generally is better low (here is at 0.975v) If you wanna do a truly helpful video grab an oscilloscope and try to find peaks at LLC4 / LLC5. There is a huge debate around Ryzen about this.
There is a debate whether LLC4 and LLC5 are harmful or not and to determine that we need to know if there are peaks while transitioning from load to idle. So far only one person measured and found no peaks but we need a definite confirmation. For that Gamers Nexus needs a high quality oscilloscope. (maybe rented) I'm on a 1800x at 4.05Ghz 1.418v vcore LLC4, 0.975 v SOC LLC30
If the peaks are short enough to need a good oscilloscope to see it (I take it >100MHz) then they don't matter because there are extra caps on the CPU package that can clean up blips in voltage that brief. The problem is that he is stabling the black lead of the multimeter into the PSU, this adds the voltage drop of the PSU cables in and throws off the reading by as much as over 100mV. Doing that is fine for testing if your 12V rail drops. But its NOT fine when you are mesuring voltages so tiny.
Yea almost kill my CPU until I was informed about this. Running my 2700X at what I thought was 1.75V but the MoBo was actually pushing closer to 1.4V. Realized I had followed bad advice and 1.25V (which is pushing closer to 1.35V) is enough voltage to sustain a stable 4.2GHz overclock. To many people in forums and online in general recommend just whacking up the voltage to "stabilize" your overclock.
You should take the mainboard ground as negative reference, not the power supply cable. There is a voltage drop across the power cables so the voltage read will be higher than actual Vcore voltage. Or even better measure directly on a capacitor of the respective voltage.
Excellent point. When I was trying to scope out my motherboard I did clip the ground to a grounded piece of metal on the motherboard, but I was only checking the outputs of the VRMs for stability.
Sorry, but I disagree with you, HWmonitor, in the CPU section read 1.1V from the Sensor in the CPU, below, in the mobo section (CPU Vcore... SOC, Sensor in the mobo) read 1.368V, VRMs Output, equal of the measurement from the multimeter. The difference between both values is the Vdrop, compensate by the LLC setting in the BIOS. Values taken in the minute 12:58 . (Sorry my English)
I had boards that were set to 1.48v and had a hard time to let the sensors hit 1.42 idle/1.40 under load Got better after some vdrop pencil mods. Agree that the important voltage is those actually hitting the chip, but voltage sensors are just as inaccurate as temp sensors.
True, not to mention using a cheap meter and can the motherboard.Noticed in my Msi bios the voltage is 0.01 higher then whatever voltage you set it to.
So he read vddcr CPU voltage with the multimeter and mixed them up? is he sure that he is taking soc voltage measurement? if he showed the CPU Voltage and show thats not it ( a higher value... Over there). We need a captain
Where you measure these voltages is imperative. There are resistive losses between the voltage source and the device being powered. I believe the reading in hwinfo is coming from the ADCs on the cpu itself. Therefore it's getting a more accurate reading of voltage that arrives at the silicon. This is why all the hwinfo voltage readings are lower than what your meter is reporting. There's lots of variables that affect measurement, but the main issue is the difficulty in accurately measuring voltages in the millivolt range.
Gamers Nexus hah, you're right. I didn't watch the video all the way through. Gamers get this stuff wrong so often that I just assumed you weren't aware of these factors. You've done your research. You're no Linus. ;)
I think there is a lot of confusion about just saying "this voltage will kill your CPU" because the truth is that the answer really depends on thermals (and if you're skilled/knowledgeable about overclocking ie: you don't try to use 1.5v+). I'm not saying it is impossible to to kill a CPU as I realize that it is possible however I disagree that it is "easy" to do. I think there is honestly way too much "Myth" about "CPU degradation" over time even with CPUs as old as 45nm @ 1.4x volts. If you know what you are doing and have the proper cooling for high voltage (at least an AIO with a decent CPU block) it is actually difficult to harm a CPU with voltage less than 1.5v. This topic has been beaten to death long before RUclips. The only thing left is for it to be proven in a controlled test as there's not much room left for words at this point (it's just a really old debate). I would love to see a controlled test with various identical systems running 24/7 with different voltages applied along with air cooled vs water cooled with a minimum ONE YEAR of running the systems. Run an extensive suite of tests with the systems including overclocking and then run all the same tests again at the 1 year mark. Keep it going for 2 years.. 3 years.. let's prove this myth one way or the other once and for all. GN might be the one to get it done :) IMO, the CPUs themselves, if cooled properly with less than 1.5x, will be just fine and will work forever.. I am someone that has applied 1.4x volts to many many CPUs over the years (always been an overclocker) and ran them for YEARS at a time and never ever saw any of this "mythical degradation".
You're speaking out your ass. It is no myth, and as junctions get smaller and smaller the problem WILL inherently get worse and WORSE. Which is likely the real reason you're seeing "old" CPUs doing just fine today at near the same frequencies. Voltage does impact transistor junction breakdown (thermal breakdown is not NEARLY as much a problem) and as opposed to thermal meta-stable changes, voltage breakdown causes permanent stable changes, aka. permanent shorting. A lot of it is the chip lottery because the doping process is not "perfect" for better or worse. Not only that, but a lot of VRMs have difficulty maintaining "stable" high voltages, where the margin for error is not linear as you increase it, and you can easily over-volt up to hundreds of millivolts higher than intended when overvolting, very potential breaching breakdown of the dielectric doping...and thus you have permanently damaged the chip. Will it not work at all anymore? No. The breakdown usually isn't that catastrophic. If the VRM is very stable and you're under 1.4V? Most CPUs will be ok. Unstable VRM and above 1.4V? You're asking to ruin it...no matter your cooling. Can you ever overclock that high with that voltage again? Will it get worse over time now until it IS catastrophic? Have you shortened it's lifespan from 10 years to 2? YES. Your cooler isn't a superconductor, even running at sub zero temps overclock extremers are ruining the chips, but for that sport longevity is inherently not the point. Besides, on the whole heat and voltage is bad. But voltage is markedly many times worse and riskier.
euhm. the correct voltages the mobo puts out are in the motherboard section of HWinfo (it shows 1.2 when you first pointed it out at CPU VCORE SOC.) the CPU shows lower numbers because of internal resistance I reckon. so that just shows the vdroop between what the MOBO delivers and what is actually used by the chip.
So when monitoring that SOC on my Ryzen 1700 is at max 1.2v, I'm correct at looking at the "CPU VCORE SOC" under mobo readings in HWInfo? This video made me second guess myself by checking the CPU readings of "SoC Voltage (SVI2 TFN)" which reads at 1.09v.
I have the Gigabyte AB350M Gaming 3 (the mATX variant of your mobo). My avg voltage difference between the two is around the same at 0.048v, and max voltage differences 0.094v. Granted not a lot, but to certain users, could be enough between sweet spot and degradation.
One thing that I would point out just for new users to also understand is that the meter they use could also be several % off from what it is reading/displaying and that accuracy varies by range. An inexpensive hardware store meter could be out as much as half a volt or possibly more. This is especially common for cheap meters that cost less than $250-500 US and don't come with a lab calibration sticker or statement of calibration from an independent lab. The lowest precision lab calibrated meter I would consider using for overclocking would have a 4.5 digit display and 20K counts but a 5.5 digit display and 200K counts would still be a better option for seeing problematic voltage variations during operation.
Had a mild yet annoying audio stuttering problem with my new Lenovo, the support team tried to help but couldn't, however we did find in the user forums some discussions that led me to conclude the problem might be an undersized power supply, 400w not nearly enough for a gaming desktop. (No wonder it was such a bargain!) What made this more credible is that I read numerous solutions from people that only occasionally worked. When I disabled my WLAN I found the audio stutter I was fighting diminished by about 90% further making me conclude that this is a power issue.
Kyle Van small scale mining operations fronted by individuals will die out once cryptocurrency and blockchain become so widely accepted that "casual" individuals will be using it. Once the general populations begin using something scams become more widespread and soon enough federal regulations are placed upon it. Regulations that can track usage of coin and will eventually find a way to track individual transactions then the big original advocates of it drop off and then it becomes an everyday kind of thing that is no longer feasibly "mined" on a GPU array due to exponential growth, taxation, and regulation. Blockchain and cryptocurrency are logical evolutions, with evolution come changes.
Keith Berwanger you seriously think that's happening in the next few months? I have a hard time visualizing major changes before the next halving. No one really cares what's going to happen in 2025. People are going to be buying things based on this year. Bitmain owns such a crazy amount of crypto they by themselves can fluctuate the market and accelerate or reduce demand. I don't honestly understand why so many miners are frothy over the video cards anyway. You can get the T9 and S9 all day long and you don't have to worry about multiple motherboards or PC setups to run video cards
I don't /seriously/ project anything happening. I do /seriously/ think anything can happen with such a volatile concept. I won't pretend to know the nuances of it all, but anything that can grow exponentially can mutate exponentially.
Thank you for this video. People were going crazy over the internet because it was so easy to get 1600 mhz with 1.25v. At least 1 channel recognize this and explain why is particularly dangerous to believe in software to measure voltages.
This is some fantastic content, this will be coming in handy today as my Apu just arrived pretty crazy the differences between software and the actual reading did not expect that huge of a jump but will keep that in mind going forward thanks guys!
That isn't the correct probing method to get accurate CPU voltage: it doesn't account for voltage droop across the ground plane between the VRM and CPU, possibly some more due to droop on wiring between the PSU and motherboard. Your readings may be off by 100mV simply from taking your ground from so far away from the load.
On all the boards? I'd expect that to vary a fair amount between cheaper and more expensive boards with a different number of ground planes, different plane thickness and possibly split planes. BTW, when chip manufacturers specify supply voltages, that's usually directly at the chip package's pins/pads. The closest you can get to measuring that would be to solder jumpers to one of the capacitors either under the IHS or under the CPU unless the motherboard is re-purposing a pair of VCC-GND pins for VRM feedback to read package voltage.
my old Ryzen 3600x was at 1.4v on for time before I wanted to upgrade. Never missed a beat. I gave it to my mate cheap and his still using it on the same board with same OC still going strong.
Excellent Video Steve. You illustrate clearly why Overclockers should not trust the voltage readings that come from Software Apps. The excessive fluctuations in the A-to-D converters used in the hardware on the board. I know I would like to see, at the very least, a standard adopted where all Mobo Manufacturers supply overclocking motherboards with the SOC Voltage and Ground brought to a clearly labeled Hard Post on the top side of the board. In addition the more diligent manufacturer should supply a precision Reference Standard set to the CPU Manufacturers "never go above" SOC Voltage. Good quality DMMs run anywhere from 80 to 200 bucks. That isn't a great deal of extra money to spend considering the user will only have to do it once.
Great stuff as usual, Steve! You have to be THE MOST TRUSTWORTHY GUY ON THE NET. Your level of information far outstrips most people out there. Your dedicated, and accurate, and you have great hair! Keep up the great work, Matt
In general, other than getting a higher benchmarking score it's usually not worth the extra voltage or higher frequency to run your components at and the cooler the better as long as it's not affecting your overall experience using the computer. I try to run my computer as quiet and as cool as possible I do test out how hard and far and extreme I can get it but for 24/7 use my realistic settings are much more conservative. This whole video definitely confirms my ideology when it comes to building and running one's PC for benchmarking versus daily use. As always Steve, great work!
Great job gettin in on the details. Interesting the variance between manufacturers and from different measurement points.. I can't help but wonder if there is some sort of unspoken industry standard, or if they really just sort of all do their own thing. Kinda makes you wonder which voltage is really the important one, because even if you are seeing a certain voltage on the fet if there is loss on the way to the sensitive component then I would think that the loss would have to be kept in consideration and compensated for.. But it seems like they all act so differently its almost impossible to say..
Reminds me of my old P35 board where I went with a pencil mod to counter the immense VDroop woth a quad. Still find it crazy when people put something like 1.35-1.4v onto recent chips. That was considered high to threshold 10 years ago and todays chips are manufactured way smaller.
If you deeply know about mosfet drive controls and inverters.Because after the constant drive voltages and frequencis you can increase the source voltage on mosfet because the source voltages and drive voltage on mosfets are not that interlinken with each others.But finally i want to say that these things are easy if you are able to recircuit and resolder your whole motherboard.
You really should connect ground on the board and not through the PSU. Technically they are the same, but that wire length is going to make the readings off.
Haha thats exactly what i just did to my asshat *cough* asus cheap prime b350 plus with my ryzen 7 1700 and my friend accidently let ryzen master kill his tomahawk ab350 pro with ryzen 1700 so we have to do it again. And just shoot the shit with them and dont say you overclocked it. If anything blame a bios update for changing voltage lulz
Either HWMonitor is mislabeling stuff or you are misreading it. HWM was almost exactly matching your multimeter readings, if you looked at the CPU VCORE voltage. Take a look at 14:53. VCORE voltage: 1.284, measured voltage: 1.28-ish.
I haven't watched the whole video yet, but by using a disconnected Molex ground line as the reference, you're adding in the voltage droop across the other ground lines that are connected to the board which are carrying some current. This happens because every wire has some resistance, and there's current passing through them to get to power supply ground, so they generate a voltage. V = I * R. This won't be much (probably less than 1/10 of a volt), but it can add to it. If you want to measure more accurately you should probe as close to the CPU as possible.
Not surprised this is happening, many people on forums have been advising me to run 1.4v as 24/7 use on my new coffee lake, and you got specialist companies selling pre binned chips that are rated at really high voltages like 1.37-1.42v which is crazy for daily use. Even worse when you consider that chips actual voltage tends to be higher than whats set in bios and what software tools report. There seems to be a big misconception that as long as you can cool a high volted chip then its all good, as if only heat kills cpus.
one thing to point out is due to the distance between that molex gnd and the vcore pin the value could be slightly off due to the molex gnd not actually being 0V
I think you are looking at the wrong number. There is a stat for CPU VCORE SOC down there, which I think measure from the board instead of the CPU. It gave a better read and closer to the real voltage
Take the crappiest B350 board and set the SOC to 1.2V and v-core to 1.425V and keep the CPU cool and i bet you can break the VRM and then break the CPU. I say this before i watch this video. I love gamer nexus they really do great tests and i'm happy they do it
Might, sure but most decent power supplies would turn off the moment your VRMs burn out. Cant guarantee that your cpu wont get damaged in that time though. High voltages roast it really fast.
He says a lot of things but I waited and waited to hear data or even an explanation about why high voltage actually kills a chip and.. like every other neckbeard, thats where the details get suspiciously thin.
So, as someone who has a B350 Gigabyte Ryzen board. I always read off the Motherboard's reported VCORE SOC voltage in HWinfo64. Which you have set to obscure the SoC denomination in your sensors bar. It's always a lot different than the CPU reported. But it's being fairly close to what you're measuring on the back. THIS is something I wish you'd talk about more. It's something I always warn people on the subreddit on the B350 Gigabyte boards. Read the motherboard's voltage, not the CPU's voltage, because the offsets in BIOS will also stack on top of the voltage you set in Ryzen Master. Basically: You should note you're reading the CPU reports, and not the motherboard reports. The motherboard here is actually reporting the correct voltage on that gigabyte board.
Thank you so much for making this video Steve! I've been telling my cousin for over a year now, don't trust the software, its usually not accurate. So many people believe it like its written in stone, even the impossible (like sub-ambient temps on FX, just physically impossible), while I'm laughing at the 1 million volts on one of my GPUs show. Always check with a multi-meter to know for sure, never assume what you type is accurate with what happens. Great video Steve, keep up the awesome work!
The HWiNFO64 software is accurate, this video proves that. Steve used a digital multymeter to cross check and validate the readings. Fo example if you set your OC software at ~1.1 volts, check the HWiNFO64 motherboard section to see if the voltage on those components exceeds the safety numbers the manufacturer provided
Ok, so 1 million volts was correct... And 8c CPU in a 20c room is also correct. And even though I measured with a multi-meter, and it showed Vdroop of 0.01v, software was showing 0.1v (order of magnitude different), as well as 1.3v instead of 1.55v like I typed and was reading with DMM, then you're saying software was right the whole time? Mind blown! I just invented a CPU that runs cooler than its surroundings (so it must be absorbing thermal energy, and creating electricity), while capable of mining. I just solved global warming and mining at the same time! Where do I sign up for my Nobel Peace prize?
But how can I? I just invented something better than perpetual motion! I'm off to save the world, providing energy for 3rd world countries, Hash power for miners, and solving global warming. Its a miracle!
I uhh... I'm pretty sure people are missing the point of this video. But so am I. GN is talking about what people recommend as a safe voltage. What does it mean? This isn't commonly communicated as where its measured, just what the voltage is. Which doesn't really tell you anything. All the same, I do not think that the voltage readout on the back of the board has anything to do with components past the socket. Those parts report their voltage via software, and should always do so accurately. But of course I am wrong, as voltage droop across the socket is extra heat generated. This is something to consider, as voltage is only one of two factors in lethality. _But its quite easy to check core temps_ The other X-factor are VRM temps, which are going to go up according not to what the socket uses, but to what the VRM puts into the socket. Why worry about this at all? GN I think you need to explain how the CPU can incorrectly report its operating voltage, or if it isn't doing that, why is this dangerous?
I put 1.45v (ASUS maximus V extreme Bios)through a 2600k and ran it @5.1ghz 24/7 on my main rig for over 5 years With a maximum Core Temperature of 60°c(realtemp) without issue at which point I sold the rig on to one very happy camper. I think a key point you are missing here is that a "Safe voltage" can only be given in conjunction with operating temperature and intended life cycle. by "Safe" I mean CPU will maintain OC performance as well as stock operation for the entirety of its useful life ~ 6years IMHO. As I understand it degradation is a wear rate and occurs regardless but is accelerated by increased voltage and temperature. In any event I intend to push similarly high voltage through my 3700x @ a max 60°c core. And would be interested in collaborating to help to establish a degradation curve for the following individual factors. "Voltage" "temperature" "Node Size in nm" Conventional wisdom would suggest die shrinks are more susceptible to degradation as its a Physical Erosion caused by Electromigration In other words do you want to roast some 3700x's like chestnuts in time for Christmas it will be fun and it will benefit the community
Although I've since owned a q6600 and 6700k, my first self built pc was an AMD Athlon XP 3200+ @2.2ghz. back then I didn't know how to overclock and I never did in the time I owned that but when I bought my q6600 and sold the AMD to a friend, he fried it within 2 weeks, just ramping up voltages because I told him how good the q6600 was. I kinda wish I kept it for sentimental value
Great video. I've got an Asus B350-I with an 1700 in a Phanteks Shift case (gonna need to print a top extraction fan mount at one point). After seeing the thermals on the board. I'm sticking to 3.75 even with a 120 Coolermaster AiO. Higher I could probably put it horizontal and grill sausages. Thankfully it's only purpose is for HTPC and as a rendering slave for v-ray distributed when needed (cheaper to have a few 1700 than less Threadripper systems). Next up is a 1700 NAS system hahaha. Can't wait for kill the board series. I need a video so I can shut up and just show people :P
What about a voltage drop on a path from cpu to ground? I don't know how much resistance in between them, but they may affect your results. Could you find a pin that located at exiting current from cpu and make your results a more precise? Ah, and also voltage drop from resistance of a socket.
Can you do a video on transient voltage spikes when using high LLC? For the new z390 gigabyte boards there is no know info about how much the voltage spikes when coming off load with high LLC. Im running 1.345V on my 8086K for 5.1 Ghz, but I am afraid that my voltage will spike +0.200mv to 1.645 for a millisecond when coming off load. That will slowly kill my cpu in a year.
Hi man, I'm just curious. Did you check the voltage difference between the 8 pin ground and the mesurments capasitor ground? the difference is still so much that there is over volt, but if the ground points is (because the resistance in the motherboard) higher than the 8 pin ground, the cpu will see lower voltage. It might be marginal, but like I said, just curious All the best and thanks for the great work!
So how does this compare for Intel? It’s pretty common to set voltages as high as 1.350 for core voltage if your temps are okay. You usually can’t even achieve the expected overclock without going at least 1.3
Have you checked that the voltage you read in the socket with the multimeter does not pass the voltage limiter( MOSFET) before feeding the cpu? The voltage limiter may limit the excess voltage that enters the socket. Of course, measuring in hardware is much more reliable than software, but if you go through a voltage limiter the difference may not be that big. As you know these little details can make all the difference. Love guys, Keep it RAW!
Hey GN I'm calling bs on that video because HWInfo was reading the correct value for the gigabyte board. At 13:45 it's saying 1.38V for CPU Vcore and after that you're measuring 1.375V. Seems completely accurate to me. I dunno what the SoC Voltage it shows is but my guess is it's as useless as the VID reading. I've always used CPU Vcore in the motherboard section, from what a lot of people told me that is the number you wanna read.
My Msi Mpg x570 Atx gaming plus when left at auto pumps Over 1.4v while my Ryzen 7 3700x runs at Stock 🧐 i manually set it to 1.35v and use Ryzen master to OverClock my Chip to 4.4ghz @ 1.35v = Stable on 90% of my Triple A games 🤔 A C Odyssey seems to Crash after 1 hour of gameplay ...... any Recommendations are appropriate & Ty 🇵🇹
Thank you Steve, you are doing a great job at explaining things. I watched your video in a discord channel and had to go to the browser so I could upvote. Currently I'm not doing any OC except for XMP on RAM. I pushed 1.36V in BIOS to make it more stable. I even had to push 1.38V on a friend's rig because it only POSTed at 1.36V, but it could still be unstable. I hope it is "safe".
i would have thought that the chip readings would be the more accurate voltage. Where you're reading is before the contacts and there could be more resistance between the socket and the back of the silicon chip compared to where the chip is reading. that doesnt change the fact voltage is still higher than the setpoint, but it might not be as bad as you are seeing.
Using a molex connector ground pin doesn't account for voltage drop across traces on the PCB. It's a bit more fiddly, but you should measure the voltage across both sides of that SMD capacitor. You will probably find that it's not as high as 1.4V.
OctoMan PC's 3.9 1.32 on my 1600. At stock I have even seen the auto voltage setting spike to 1.48. Asus mb, same thing kept happening to my z170 asus rog board.
1.2 Vcore ? If that is correct I need your address to arrest you for: "way more luck in the silicon lottery than me". I can barely get my 8700K stable at 1.35 V Edit: word
could be the mb or even psu. while you might just be REALLY unlucky you can usually get an 8700k stable on 1.28v or less at 4.8ghz on adaptive and a little bit of LLC. vdroop is a killer of stable OC's. long term stability on mine came at 1.232v @ 4.8, 6hrs prime 95, no problem. with a hero x MB and seasonic titanium PSU. you can limit short and long term wattage to keep temps under control if you're on air. Lastly it could be possible that you have a motherboard that just does not report voltage correctly. I have used the Asus Hero on the past 3 builds I've done simply for predictability. I know what it's going to do and I know the voltages are pretty close to accurate. As you can see in this video some motherboards don't do that
Somehow I missed this video. I have i7 7700K with a Gigabyte Z270X-UD5 motherboard and have the Vcore set at 1.235 and onboard graphics set to 0.600v. I do not use the IGP, I have a MSI 980Ti Lightning. These voltages seem to be fine as I have had this build for just under 1 1/2 years.
Cant they put an integrated multimeter for critical parts? I would like to know the physical reason about this difference, why motherboard cant read nor write correct voltages.
Any proof for the degrading? Just because you measure 1.35V on socket, doesn't mean that there is same voltage in the DRAM controller. I'd personally trust internal sensors over external measurement.
I think this is something the MOBO manufacturers are responsible for and should be on top of. If people can't rely on the information they get from their MOBO they can make a lot more damaging mistakes which i would say are out of their control. Of cause their will probably be some margin of error but if it can be the difference between a safe and definitely unsafe voltage the difference is simply too big also if you are gonna have a "big" difference at least be on the safe side specifically referring to Gigabyte.
It would seem to me that 1.2v is a good empirical (middle of the road) value for SoC in that there so much variance between manufacture architecture, and electronic component quality. That said, t would seem to me that the absolute or max SoC voltage allowable is a need to know requirement - would that be 1.35 volts? Question: Curious to know, to stabilize SoC voltage fluctuation, did your testing include changing LLC - I saw from your excellent presentation that there was quite a bit of volt flux?
Im Running my I5 8600k on 5.1Ghz 4500Mhz on my Cache @ 1.440v. Let´s see what happens in 1 Year or 1 and a Half.... (Cpu is Delidded Stable @ 80-85°C @ 30minute stress testing depending on the room Temperature) In Gaming it wont go further then 65-70°C. Im using a Broken Eco cooler..... Motherboard is a MSI Z370-A-Pro With the latest Bios (MS-7B48) Maybe another Mainboard will give me lowervoltage on the same settings. But it is not worth an 160€ upgrade.... My Old FX8350 runned stable @ 5Ghz with 1.45v for over 1 and a half Years. And it is still working.
Isn't there a considerable vdrop on that molex ground vs. ground near the cpu? I will answer, yes it is. You should check that whats the difference. Also what is your multimeter tolerance? It looks like 0,05% accuracy which in 2 volt meassurment means 0,1V. After what I've seen, I think that readings are way off.
I have a question, many OCers just raise the speed only and leave the voltage at "stock" or auto setting- so how accurate/safe is the stock/auto option regarding actual delivered voltage? It seems with these apus you can get 1.4 or so more speed, but only 1.2 x in performance (the square root). It may not be worth the last 100-200 mhz if you need to raise voltage for stability with the really unknown actual output sent to the socket. Not everyone is going to pull mobo and measure things- I think there might be a shorting danger with the small and crowded under-socket area.
Is this the scenario with only apus or is applicable to ryzen cpus aswell? If so , how ryzen behaves to it? Same variance or is it less as theres no igpu.
Question, had my i5 10600k At 5 gig adaptive offset 0.050@ 88 degrees was highest temp on core 2. I used MSI try it now on ram, left xmp, try it now on my vengeance RGB pro 3200 DDR to 3400, rebooted system said ram couldn't clock, reset Mobo to default. Put xmp back out same settings on over clock now when I run test core 2 has hit 97. What gives, and my voltage is auto, but looks higher than before? Did that ram oc screw up cpu potential? Please let me know if you can
Why would you not want Load Line Calibration to be set to its max? I'm not sure why its even a setting in Bios. Wouldn't delivered load voltage be closer inline to configured and reduce the guess work?
could you make a video of how accurate are the z370 mainboards are messuring the vcore of the cpu? would be very nice. i do not have a multimeter to check that. important are the models of asrock (taichi and fatality itx) and asus maximus x hero and apex.
i've got 2 questions 1 how low would you asume mx cpu will get if i direcht die cool it with a pleter witch is cooled by a 65mm quad rasiator but there are dual 1080ti's on the same loop and 2 at those temperatues what is a save core voltage (cpu=i5 8600k)
I don't doubt that there are differences between the voltage reading in the settings and the "actual" voltage. But I doubt whether the voltage meter is calibrated or not, and when this was done.
Article: www.gamersnexus.net/guides/3251-raven-ridge-soc-voltage-guidelines-how-to-kill-cpu-with-safe-voltage
How to Kill Your Motherboard: ruclips.net/video/vLMd-5yxTAc/видео.html
What is LLC? ruclips.net/video/NMIh8dTdJwI/видео.html
So high voltage can be dangerous, and as you name only voltages(and i can see that's pretty universal in pc hardware) i assume what vrms driving soc outputs constant current? If no, why do we care about voltage but not about current?
Use the killed cpu as a box opener
@Gamers Nexus So my question was partially answered with the LLC video with buildzoid, but I'm still unsure. The short version is simply, since voltage reporting is mostly highly accurate "estimations", how possible is it that the reported cpuz voltage is higher than the actual voltage, the opposite case of which you're showing here.
Reason I ask is I have a R7 1700 under Arctic's 240mm aio, and was playing around with the OC seeing how far I could push it. At one point I had cpuz reporting 1.51v and yet prime95 only got the cpu to 75c (this wasn't an everyday clock, I turned it way down since, calm down comment section). Since this performance is much higher than I've expected compared to multiple other reported overclocks, I'm curious if that 1.51v figure reported by cpuz is accurate or not, and by how much.
This is a below standard quality from what I've been used to from GN, all I've seen in this video is wild speculation that you may be putting more voltage into your CPU than you actually are without knowing what sensors (and where they're placed and how they work) are reporting what voltages to what software and then comparing that with non-representative measurements taken from the socket (because socket voltages aren't necessarily the same as what actually goes into the CPU due to series resistance and even though one of the probes was actually on a molex and not in the socket. If you had 6v going in and 5v going out of the cpu the cpu would have 1v, but your molex-socket measurement would report 6v, I'm not saying the measurement is necessarily incorrect of gives necessarily incorrect results but it does introduce variables that could be easily removed or controlled for.).
very good video. especially after most YT reviews showing their overclocking procedure setting both soc and gfx voltages to 1.35V :-o
So... I shouldn't round it up to 2V?
Unless it's a 486, no.
5v is ok too, just dont go further ;)
lol 10v is pro gamer mode 😎
85v is safe for celerons. They deserve that voltage.
Set VCore to 220v in BIOS and will enable FUTURE mode in BIOS. Make sure your CPU has a flux capacitor and a cooler compatible with TDP 1.21 gigawatt.
The speedometer on my car shows "50" when I'm really driving 75 mph. That's not a fault, just a behaviour.
you should buy a new gauge motor or whatever they are called unless its digital then it might be the instrument cluster thats bad
@@bizzy156246 I'm pretty sure they were joking...
I dare say that the MB manufacturers ARE responsible, they should do everything in their power to tune the boards so they report correct numbers. At the very least make sure that you don't underreport it, that's a recipe for disaster. Which incidentally is why I never recommend Gigabyte to novice builders... Hey Gigabyte, go buy your engineers some multimeters ok? (Well the other guys as well of course...)
Edit: I was intending to add something but I can't remember what it was. Oh yeah, the state of bios support for the APU's are still quite rough, that the MSI board got non functioning GPU voltage control right now is probably just that. Let's hope they improve on it on current boards and don't gloss over it to push sales of the new chipset boards that are coming...
D Bucky You should be able to expect them to be somewhat accurate.
Fartonaut resistance varys from board to chip though even. So getting it exactly accurate or llc perfectly on the nose would add significant cost due to the fact it would take a hardware monitor similar to volt metter except it's inlined with Internal wiring which ya also would have to either run it to a standout point and hope a user would properly ground it to the standoff when they install it in a case or have two wires coming off it. One to be taped on where ya wanna get x voltage reading and one for ground via molex or some other way. This will not work and adds extra cables. It's best to just use a volt metter and take off the side panel of your case. Easy peasy, what they should do is clarify such things as Steve has mentioned since overclocking seems to be championed (I'm unsure if that's too dramatic or not?)
That's simply absurd. You're basically asking that every MoBo have a high accuracy voltmeter built in. Go price such. Would you be willing to pay that premium on every MoBo? They use reasonably accurate techniques that are sufficient for 99% of end users needs. If you need more accurate readings then get a good voltmeter and take direct readings.
They should be more or less be around the correct voltage (most are what +/- 5%) not 0.1v higher then what you entered
Except... the motherboard is reporting the correct voltages... the CPU isn't. You can see it in the video if you're familiar with the sensors on HWinfo64. The motherboard VCORE SOC sensor, which is cut off to read VCORE... is showing the correct voltages.
Oh shit *runs to downclock his IntelHD*
FeelsBadMan
That IntelHD Graphics is going to overpower your monitor and cause it to detonate.
That is not the right way to measure the core voltage !!!
You are not only measuring the core voltage but also adding in all the voltage drop on the cables from the motherboard to the PSU. Might not sound like much but there is easily 20A flowing back trough the ground cables (12V 5V 3.3V rails combined) and there can be easily 0.01 Ohm of resistance in the cables this can cause an extra 0.2V to appear (Amps*Ohms=Volts)
The correct way to do it is to touch the other multimeter lead to a grounded pin ON the actual motherboard, ideally touch it to the other side of the capacitor you are measuring.
If you repeat your measurements like that you will see a difference.
Agreed. And also, the recommendations are surely based on the board's own measured values, not externally measured values on a multimeter.
I'd have to see long term failure rates to determine if these risks are accurate.
@@thequantaleaper Yeah this way of measuring didn't sit right with me either. Plus the guidelines are there and they take everything into account.
The correct way is to connect the power inside the CPU to two unused pins (has to be done by AMD), which don't draw any significant current,
Absolutely right, thanks Asus for the probe points that do it right!
In regard to the comment by Casey Goode, the voltages quoted are ABSOLUTE voltages, not "read by software or MB", absolute. The meaning is not ambiguous, the chip manufacturer cannot account for the different MB constructors voltages, they can only quote what their own equipment will take.
If you have a board that delivers 200 mV over target, well, take it up with the MB manufacturer.
What is FAR more important... over/under shoot. How far over and how far under will do serious damage over time.
berni8k You’re right. With those low voltages every little bit added counts. Wondering if he has a BSEE. Nothing against him if he doesn’t. Just want to see what passes for engineering these days.
Still waiting on my 4690K to die its been running 1.51V for over 3 years.
If it's cool, it'll last quite a while.
Yea its delidded with LM tim on a custom loop.
No degrading?
Haven't had to do anything its been fine. I push all my hardware to the limit. Need an excuse for upgrades :P
CPU are made to last for a couple of decades when running within specs, even with mild voltage degradation their useful lifespan won't be affected. Most of the time the motherboard is more likely to die before the CPU does.
This is why I like top ROG boards. Voltage contacts on the front of the board and cleaner/more consistent power delivery. It would be interesting to see a board like the Apex and the most basic Z370 board. It would be interesting to add this metric to mobo reviews.
Although, this doesn't really matter for discrete GPUs since all modern ones are voltage capped and don't perform better with higher voltage since clocks depend more on temps.
paskowitz when I built my Ryzen build around cyber Monday deals last year the reasons you stated about Asus ROG boards is why I went with the Asus ROG X370 Hero board. Before that build all the ROG components I had or used in builds for friends and family were absolute beasts and were parts I could rely on!!
I must say, the Z270 Maximus IX Apex is the best. So excessive for sure, but also for what it was built for, I don't think it can be beat. I also have the z270e strix board, and to be fair, if you want to OC, the heat sinks are decent on the VRM, and the VRM works great, with most of the options that other ROG boards have.
Yeah I love my ROG RVE10 and I'll always buy high end ROG boards for all my builds, the only thing to be careful of is that you make sure you go in an manually set ALL voltages as on the auto settings ESPECIALLY with x.m.p enabled, these boards do push a lot of voltage and there are lots of reports of these auto setttings killing cpus.
This is the kind of video you would expect to see on a enthusiast/technical channel. But only Gamers Nexus does it.
Thank you Steve! ♥
I've been taking CPU vcore measurements out of the socket since Ryzen launched (Prime X370 Pro)
While they vary compared to HWinfo some say that there could be a drop from the back of the socket to the actual die. So things have to be taken with a grain of salt.
I generally recommend not to go over 1.425v for a Ryzen chip. SOC never over 1.25 and generally is better low (here is at 0.975v)
If you wanna do a truly helpful video grab an oscilloscope and try to find peaks at LLC4 / LLC5. There is a huge debate around Ryzen about this.
There is a debate whether LLC4 and LLC5 are harmful or not and to determine that we need to know if there are peaks while transitioning from load to idle. So far only one person measured and found no peaks but we need a definite confirmation. For that Gamers Nexus needs a high quality oscilloscope. (maybe rented)
I'm on a 1800x at 4.05Ghz 1.418v vcore LLC4, 0.975 v SOC LLC30
If the peaks are short enough to need a good oscilloscope to see it (I take it >100MHz) then they don't matter because there are extra caps on the CPU package that can clean up blips in voltage that brief.
The problem is that he is stabling the black lead of the multimeter into the PSU, this adds the voltage drop of the PSU cables in and throws off the reading by as much as over 100mV. Doing that is fine for testing if your 12V rail drops. But its NOT fine when you are mesuring voltages so tiny.
yeah, you really need to sense ground near the cpu, preferably on the other side of the cap
when there are triple digit amps flowing milliohms matters
Yea almost kill my CPU until I was informed about this. Running my 2700X at what I thought was 1.75V but the MoBo was actually pushing closer to 1.4V. Realized I had followed bad advice and 1.25V (which is pushing closer to 1.35V) is enough voltage to sustain a stable 4.2GHz overclock. To many people in forums and online in general recommend just whacking up the voltage to "stabilize" your overclock.
You should take the mainboard ground as negative reference, not the power supply cable. There is a voltage drop across the power cables so the voltage read will be higher than actual Vcore voltage. Or even better measure directly on a capacitor of the respective voltage.
Excellent point. When I was trying to scope out my motherboard I did clip the ground to a grounded piece of metal on the motherboard, but I was only checking the outputs of the VRMs for stability.
Sorry, but I disagree with you, HWmonitor, in the CPU section read 1.1V from the Sensor in the CPU, below, in the mobo section (CPU Vcore... SOC, Sensor in the mobo) read 1.368V, VRMs Output, equal of the measurement from the multimeter. The difference between both values is the Vdrop, compensate by the LLC setting in the BIOS. Values taken in the minute 12:58 . (Sorry my English)
So we should focus on mobo section "CPU VCore SOC", correct? (Mine is at max 1.2v so I think I'm good).
I had boards that were set to 1.48v and had a hard time to let the sensors hit 1.42 idle/1.40 under load
Got better after some vdrop pencil mods.
Agree that the important voltage is those actually hitting the chip, but voltage sensors are just as inaccurate as temp sensors.
True, not to mention using a cheap meter and can the motherboard.Noticed in my Msi bios the voltage is 0.01 higher then whatever voltage you set it to.
Don't really know if they make those, but a multi do uses on
So he read vddcr CPU voltage with the multimeter and mixed them up?
is he sure that he is taking soc voltage measurement? if he showed the CPU Voltage and show thats not it ( a higher value... Over there). We need a captain
Where you measure these voltages is imperative. There are resistive losses between the voltage source and the device being powered. I believe the reading in hwinfo is coming from the ADCs on the cpu itself. Therefore it's getting a more accurate reading of voltage that arrives at the silicon. This is why all the hwinfo voltage readings are lower than what your meter is reporting. There's lots of variables that affect measurement, but the main issue is the difficulty in accurately measuring voltages in the millivolt range.
trahim2 That was the entire point of the content, yes.
Gamers Nexus hah, you're right. I didn't watch the video all the way through. Gamers get this stuff wrong so often that I just assumed you weren't aware of these factors. You've done your research. You're no Linus. ;)
trahim2 So basically hwinfo's reading is accurate ? That's a relief
trahim2 He's not Linus
I think there is a lot of confusion about just saying "this voltage will kill your CPU" because the truth is that the answer really depends on thermals (and if you're skilled/knowledgeable about overclocking ie: you don't try to use 1.5v+). I'm not saying it is impossible to to kill a CPU as I realize that it is possible however I disagree that it is "easy" to do. I think there is honestly way too much "Myth" about "CPU degradation" over time even with CPUs as old as 45nm @ 1.4x volts. If you know what you are doing and have the proper cooling for high voltage (at least an AIO with a decent CPU block) it is actually difficult to harm a CPU with voltage less than 1.5v. This topic has been beaten to death long before RUclips. The only thing left is for it to be proven in a controlled test as there's not much room left for words at this point (it's just a really old debate). I would love to see a controlled test with various identical systems running 24/7 with different voltages applied along with air cooled vs water cooled with a minimum ONE YEAR of running the systems. Run an extensive suite of tests with the systems including overclocking and then run all the same tests again at the 1 year mark. Keep it going for 2 years.. 3 years.. let's prove this myth one way or the other once and for all. GN might be the one to get it done :)
IMO, the CPUs themselves, if cooled properly with less than 1.5x, will be just fine and will work forever.. I am someone that has applied 1.4x volts to many many CPUs over the years (always been an overclocker) and ran them for YEARS at a time and never ever saw any of this "mythical degradation".
You're speaking out your ass. It is no myth, and as junctions get smaller and smaller the problem WILL inherently get worse and WORSE. Which is likely the real reason you're seeing "old" CPUs doing just fine today at near the same frequencies.
Voltage does impact transistor junction breakdown (thermal breakdown is not NEARLY as much a problem) and as opposed to thermal meta-stable changes, voltage breakdown causes permanent stable changes, aka. permanent shorting. A lot of it is the chip lottery because the doping process is not "perfect" for better or worse.
Not only that, but a lot of VRMs have difficulty maintaining "stable" high voltages, where the margin for error is not linear as you increase it, and you can easily over-volt up to hundreds of millivolts higher than intended when overvolting, very potential breaching breakdown of the dielectric doping...and thus you have permanently damaged the chip.
Will it not work at all anymore? No. The breakdown usually isn't that catastrophic.
If the VRM is very stable and you're under 1.4V? Most CPUs will be ok.
Unstable VRM and above 1.4V? You're asking to ruin it...no matter your cooling.
Can you ever overclock that high with that voltage again? Will it get worse over time now until it IS catastrophic? Have you shortened it's lifespan from 10 years to 2? YES.
Your cooler isn't a superconductor, even running at sub zero temps overclock extremers are ruining the chips, but for that sport longevity is inherently not the point.
Besides, on the whole heat and voltage is bad. But voltage is markedly many times worse and riskier.
euhm. the correct voltages the mobo puts out are in the motherboard section of HWinfo (it shows 1.2 when you first pointed it out at CPU VCORE SOC.)
the CPU shows lower numbers because of internal resistance I reckon. so that just shows the vdroop between what the MOBO delivers and what is actually used by the chip.
noticed that too
Indeed, saw it too. Strange that Steve didn't notice this.
So when monitoring that SOC on my Ryzen 1700 is at max 1.2v, I'm correct at looking at the "CPU VCORE SOC" under mobo readings in HWInfo? This video made me second guess myself by checking the CPU readings of "SoC Voltage (SVI2 TFN)" which reads at 1.09v.
for good or for bad my mobo(gigabyte ab350 gaming 3) shows both sensors at around the same voltage with differences around 0.05 volts
I have the Gigabyte AB350M Gaming 3 (the mATX variant of your mobo). My avg voltage difference between the two is around the same at 0.048v, and max voltage differences 0.094v. Granted not a lot, but to certain users, could be enough between sweet spot and degradation.
One thing that I would point out just for new users to also understand is that the meter they use could also be several % off from what it is reading/displaying and that accuracy varies by range. An inexpensive hardware store meter could be out as much as half a volt or possibly more. This is especially common for cheap meters that cost less than $250-500 US and don't come with a lab calibration sticker or statement of calibration from an independent lab. The lowest precision lab calibrated meter I would consider using for overclocking would have a 4.5 digit display and 20K counts but a 5.5 digit display and 200K counts would still be a better option for seeing problematic voltage variations during operation.
Had a mild yet annoying audio stuttering problem with my new Lenovo, the support team tried to help but couldn't, however we did find in the user forums some discussions that led me to conclude the problem might be an undersized power supply, 400w not nearly enough for a gaming desktop. (No wonder it was such a bargain!)
What made this more credible is that I read numerous solutions from people that only occasionally worked. When I disabled my WLAN I found the audio stutter I was fighting diminished by about 90% further making me conclude that this is a power issue.
Next up: how to kill the GPU you are going to sell used to someone after block chain mining becomes obsolete. FT bitcoin and 95+ efficiency.
To be fair, you kill power components on the card before the GPU dies, but it's a joke so I'm sorry
what will make it obsolete?
Kyle Van small scale mining operations fronted by individuals will die out once cryptocurrency and blockchain become so widely accepted that "casual" individuals will be using it. Once the general populations begin using something scams become more widespread and soon enough federal regulations are placed upon it. Regulations that can track usage of coin and will eventually find a way to track individual transactions then the big original advocates of it drop off and then it becomes an everyday kind of thing that is no longer feasibly "mined" on a GPU array due to exponential growth, taxation, and regulation.
Blockchain and cryptocurrency are logical evolutions, with evolution come changes.
Keith Berwanger you seriously think that's happening in the next few months? I have a hard time visualizing major changes before the next halving. No one really cares what's going to happen in 2025. People are going to be buying things based on this year. Bitmain owns such a crazy amount of crypto they by themselves can fluctuate the market and accelerate or reduce demand. I don't honestly understand why so many miners are frothy over the video cards anyway. You can get the T9 and S9 all day long and you don't have to worry about multiple motherboards or PC setups to run video cards
I don't /seriously/ project anything happening. I do /seriously/ think anything can happen with such a volatile concept. I won't pretend to know the nuances of it all, but anything that can grow exponentially can mutate exponentially.
No one is doing these kinds of vids! You guys deserve FAR more credit.
Thank you for this video. People were going crazy over the internet because it was so easy to get 1600 mhz with 1.25v. At least 1 channel recognize this and explain why is particularly dangerous to believe in software to measure voltages.
As an XOCer myself. This was a good watch. Lots of good info here.
This is some fantastic content, this will be coming in handy today as my Apu just arrived pretty crazy the differences between software and the actual reading did not expect that huge of a jump but will keep that in mind going forward thanks guys!
That isn't the correct probing method to get accurate CPU voltage: it doesn't account for voltage droop across the ground plane between the VRM and CPU, possibly some more due to droop on wiring between the PSU and motherboard. Your readings may be off by 100mV simply from taking your ground from so far away from the load.
No. The difference is about or less than 20mv.
On all the boards? I'd expect that to vary a fair amount between cheaper and more expensive boards with a different number of ground planes, different plane thickness and possibly split planes. BTW, when chip manufacturers specify supply voltages, that's usually directly at the chip package's pins/pads. The closest you can get to measuring that would be to solder jumpers to one of the capacitors either under the IHS or under the CPU unless the motherboard is re-purposing a pair of VCC-GND pins for VRM feedback to read package voltage.
On the boards we tested.
my old Ryzen 3600x was at 1.4v on for time before I wanted to upgrade. Never missed a beat. I gave it to my mate cheap and his still using it on the same board with same OC still going strong.
Excellent Video Steve. You illustrate clearly why Overclockers should not trust the voltage readings that come from Software Apps. The excessive fluctuations in the A-to-D converters used in the hardware on the board. I know I would like to see, at the very least, a standard adopted where all Mobo Manufacturers supply overclocking motherboards with the SOC Voltage and Ground brought to a clearly labeled Hard Post on the top side of the board. In addition the more diligent manufacturer should supply a precision Reference Standard set to the CPU Manufacturers "never go above" SOC Voltage. Good quality DMMs run anywhere from 80 to 200 bucks. That isn't a great deal of extra money to spend considering the user will only have to do it once.
I know the input values aren't exact, but I didn't expect this much of a difference. Interesting insight, thanks Steve!
Great stuff as usual, Steve! You have to be THE MOST TRUSTWORTHY GUY ON THE NET. Your level of information far outstrips most people out there. Your dedicated, and accurate, and you have great hair!
Keep up the great work,
Matt
Calm your ti ts.
In general, other than getting a higher benchmarking score it's usually not worth the extra voltage or higher frequency to run your components at and the cooler the better as long as it's not affecting your overall experience using the computer. I try to run my computer as quiet and as cool as possible I do test out how hard and far and extreme I can get it but for 24/7 use my realistic settings are much more conservative. This whole video definitely confirms my ideology when it comes to building and running one's PC for benchmarking versus daily use. As always Steve, great work!
Video is a bit old, modern CPUs seem more robust, and you will notice they can run at very high voltages stock
awesome work, addressing the real issues of overclocking.
Great job gettin in on the details. Interesting the variance between manufacturers and from different measurement points.. I can't help but wonder if there is some sort of unspoken industry standard, or if they really just sort of all do their own thing. Kinda makes you wonder which voltage is really the important one, because even if you are seeing a certain voltage on the fet if there is loss on the way to the sensitive component then I would think that the loss would have to be kept in consideration and compensated for.. But it seems like they all act so differently its almost impossible to say..
Reminds me of my old P35 board where I went with a pencil mod to counter the immense VDroop woth a quad.
Still find it crazy when people put something like 1.35-1.4v onto recent chips.
That was considered high to threshold 10 years ago and todays chips are manufactured way smaller.
What does SoC refer to in non-integrated graphics CPUs such as Ryzen 3900x?
memory controller voltage
If you deeply know about mosfet drive controls and inverters.Because after the constant drive voltages and frequencis you can increase the source voltage on mosfet because the source voltages and drive voltage on mosfets are not that interlinken with each others.But finally i want to say that these things are easy if you are able to recircuit and resolder your whole motherboard.
You really should connect ground on the board and not through the PSU. Technically they are the same, but that wire length is going to make the readings off.
hawkshot2001 Max 20mv, same idea and same concepts apply.
I told steve not to bother since it generally doesn't make that much difference(10s of mv at most).
Yeah the voltage drop between the CPU socket and the power supply might account for the discrepancy between HWinfo and their measurements
No. No it does not. Read the reply we already posted to this thread.
How to kill your CPU, aka how to take advantage of Micro Center's in-store replacement plan and get an upgrade every year or two.
OC voids the warranty though
Akalion Manufacturers warranty, sure. But Micro Center's in store replacement plan isn't the same as the manufacturer warranty.
Hodor they cant tell that its been overclocked anyways
Haha thats exactly what i just did to my asshat *cough* asus cheap prime b350 plus with my ryzen 7 1700 and my friend accidently let ryzen master kill his tomahawk ab350 pro with ryzen 1700 so we have to do it again. And just shoot the shit with them and dont say you overclocked it. If anything blame a bios update for changing voltage lulz
@RectalDiscourse thank you
Either HWMonitor is mislabeling stuff or you are misreading it. HWM was almost exactly matching your multimeter readings, if you looked at the CPU VCORE voltage. Take a look at 14:53. VCORE voltage: 1.284, measured voltage: 1.28-ish.
A great research piece again! Well done!
I haven't watched the whole video yet, but by using a disconnected Molex ground line as the reference, you're adding in the voltage droop across the other ground lines that are connected to the board which are carrying some current.
This happens because every wire has some resistance, and there's current passing through them to get to power supply ground, so they generate a voltage. V = I * R. This won't be much (probably less than 1/10 of a volt), but it can add to it.
If you want to measure more accurately you should probe as close to the CPU as possible.
Not surprised this is happening, many people on forums have been advising me to run 1.4v as 24/7 use on my new coffee lake, and you got specialist companies selling pre binned chips that are rated at really high voltages like 1.37-1.42v which is crazy for daily use.
Even worse when you consider that chips actual voltage tends to be higher than whats set in bios and what software tools report.
There seems to be a big misconception that as long as you can cool a high volted chip then its all good, as if only heat kills cpus.
one thing to point out is due to the distance between that molex gnd and the vcore pin the value could be slightly off due to the molex gnd not actually being 0V
I think you are looking at the wrong number. There is a stat for CPU VCORE SOC down there, which I think measure from the board instead of the CPU. It gave a better read and closer to the real voltage
Ok then, now I need to look up how to measure voltage from the back of the board. GN the best tech channel on RUclips strikes again!
Take the crappiest B350 board and set the SOC to 1.2V and v-core to 1.425V and keep the CPU cool and i bet you can break the VRM and then break the CPU. I say this before i watch this video.
I love gamer nexus they really do great tests and i'm happy they do it
Might, sure but most decent power supplies would turn off the moment your VRMs burn out. Cant guarantee that your cpu wont get damaged in that time though. High voltages roast it really fast.
@Gamer Nexus if you look at cpu core voltage, it is a correlation between it and the one that you modify in amd master and what you measure.
Thanks for existing. Keep up the great work!
I have my i7 990x overclocked to 4.62Ghz at 1.448 volts, for over more than 10 years.
Still runs like a charm.
He says a lot of things but I waited and waited to hear data or even an explanation about why high voltage actually kills a chip and.. like every other neckbeard, thats where the details get suspiciously thin.
So, as someone who has a B350 Gigabyte Ryzen board. I always read off the Motherboard's reported VCORE SOC voltage in HWinfo64. Which you have set to obscure the SoC denomination in your sensors bar. It's always a lot different than the CPU reported. But it's being fairly close to what you're measuring on the back. THIS is something I wish you'd talk about more. It's something I always warn people on the subreddit on the B350 Gigabyte boards. Read the motherboard's voltage, not the CPU's voltage, because the offsets in BIOS will also stack on top of the voltage you set in Ryzen Master.
Basically: You should note you're reading the CPU reports, and not the motherboard reports. The motherboard here is actually reporting the correct voltage on that gigabyte board.
Thank you so much for making this video Steve! I've been telling my cousin for over a year now, don't trust the software, its usually not accurate. So many people believe it like its written in stone, even the impossible (like sub-ambient temps on FX, just physically impossible), while I'm laughing at the 1 million volts on one of my GPUs show. Always check with a multi-meter to know for sure, never assume what you type is accurate with what happens. Great video Steve, keep up the awesome work!
The HWiNFO64 software is accurate, this video proves that. Steve used a digital multymeter to cross check and validate the readings. Fo example if you set your OC software at ~1.1 volts, check the HWiNFO64 motherboard section to see if the voltage on those components exceeds the safety numbers the manufacturer provided
Ok, so 1 million volts was correct... And 8c CPU in a 20c room is also correct. And even though I measured with a multi-meter, and it showed Vdroop of 0.01v, software was showing 0.1v (order of magnitude different), as well as 1.3v instead of 1.55v like I typed and was reading with DMM, then you're saying software was right the whole time? Mind blown! I just invented a CPU that runs cooler than its surroundings (so it must be absorbing thermal energy, and creating electricity), while capable of mining. I just solved global warming and mining at the same time! Where do I sign up for my Nobel Peace prize?
Calm your t its
But how can I? I just invented something better than perpetual motion! I'm off to save the world, providing energy for 3rd world countries, Hash power for miners, and solving global warming. Its a miracle!
Stahp
I uhh... I'm pretty sure people are missing the point of this video. But so am I.
GN is talking about what people recommend as a safe voltage. What does it mean? This isn't commonly communicated as where its measured, just what the voltage is. Which doesn't really tell you anything.
All the same, I do not think that the voltage readout on the back of the board has anything to do with components past the socket. Those parts report their voltage via software, and should always do so accurately.
But of course I am wrong, as voltage droop across the socket is extra heat generated. This is something to consider, as voltage is only one of two factors in lethality. _But its quite easy to check core temps_
The other X-factor are VRM temps, which are going to go up according not to what the socket uses, but to what the VRM puts into the socket.
Why worry about this at all? GN I think you need to explain how the CPU can incorrectly report its operating voltage, or if it isn't doing that, why is this dangerous?
I remember casually setting 5V on a CPU that’s designed for 3.3V, just because the BIOS didn’t support more options. How far we’ve come! 👨🏻🦳
Hmm I think I should dial back my 1.52V Overclock... 🤔
Momi ye got the 8150. Won't stay at 4.2 without at least 1.5V
Momi u didnt get the 420 joke did u?
lol
I put 1.45v (ASUS maximus V extreme Bios)through a 2600k and ran it @5.1ghz 24/7 on my main rig for over 5 years With a maximum Core Temperature of 60°c(realtemp) without issue at which point I sold the rig on to one very happy camper. I think a key point you are missing here is that a "Safe voltage" can only be given in conjunction with operating temperature and intended life cycle. by "Safe" I mean CPU will maintain OC performance as well as stock operation for the entirety of its useful life ~ 6years IMHO. As I understand it degradation is a wear rate and occurs regardless but is accelerated by increased voltage and temperature. In any event I intend to push similarly high voltage through my 3700x @ a max 60°c core. And would be interested in collaborating to help to establish a degradation curve for the following individual factors.
"Voltage"
"temperature"
"Node Size in nm" Conventional wisdom would suggest die shrinks are more susceptible to degradation as its a Physical Erosion caused by Electromigration
In other words do you want to roast some 3700x's like chestnuts in time for Christmas it will be fun and it will benefit the community
Although I've since owned a q6600 and 6700k, my first self built pc was an AMD Athlon XP 3200+ @2.2ghz. back then I didn't know how to overclock and I never did in the time I owned that but when I bought my q6600 and sold the AMD to a friend, he fried it within 2 weeks, just ramping up voltages because I told him how good the q6600 was. I kinda wish I kept it for sentimental value
Great video. I've got an Asus B350-I with an 1700 in a Phanteks Shift case (gonna need to print a top extraction fan mount at one point). After seeing the thermals on the board. I'm sticking to 3.75 even with a 120 Coolermaster AiO. Higher I could probably put it horizontal and grill sausages. Thankfully it's only purpose is for HTPC and as a rendering slave for v-ray distributed when needed (cheaper to have a few 1700 than less Threadripper systems). Next up is a 1700 NAS system hahaha.
Can't wait for kill the board series. I need a video so I can shut up and just show people :P
What about a voltage drop on a path from cpu to ground? I don't know how much resistance in between them, but they may affect your results. Could you find a pin that located at exiting current from cpu and make your results a more precise? Ah, and also voltage drop from resistance of a socket.
Indeed, this is how not to measure Voltage
Can you do a video on transient voltage spikes when using high LLC? For the new z390 gigabyte boards there is no know info about how much the voltage spikes when coming off load with high LLC. Im running 1.345V on my 8086K for 5.1 Ghz, but I am afraid that my voltage will spike +0.200mv to 1.645 for a millisecond when coming off load. That will slowly kill my cpu in a year.
Steve really seems to be into killing hardware, or at least making them suffer! 🤔
Hi man, I'm just curious. Did you check the voltage difference between the 8 pin ground and the mesurments capasitor ground?
the difference is still so much that there is over volt, but if the ground points is (because the resistance in the motherboard) higher than the 8 pin ground, the cpu will see lower voltage. It might be marginal, but like I said, just curious
All the best and thanks for the great work!
So how does this compare for Intel? It’s pretty common to set voltages as high as 1.350 for core voltage if your temps are okay. You usually can’t even achieve the expected overclock without going at least 1.3
@@djangojansen7648 You can do 1.4 but it's not really a good idea as that will significantly impact the lifespan
Learned a lot, thx GN!
Have you checked that the voltage you read in the socket with the multimeter does not pass the voltage limiter( MOSFET) before feeding the cpu? The voltage limiter may limit the excess voltage that enters the socket.
Of course, measuring in hardware is much more reliable than software, but if you go through a voltage limiter the difference may not be that big. As you know these little details can make all the difference.
Love guys, Keep it RAW!
Check the CPU Vcore at the motherboard's readings...
Hey GN I'm calling bs on that video because HWInfo was reading the correct value for the gigabyte board.
At 13:45 it's saying 1.38V for CPU Vcore and after that you're measuring 1.375V. Seems completely accurate to me.
I dunno what the SoC Voltage it shows is but my guess is it's as useless as the VID reading.
I've always used CPU Vcore in the motherboard section, from what a lot of people told me that is the number you wanna read.
Did anybody get any response to this by now?
My Msi Mpg x570 Atx gaming plus when left at auto pumps Over 1.4v while my Ryzen 7 3700x runs at Stock 🧐 i manually set it to 1.35v and use Ryzen master to OverClock my Chip to 4.4ghz @ 1.35v = Stable on 90% of my Triple A games 🤔 A C Odyssey seems to Crash after 1 hour of gameplay ...... any Recommendations are appropriate & Ty 🇵🇹
Thank you Steve, you are doing a great job at explaining things. I watched your video in a discord channel and had to go to the browser so I could upvote. Currently I'm not doing any OC except for XMP on RAM. I pushed 1.36V in BIOS to make it more stable. I even had to push 1.38V on a friend's rig because it only POSTed at 1.36V, but it could still be unstable. I hope it is "safe".
Was your multimeter calibrated ? They can be quite off while measuring low voltages.
What is to keep me from soldering on lead to the SOC, another to ground, and keeping a cheap multimeter on full time while OCing?
Very informative !!
Thanks for this.
Shit.. and now I gotta use my multimeter to overclock. . and it's covered by the mounting bracket. This makes me so confident
i would have thought that the chip readings would be the more accurate voltage. Where you're reading is before the contacts and there could be more resistance between the socket and the back of the silicon chip compared to where the chip is reading.
that doesnt change the fact voltage is still higher than the setpoint, but it might not be as bad as you are seeing.
Using a molex connector ground pin doesn't account for voltage drop across traces on the PCB.
It's a bit more fiddly, but you should measure the voltage across both sides of that SMD capacitor. You will probably find that it's not as high as 1.4V.
now im scared for my 1700 @ 3.9 with 1.4v OC!
Crazy shit i have my 8700K @ 4.8 with 1.2v
OctoMan PC's 3.9 1.32 on my 1600. At stock I have even seen the auto voltage setting spike to 1.48. Asus mb, same thing kept happening to my z170 asus rog board.
Running my old Phenom @4.2ghz at 1.6v
1.2 Vcore ? If that is correct I need your address to arrest you for: "way more luck in the silicon lottery than me". I can barely get my 8700K stable at 1.35 V
Edit: word
could be the mb or even psu. while you might just be REALLY unlucky you can usually get an 8700k stable on 1.28v or less at 4.8ghz on adaptive and a little bit of LLC. vdroop is a killer of stable OC's. long term stability on mine came at 1.232v @ 4.8, 6hrs prime 95, no problem. with a hero x MB and seasonic titanium PSU. you can limit short and long term wattage to keep temps under control if you're on air. Lastly it could be possible that you have a motherboard that just does not report voltage correctly. I have used the Asus Hero on the past 3 builds I've done simply for predictability. I know what it's going to do and I know the voltages are pretty close to accurate. As you can see in this video some motherboards don't do that
Somehow I missed this video. I have i7 7700K with a Gigabyte Z270X-UD5 motherboard and have the Vcore set at 1.235 and onboard graphics set to 0.600v. I do not use the IGP, I have a MSI 980Ti Lightning. These voltages seem to be fine as I have had this build for just under 1 1/2 years.
Cant they put an integrated multimeter for critical parts? I would like to know the physical reason about this difference, why motherboard cant read nor write correct voltages.
Any proof for the degrading? Just because you measure 1.35V on socket, doesn't mean that there is same voltage in the DRAM controller. I'd personally trust internal sensors over external measurement.
I think this is something the MOBO manufacturers are responsible for and should be on top of. If people can't rely on the information they get from their MOBO they can make a lot more damaging mistakes which i would say are out of their control. Of cause their will probably be some margin of error but if it can be the difference between a safe and definitely unsafe voltage the difference is simply too big also if you are gonna have a "big" difference at least be on the safe side specifically referring to Gigabyte.
Can you show us which point you're measuring with the multimeter probe or how to determine that point?
It would seem to me that 1.2v is a good empirical (middle of the road) value for SoC in that there so much variance between manufacture architecture, and electronic component quality. That said, t would seem to me that the absolute or max SoC voltage allowable is a need to know requirement - would that be 1.35 volts? Question: Curious to know, to stabilize SoC voltage fluctuation, did your testing include changing LLC - I saw from your excellent presentation that there was quite a bit of volt flux?
Good old 80486 handles 3.6v.
Im Running my I5 8600k on 5.1Ghz 4500Mhz on my Cache @ 1.440v. Let´s see what happens in 1 Year or 1 and a Half.... (Cpu is Delidded Stable @ 80-85°C @ 30minute stress testing depending on the room Temperature) In Gaming it wont go further then 65-70°C. Im using a Broken Eco cooler..... Motherboard is a MSI Z370-A-Pro With the latest Bios (MS-7B48) Maybe another Mainboard will give me lowervoltage on the same settings. But it is not worth an 160€ upgrade.... My Old FX8350 runned stable @ 5Ghz with 1.45v for over 1 and a half Years. And it is still working.
What about VCORE hwinfo readings?Are they accurate?
Very good video as usual. I would have liked to see results from a good multimiter.....
That wobbly monitor is a constant in the latest videos :)
Isn't there a considerable vdrop on that molex ground vs. ground near the cpu? I will answer, yes it is. You should check that whats the difference. Also what is your multimeter tolerance? It looks like 0,05% accuracy which in 2 volt meassurment means 0,1V. After what I've seen, I think that readings are way off.
I have a question, many OCers just raise the speed only and leave the voltage at "stock" or auto setting- so how accurate/safe is the stock/auto option regarding actual delivered voltage? It seems with these apus you can get 1.4 or so more speed, but only 1.2 x in performance (the square root). It may not be worth the last 100-200 mhz if you need to raise voltage for stability with the really unknown actual output sent to the socket. Not everyone is going to pull mobo and measure things- I think there might be a shorting danger with the small and crowded under-socket area.
@9:28 How do you know which correct cap will read the soc voltage? multimeter is blocking view. Also does it matter which molex ground?
Is this the scenario with only apus or is applicable to ryzen cpus aswell? If so , how ryzen behaves to it? Same variance or is it less as theres no igpu.
dude, that was insanely important! i need to make sure the voltage of my cpu (currently a fx8320 at 1.39) if that was really 1.39.
I love this channel so much...
Question, had my i5 10600k At 5 gig adaptive offset 0.050@ 88 degrees was highest temp on core 2. I used MSI try it now on ram, left xmp, try it now on my vengeance RGB pro 3200 DDR to 3400, rebooted system said ram couldn't clock, reset Mobo to default. Put xmp back out same settings on over clock now when I run test core 2 has hit 97. What gives, and my voltage is auto, but looks higher than before? Did that ram oc screw up cpu potential? Please let me know if you can
Great video can you do another using maybe an i7 8700k?
recommend to use GND from MB or measure ground reference from MB.
Would this method of checking voltage be the same for most MOBO/CPU combos? IE checking my OC on my 2700k vs what hardware info is showing?
And when was the multimeter calibrated?
Why would you not want Load Line Calibration to be set to its max? I'm not sure why its even a setting in Bios. Wouldn't delivered load voltage be closer inline to configured and reduce the guess work?
could you make a video of how accurate are the z370 mainboards are messuring the vcore of the cpu? would be very nice. i do not have a multimeter to check that. important are the models of asrock (taichi and fatality itx) and asus maximus x hero and apex.
i've got 2 questions 1 how low would you asume mx cpu will get if i direcht die cool it with a pleter witch is cooled by a 65mm quad rasiator but there are dual 1080ti's on the same loop
and 2 at those temperatues what is a save core voltage (cpu=i5 8600k)
Is this behaviour observed on Intel as well? (and for any APU/CPU?)
I don't doubt that there are differences between the voltage reading in the settings and the "actual" voltage. But I doubt whether the voltage meter is calibrated or not, and when this was done.