Come on dude You make me worried now should i worry abaut mu 3600 on stock settings?😱 I have phanteks ph-tc14pe CPU Cooler I dont oc until the varranty is no more then MAYBE I START playing oc If i know What i do 🤣
For GPUs, I tend to do a mild(ish) overclock combined with an undervolt these days. Then maybe a memory overclock (at stock voltage). You guessed it, I mainly use Radeon GPUs!
@@DJ_Dopamine its not a bad trade i guess 100 dollars less because all of the tuning wasnt done at factory, fun for tweakers not the best stock products tho
Amazing video. I've been getting RUclips recommendations from this channel for months, and ignoring them because I didn't realize this was Buildzoid's channel. Glad I finally wised up.
We love these vids, definitely not boring. The only thing that'd make them better is dark mode (I watch a lot before bed so not having a white background in my eyes would help)
Yes!!!! Love this overckockers' general knowledge content! Moar! Also, I remember a video from Roman (I think it was the 9900k review) where he explored the relationship between temperature (all the way to LN2) and power efficiency; therefore exposing how leakage happens.
I've always wondered why that was never a thing. Competitions like: 'highest performance per watt for a given chip' Or 'highest performance from a system that draws no more than 100 watts' With variations with increasingly tighter power budgets. Oldschool challenge; How much performance can you squeeze out of a system that has power limits similar to 80's and 90's PC's? (that's about 300 watts at most) Maybe... What's the most possible performance you can safely get out of a system that uses passive cooling only? Or even better, can you get a modern system to ever run in a stable configuration with no cooling whatsoever, not even heatsinks? There's so many options for these kind of things, but it seems 'biggest overclock ever' is the only one anyone bothers with...
@@KuraIthys On the point of "highest performance per watt for a given chip", you can just run slow (BCLK multiplier set to 1, or the lowest available) and have the chip sip incredibly low amounts of watts, to the point of being unmeasurable between idle and loaded. On the point of "highest performance from a system that draws no more than 100 watts", you begin to run into the problem of having to take EVERYTHING into consideration, and it becomes a contest of "who can throw the most money at the wall" to get the best efficiency from your PSU, motherboard, CPU, RAM, GPU, OS... even the hard drive will be an SSD that copies a barebones Linux distro on a RAM disk and runs from there, allowing shutting down the SSD to save a couple watts. Let's be crazy and add LN2 to get the lowest current leakage possible. There's also the issue of measuring power consumption accurately: _(in my best impression of Buildzoid)_ motherboard probes are teeeeerrible, so you'd need external hardware (with low tolerances may I add, so it will be expensive), a way to connect it to your PC, and it being standard for a competition to happen. The last part will be quite tricky. I agree this is interesting to explore, but to make it a competition on par with balls-to-the-wall overclocking? Nay. I see getting every drop of efficiency more as a personal project for your particular system, the kind that won't see any benefit from overclocking or just can't (like locked Intel SKUs or based on A320 AMD motherboards), than a competition project. (And obviously the biggest downside with such a low-wattage competition: no RGB. /s)
As a true Tech geek, I enjoy your videos (and understand what you are explaining) :) I always undervolt if not OC'ing. hat gives you a lot better efficiency. I have a retro rig with a QX6850. Running that beast below 1.1V. stock speed. (not using such a good cooler) Temps has dropped 10-15c during stress. Now that is eficiency increase!! BTW. Sometimes you can actually gain efficiency with a slight OC. I remember the good old days with C2D and C2Q (20% OC) with undervolt as well. Was using about the same power as stock clock/volts. That equal 20% better power efficiency with OC. Real OC fun is not the same (fun) anymore. Today the chips do a decent job themselves (Turbo , MCE). Very dificult to get better effiency with OC.
I remember undervolting my 3570k when I recently got it and It ran really efficient. At some point however I started playing planetside 2 a lot, so up to 4.7ghz it went.
I reduced the power consumption of my Threadripper 3970x by -35% (from 280W to 180W) by capping it to 180W PPT (stock is 280W) in the PBO menu in BIOS. As a result I lost only ~10% multi-threaded performance while the single and lightly threaded performance stayed the same. I think I'm going to run it like this with BOINC since it's massive power savings for minimal performance loss. 🙂
that center chart, and the right one, is mostly why portable devices tend to be so limited in performance, battery limits and thermal dissipation limits the power and clocks of the thing. Great vid. B)
Pretty interesting graphs... Bottlenecks can be fascinating though. I remember years ago I started using unofficial drivers with my laptop because the manufacturer literally never provided even a single graphics driver update. and the one it did provide caused constant stuttering in games. So... The chip in that laptop was an x1400 - which, I mean, for the time, having discrete graphics in and of itself put you well ahead of most laptops, but that chip is one step removed from the lowest end chip Ati produced for laptops. In particular, an x1400 (which only exists in laptops) is an x1300 with a memory bus that's twice as wide. Anyway, as a side effect of installing unofficial drivers, I got access to a whole bunch of settings no laptop driver software would generally ever let you touch. One of these was the desktop overclocking software. And, that leads to the bottleneck in question; See, in spite of having double the memory bandwidth of the otherwise identical x1300, (And yes, it IS about twice as fast)... In examining the overclocking and underclocking results... Well... It got very interesting indeed. First of all, the clocks could be increased by something like 80% and it still worked (this is without having any kind of voltage control). Granted, if you overclocked both the core AND memory by that much it got unstable, but you could usually push one or the other nearly that high and push the other by about 20% and get it to run fairly stable at those settings. Of course, the cooling system here means the RAM has nothing at all cooling it and the GPU shares a single cooler with the CPU. (thus CPU load directly affects cooling performance available to the GPU as well.) So... Anyway, the result that interested me here was the actual performance metrics in games. Basically, altering the core clock did pretty much nothing in any games I tested. Overclocking it by 80% did nothing. Underclocking it by about 75% (anything less than that and the chip became unstable for some reason) also did nothing. Nearly double the core clock, or cut it to 1/4 it's stock speed and nothing changes whatsoever in games of the era... Meanwhile, change the memory clocks and performance in games scales almost linearly with memory speed... Bottlenecks huh. XD To be fair, Ati badly misjudged their GPU designs from that era. While the x1950 did really well, most of the rest of their range did not. With the x1600 being especially poor. The problem is, Ati had assumed games would use a lot of shaders; Procedurally generated textures, vertex shaders, and so on. As such, they had put a lot into giving these chips (Even the low end ones) a LOT of shader performance. Unfortunately... What happened in practice was that games of that era doubled down on textures and similar effects that hammer the chip's memory bandwidth and texturing units... And Ati's chips had not been built with that in mind. In other words they were over-specified for something games didn't really make much use of, and under-specified for things they did. And the low end, but especially the midrange chips were by far the worst off in this regard... Anyway, just interesting sometimes what overclocking can do in practice when a bottleneck is involved. You could find yourself with a truly massive overclock and yet the practical benefit is... Nothing whatsoever...
on curve #2: power increases exponentially with voltage because as you increase voltage, the resistance of silicon falls, increasing the current non-linearly. meaning, at 5% higher voltage you're not getting 5% more electrons flowing per second (current) like a resistor would behave, you're getting say 10% more current, meaning 15.5% more power (U*I)
i love this shit. Nice articulation at 18:15 saying power consumption doesn't go up too bad when the cpu is waiting for ram to respond, since the subsets of cpu logic in question aren't dumping charge into ground writing 0s with calculations.
Not true, because the standard clocks at overclocked HBM is 15-20% faster, and you ate 20W in HBM. Besides that flat boost caused by Infinity Fabric behavior (not memory itself), games tend to scale linearly with clock. (At 4k.) So in many games you will get about 5% slower with your settings. (1500 memOC vs 1680 MHz 1.2V. You can run stock boost 1630 at 1.18V for 50W savings and minimal performance drop. Has space for memory OC.) But about 100W cooler. Presuming the card is stable at 0.9V 1500 MHz, not all are. (Typical is 0.95V, and usually it is capped by IF bus voltage, which has to be increased to 1.025V or 1.05V to run that 1111 MHz memory clock to trigger 1200 MHz IF speed.) Source: have 2x Vega 64 overclocked. They tend to random brownout and crash in specific titles when undervolted too hard. I can "run" stock settings at 1.135V, but specific games will reliably crash. By the way, locking clock to P7 gets 5% boost stock. Card is also more stable allowing for more precise undervolt. The cards do feature some sort of LLC.
You've reminded me of one more graph I haven't tested in detail, it was more a small test, gpu sample of one. It was a long(ish) time ago, but I'm pretty sure that lowering the temperature of an R9 380 by 5 degrees gave me about 5W drop in consumption measured from the wall. Really don't remember the fan speeds, also it was undervolted quite a bit. Nothing to write home about, but it was measurable and consistent. Maybe you had more extreme scenarios.
@@dpokor well i havent tried 1.5, but tried 1.4 with loose timings and didn't like it, tried lowering the stock timings with increased voltage, same thing, the memory was corsair led 3000mhz cl15-17-17-35 2n and i could only change 2n to 1n and was stable anything else i couldn't lower them not even by 1 not even trfc by 1...Tho i could oc them to 3333mhz cl15-17-17-35 1n at 1.35v .That was the best i could do.
Hey man, I'd be interested in a video that explains how different types of calculations or the use of other instruction sets (like AVX) affect the stability and the power consumption of a CPU. The voltage and the frequency remain the same so I wonder why AVX load is so much harder for a CPU to run. What do I have to look out for when overclocking? Do I have to test stability for all instruction sets if I want true 24/7 settings?
You don't need to test all the instruction sets, only AVX. AVX is special because it adds vector instructions - a normal add does A = A + B, but an AVX add is capable to do 4x-16x of these additions AT THE SAME TIME. For example Zen 2 CPU with a 256 bit wide vector unit is able to do 8x 32 bit additions. Skylake-X has 512 bits wide vector unit, so it can do 16x 32 bit additions. You get 4x-16x times the performance, but you also get crazy high power consumption (a lot of work equals a lot of power). If you want more info, just google AVX or SIMD (Single Instruction Multiple Data). Some other instruction sets (something like TSX maybe) will increase power draw too, but nearly not as much as AVX/other vector extensions do.
Well liquid helium is less hot than liquid nitrogen just much more expensive per oz. So with luiqud helium you can get colder than with liquid nitrogen just it will cost you more money to do so for the chance of getting a little higher.
undervolting even seems to do me favors in overclocking, at least regarding turing cards. my 2060 hit like 1985mhz solid at 950 mv, stock being 1885 mhz at 1000-ish mv. i think that is when temperature comes into play.
I've played with subzero at normal voltages. I find that below a certain voltage depending on architecture it makes very little difference whether you're at 100 or -100 degrees. it's like maybe 50 mhz difference best case. I haven't done it with any recent chips though, so take this with a grain of salt, but I imagine most newer stuff will be similar.
As we know th left curve is tempture dependent on chip hotspots are ram id's flipchip if so how much die is on top of the transistors if it where to be shaved down would we gain noticable clock speed increases as it is very tempture sensitive. (I believe d8bauer did a video on this with a cpu) Second are ram ics sensitive to cold tempture and if so is that because ram has tighter timing requirements than a cpu and changes in drift current velocity or the propertys of the transistor materials changes enough to greatly affect timing requirements.
memory is so low power that it wont matter much. you can use 2V or more and get away with a fan an roomtemp air. in a case it might be better to go with watercooling but thats more an exeption. over 1.45 to 1.5V you need a fan or some airflow over it. thats enough. the other i dont know enough to understand what you mean, but ram tends to like 15-35°C so roomtemp is allmost best you can get. dont know why it wont scale with lower temps, but i think resistors and stuff. On Die Termination is a thing with RAM and a resistor that changes resistance might not be ideal. but again, thats just my idea of it. CPUs and GUPs are a one chip thing. if you had 8 or so on the same board that need to talk in perfect sync, it might behave simular
Nice video man. Something I have noticed is that with a lot of CPUs if you want to run at stock speeds, or even want a 100 mhz oc or something small; You can usually get away with a slight undervolt. At-least with Ryzen. Example being my 1800x. Stock it uses 1.4v, but it runs perfectly stable at 1.25 with a +100 mhz. I can probably get a little more OC at 1.25, but at these settings it uses very little power and hardly gets hot. I would imagine you could do this with any Ryzen. I really want to see how far you can undervolt a 3000 series Ryzen, at stock speeds, and run stable; and how much less power it uses. I think I understand why the stock voltages are higher than they need to be from the factory; It's due to a safe number across the line that guarantees that all chips run stable, because maybe there is a small percentage that will not run stable at 1.25v stock speeds, but they all will at 1.4v.
Buildzoid, this was indeed interesting. I have a number of SB i7 2700Ks I run at 5GHz, which for the 2700K isn't really that high (any 2700K on an ASUS M4E will do 5GHz with just a simple TRUE, one fan, not that much voltage and good temps), but today I noticed one of them is starting to need somewhat more voltage to remain stable, even though I know its frequency potential hasn't been pushed that much. I've had to raise vcore to the point where while idle CPU-Z shows vcore as being 1.475V, while a different system (same model of CPU/mbd, same RAM setup) shows only 1.456V. Years ago the degrading unit showed only 1.456V even under load: valid.x86.fr/euhlxt It's been used a lot for benching, while the other chip is still my daily tasks system and thus is used somewhat lightly. So what you said about older 32nm CPUs having a degradation point further back on the voltage/frequency curve chimes nicely. The CPU in question normally handles gaming workloads ok, but now it goes splat if I try a CB run or a Handbrake transcode, and today it barfed while running a mere Resident Evil 6 test (a few weeks ago it wouldn't bat an eyelid running FC5, SotTR, Hitman 2, GTA V, etc.) By contrast, a different config (same model parts except for the RAM) when originally setup back in 2012 only needed 1.4V to remain stable: valid.x86.fr/show_oc.php?id=2334534 Thankfully for the benching system I have another hardly used 2700K I can move onto the mbd, so it should be ok. So yes, it was useful to see how different architectures, indeed GPUs, could end up being by default on different parts of that behavioural curve. Is it fair to say that the older the arch the further back on that curve the CPU will be? Or is this something unique to SB? What about Nehalem/Bloomfield/Gulftown/Westmere? I have an i7 950, 975 and 990X I've not tested yet, also various XEONs. Also Buildzoid, have you ever experimented with Lynnfield and Clarkdale? I have an i5 680 which, because of its high base clock, I am hoping to be able to push to at least 5GHz (not tried it yet); see: ark.intel.com/content/www/us/en/ark/products/48504/intel-core-i5-680-processor-4m-cache-3-60-ghz.html I have an i3 550 which easily runs at 4.7GHz on even a moderately average mbd (I used an Asrock P55 Deluxe), but I plan on trying the i5 680 on a far more potent mbd, the ASUS P7P55 WS Supercomputer, but what you described makes me wonder if the CPU could indeed scale just fine but may degrade easily even if it's initially perfectly stable and seemingly not that high up the scaling curve. Strange, I don't think I noticed before that the i5 680 and i7 2700K are both 32nm, somehow I regarded the i5 680 as being older. Hmm, I have some new/unused i3 550s I've not touched yet; if you'd like one to mess about with, let me know and I can post one, though of course you'd need a P55 mbd to use it. They're a fun wee chip for oc'ing, the good old days of bclks, etc. Here's the one I messed about with the most: valid.x86.fr/show_oc.php?id=2093028 Lynnfield was fun too, reaching clocks X58 owners could only dream of, especially the i5 760, but they were somewhat below Clarkdale or SB. My results: www.sgidepot.co.uk/misc/tests-jj.txt
well im running my 1080Ti on .95 Volts since over a year now and only have 3% less performance from if id overclock it so yeah. Getting nearly 10°C lower as well and running 1900Mhz stable. I guess even most Nvidia cards that get shipped these days are just super inefficient on default than anything else. I also undervolted my 3900x cause at full load its drawing way to much volts from what the cores are going to need anyways.
Many people, even regular pc enthusiasts would be quite surprised at how much power their Intel chips would draw at something like the base clock or somewhere in the middle of base and boost depending on what chip /gen with the lowest stable voltage. Especially when the chip maker is already releasing the chips at or near the max of what they're capable pf doing on normal cooling. Like a 9900ks or whatever chip that has the 5ghz turbo.
I kinda like to know more , like some specific like voltages for like architecture , like you know when you o.c like moderate if the voltage like needs a big bump like you described or like it is safe to a point like you don't get like so much performance , like you won't notice more fps like almost without o.c then if I keep o.c like to feel more diff like real fps increase like 5-10 diff then maybe you can say like yeah this is like i like to have it because like i like the performance i get. Do you like get me ? 🤷♂️
Love it bro what you think about MSI PRO B450M PRO-VDH MAX closest matx i could find to the mortar in the states would be fine for a 3600x planning on locking at whatever i get the frequency stable and temps good. The case has an absolute load of airflow for the vrms
Random thought on the frequency/voltage curve: could it be that by increasing voltage, the magnetic fields generated by the wires induce strong enough currents in other wires, generating unwanted voltages, and introducing instability?
Hi Buildzoid, I have a question about the Frequency/Power scale you listed there. Based on my testing from highly power limited Max-Q GPUs in laptops increasing the frequency just with the slider on f.e. MSI Afterburner never translated to an increase in Power consumption (or heat), I got roughly the OC i dialed in. Where does this come from? Based in my understanding the powerdraw is hugely impacted by the voltage but if you just increase frequency, you will use the OC headroom a given voltage has for your specific chip (based on silicon lottery) and as long as you are not running into instability from too less voltage you do not use more power. A quick explanation or some sources to read through would be a huge help! Thanks :)
Followed Hardware Unboxed Ryzen mem OC guide last night and got my Royal Bdie 3600 16 16 16 36 up to 14 14 15 14 28 @ 1.45v. On a taichi 3900x. My pc was crazy fast even with me rebooting to bios a few times. But this morning it looped, lost the settings, even the profile I saved in bios...
Could you go over a video explaining how/why a cpu wont clock faster when temps are not a problem? I have two 9900k's, both delid, one direct die. I can keep the chips under 80C but I hit a wall with voltages even though I do not get hot. I just dont want to pump voltages even if temps are low.....
and then there is ryzen 3000 and some special cases (reducing ppt instead of undervolting). got a 3700x on a b450i aorus pro wifi and 2x 8GB 3600mhz cl16 ram and when i drop the ppt from 88W to 65W (26% power down), i reduce the Cinebench C20 points by only 2,5% (from 4799 to 4682).
But what happens if you overclock and undervolt at the same time? I managed to get a 7820HK undervolted to -125 Mv while having it overclocked from 3.5 GHz all core to 4.2 GHz all core.
Hey dude do you know what NMODE is in the memory timings table? Default sets it to 2 and XMP Profile 1 sets it to 1. I googled NMODE and can't find anything. Thanks.
BZ, am i crazy to run 9900k direct die cooling with sanded die at 5.2Ghz at 1.42Vcore on z390 aorus master? Is it any "safe" or degradation is goin on nicely? temps are around 85C at prime95 non avx
Buildzoid I think I have a safer way to do something similar to the liquid metal shunt mod, graphite is conductive and it might work, I’m going to try it when I find a few used cards to test.
well it only speed up, but there are some voltage levels to keep in mind. problem is, that if you higher freq. that also speeds it up. 1.4V might be good for 5ghz cpu and ok for 5.2ghz. but another damaged his 9900k at 1.4v 5.4ghz. depends on architecture a lot. intel can handle more volt than amd. also more people means more "save level". on intel i would kepp the io/sa voltage 1.35V or below. some say 1.3V. older posts state 1.25V or even 1.2V. asus says 1.4V thing is. most people dont run 4000mhz+ on RAM. so they dont need 1.3 V or up. if you dont have 8. or 9. gen, you may cant even run that high. one chip might handle 1.4V good, another degrades after 1 year. if you dont want to risk it, then stay low. and allways stay as low as possible. there is no benefit to run with higher volts than you need
@@mauriceelet6376 you only need 1.15VccIO and 1.20VccSA to run 4000Mhz memory on coffee lake, but motherboard makers just couldn't care less about doing proper testing and lower the voltages: just like with core auto-overclocking they always overshoot the voltage at least 100mV
@Smucky ! There's no easy way to know, but there's some known guidelines: -on Intel pretty much from Sandy Bridge until nowadays, up to 1.4V is considered safe (with proper cooling, meaning under 90°C with max load). -on Zen 1st it's 1.4V, 2nd gen is 1.35V and 3rd gen is 1.32V. Those voltages are for all-core max load with Vdroop, meaning those should be the voltages that are physically going into the CPU, not what you punch in BIOS or what SVID reads.
I don't have enough CPUs to gather proper data from but just looking at how my 3950X behaves I wouldn't apply more than 1.25V full AVX load to that CPU.
@@ActuallyHardcoreOverclocking Would you feel safe using higher volts for lower core count parts? Or does their lower binning degree make them leak more, pulling even more current, and therefore having lower number of cores doesn't actually improve that?
Alright so is there a graph for the scaling of temperature vs frequency ? Or is that chip dependant ? With ambient cooling my 8600k will not boot at 5.4ghz even with 1.55 volts
thats part of silicon lottery or as buildzoid said its chip dependent, but if you lower the temps you will get better performance and stability in most of the cases and also increasing voltages only goes so far as he explained in the video and your chip might refuse to give good results with anything above for example 1.5v
So I shunt modded my Titan xp with LM and I have my 9900k running 5.2@1.45v extreme LLC and I'm pushing 230w and 22A. Only once my vrm is over 80 does my chip hit 93 tops but 83c if I aren't running CB for long runs. I think my CPU is not happy running 22A peak.
@@ActuallyHardcoreOverclocking sorry just saw this. Yes it is indeed "Current (IIN)" around 20-22A. Output is around 160A and 215-220w output and no avx offset for what it's worth. And it's the aurus pro wifi
What I don't understand is how I undervolt and get 1900 mhz @ 1.000 and when I'm "+110" overclocked my GPU jumps to 2005 at the same 1.000 voltage... but anything higher than 1900 Undervolt is unstable
This actually helped me notice a bad paste job on my 5700 XT! Hotspot was hitting 110C(throttle point) while edge temp was still only 50-55C. Taking it apart & properly pasting brought it down to 80C hotspot at the same edge temps. I never would have known this was happening without that hotspot reading, as the edge temps were fine!
Rahul Mahesh my xmp profile is 2666 16 18 18 36 1.35 volts. But i crank up to like 3200 with 1.4 volts. How can I test stability with 16gb of ram ? Dram calculator max is 8gb and memtest is like 2.5 gb
It's the worst for overclocking. I have an mfr kit and it doesn't behave anywhere like afr. I'm running mine at 3000mhz 13-15-14-15-21 at 1.5v, and i got very lucky with the silicon lottery. It's literal trash.
I used to have HyperX Predator mfr, was able only to tighten the timings, even +200 mhz was unachievable with reasonable voltages. Needless to say I sold it
When doing LN2 OC how come they do not also put one of those pots on the backside of the motherboard CPU socket? Would not that help with hot spots cooling?
The fact that if the socket and pin or land grid cooling in the metal can shrink them back from each other breaking contact The skill on handing 2 LN2 pots at the same time and finally "cold bugs"
Fun thinking.. I'll guess if all computers (not OC'ed) were undervoltet to the/above *limit at stock speed (*still maintaining max stability) we would shave of 3% power globally. Think... All the datacenters, office/home computeres. That is a lot of kWh "wasted" globally in the world iright now! ;) Think If each rig can shave of say 10% power consumption average..? New platforms with higher effiency can still (at least 98% of them) be tweaked to better effciency. All "stock" settings are preset to get stability to even the worst silicon batches.
I suspect quite a few datacentres may already be doing this as energy prices increase. It's also why ex-mining GPUs can be great value, as miners very quickly realised that slightly undervolting wouldn't hurt performance, but would drastically reduce power consumption, and thus increase the power/return ratio...
@@JaenEngineering I know. I have been mining a few years. All hardware still working fine (almost 24/7 running 2+ years) after best possible undervolting / efficiency tweaks possible ;)
@@Wushu-viking indeed. And that 24/7 operation also confers another advantage, mainly thermal cycling, or lack thereof. The daily expansion and contraction of parts as they go from ambient to full load (or higher if it's been OC'd) temp and back again is possibly one of the biggest killers, as the various parts expand and contract at ever so slightly different rates, placing stress on any interconnects. The only real downsides are thermal paste degradation as it slowly dries out and that the cooling fans will eventually die as their barings wear out, although neither of those is particularly expensive to fix, so should be easily outweighed by the savings .
@@JaenEngineering Exactly! Generally.. Blower fans are really durable for 24/7 running. (more so than the regular "low quality" Axial fans you find on "aftermarket" graphics cards) All my Vega 64 blower fans are still in good condition. Actually even the paste seems like its still okay regarding temperature. They must have buildt these cards quite good from AMD.
I have noticed how voltage required rises really fast above 4ghz on my 1600x. It does 4ghz with 1.23v, 4.1ghz with 1.32v. I have ran it at 4.2ghz but i cant really keep it stable, best stability seems to be at 1.42v, going higher does not seem to help at all.
my 1060 is a bees dick off of falling off that vcore/clock chart, 1.093vcore wont hold 2215 but 1.075v will hold 2202. (edit) is 1.4vcore safe on a devils canyon cpu under a 360mm rad, its second hand so its not the greatest concern
well, considering when a logic gate is on or off, there is less power bled through the gate FETs than when it's transitioning from one of the two states. the more times a second a logic gate is changing from 0 to 1 or 1 to 0, the more often it's using more power in that state between 1 and 0 (cuz both the high side FET, and the low side FET, are not totally off or on. so the current goes through the high side and low side FETs from the power rail to the ground rail when the logic gate is changing state between 1 or 0). that's why at the same Voltage at different clocks the 3rd graph is so different from the first 2.
I would not push past 1.2625v mayyyybe 1.275v, anything over is simply not that safe. You have to remember that auto oc is only providing voltage in ms time frames. The best bet is to run prime95 small fft for three minutes and see where your sbi tfn2 lands. Thats your cpu's fit voltage then target an oc below that by about 0.0125v.
And even if he is, who cares? I also like being awake at the night. Its silent around you, its dark, nobody interrupts you, and you can simply do your thing :)
ich schau mir sowas nicht an, ich kenn nur einige hintergrundsachen. aber interessant mal jemanden zu sehen der sich sowas wie das hier gibt und solch ein Zitat kennt. wie hast du hier her gefunden?
3900x ships with with 1.42v at 3.9Ghz but i managed to run it at 4.2GHz at 1.23V with level 4 LLC. If you keep your CPU at factory voltage you more likely to kill your CPU early that it should be.
Well this is now public I guess
Come on dude You make me worried now should i worry abaut mu 3600 on stock settings?😱
I have phanteks ph-tc14pe CPU
Cooler
I dont oc until the varranty is no more then MAYBE I START playing oc If i know What i do 🤣
It is good information, so worth putting out there
Was it not supposed to be public?
@@iamthegamer908 do it when you have warranty... They won't know ...if you destroy your CPU after warranty
..you are toast
@@mr.waffentrager4400 nah im okay whit it on stock no need oc Any of These ryzens its good for my gaming needs And amator video editing 🤣
Thumbs up if you want to see an EXTREME undervolt.
Not enough people do that. It's interesting to see.
For GPUs, I tend to do a mild(ish) overclock combined with an undervolt these days.
Then maybe a memory overclock (at stock voltage).
You guessed it, I mainly use Radeon GPUs!
Richard Ayy, same.
@@DJ_Dopamine its not a bad trade i guess 100 dollars less because all of the tuning wasnt done at factory, fun for tweakers not the best stock products tho
"Extreme" meaning cutting voltage so far you just blackout? I even got a t-shirt printed for doing that
Amazing video. I've been getting RUclips recommendations from this channel for months, and ignoring them because I didn't realize this was Buildzoid's channel. Glad I finally wised up.
We love these vids, definitely not boring. The only thing that'd make them better is dark mode (I watch a lot before bed so not having a white background in my eyes would help)
Especially considering the time when this video became public
Yes!!!! Love this overckockers' general knowledge content! Moar! Also, I remember a video from Roman (I think it was the 9900k review) where he explored the relationship between temperature (all the way to LN2) and power efficiency; therefore exposing how leakage happens.
I could watch these types of videos all the time. Keep them coming!
Well it's a good thing this channel isn't called "Actually Hardcore Efficiency"!
That might be cool.
I've always wondered why that was never a thing.
Competitions like:
'highest performance per watt for a given chip'
Or 'highest performance from a system that draws no more than 100 watts'
With variations with increasingly tighter power budgets.
Oldschool challenge; How much performance can you squeeze out of a system that has power limits similar to 80's and 90's PC's? (that's about 300 watts at most)
Maybe... What's the most possible performance you can safely get out of a system that uses passive cooling only?
Or even better, can you get a modern system to ever run in a stable configuration with no cooling whatsoever, not even heatsinks?
There's so many options for these kind of things, but it seems 'biggest overclock ever' is the only one anyone bothers with...
@@KuraIthys On the point of "highest performance per watt for a given chip", you can just run slow (BCLK multiplier set to 1, or the lowest available) and have the chip sip incredibly low amounts of watts, to the point of being unmeasurable between idle and loaded.
On the point of "highest performance from a system that draws no more than 100 watts", you begin to run into the problem of having to take EVERYTHING into consideration, and it becomes a contest of "who can throw the most money at the wall" to get the best efficiency from your PSU, motherboard, CPU, RAM, GPU, OS... even the hard drive will be an SSD that copies a barebones Linux distro on a RAM disk and runs from there, allowing shutting down the SSD to save a couple watts. Let's be crazy and add LN2 to get the lowest current leakage possible.
There's also the issue of measuring power consumption accurately: _(in my best impression of Buildzoid)_ motherboard probes are teeeeerrible, so you'd need external hardware (with low tolerances may I add, so it will be expensive), a way to connect it to your PC, and it being standard for a competition to happen. The last part will be quite tricky.
I agree this is interesting to explore, but to make it a competition on par with balls-to-the-wall overclocking? Nay. I see getting every drop of efficiency more as a personal project for your particular system, the kind that won't see any benefit from overclocking or just can't (like locked Intel SKUs or based on A320 AMD motherboards), than a competition project.
(And obviously the biggest downside with such a low-wattage competition: no RGB. /s)
As a true Tech geek, I enjoy your videos (and understand what you are explaining) :)
I always undervolt if not OC'ing. hat gives you a lot better efficiency. I have a retro rig with a QX6850. Running that beast below 1.1V. stock speed. (not using such a good cooler) Temps has dropped 10-15c during stress. Now that is eficiency increase!!
BTW. Sometimes you can actually gain efficiency with a slight OC. I remember the good old days with C2D and C2Q (20% OC) with undervolt as well. Was using about the same power as stock clock/volts. That equal 20% better power efficiency with OC.
Real OC fun is not the same (fun) anymore. Today the chips do a decent job themselves (Turbo , MCE). Very dificult to get better effiency with OC.
This was probably the most informative piece of computer knowledge i completely understand
I remember undervolting my 3570k when I recently got it and It ran really efficient. At some point however I started playing planetside 2 a lot, so up to 4.7ghz it went.
I reduced the power consumption of my Threadripper 3970x by -35% (from 280W to 180W) by capping it to 180W PPT (stock is 280W) in the PBO menu in BIOS. As a result I lost only ~10% multi-threaded performance while the single and lightly threaded performance stayed the same. I think I'm going to run it like this with BOINC since it's massive power savings for minimal performance loss. 🙂
great explanation, I am always picture such graphs in my head when I am thinking about various correlations
that center chart, and the right one, is mostly why portable devices tend to be so limited in performance, battery limits and thermal dissipation limits the power and clocks of the thing. Great vid. B)
This was cool, more of this
Pretty interesting graphs...
Bottlenecks can be fascinating though.
I remember years ago I started using unofficial drivers with my laptop because the manufacturer literally never provided even a single graphics driver update. and the one it did provide caused constant stuttering in games.
So... The chip in that laptop was an x1400 - which, I mean, for the time, having discrete graphics in and of itself put you well ahead of most laptops, but that chip is one step removed from the lowest end chip Ati produced for laptops.
In particular, an x1400 (which only exists in laptops) is an x1300 with a memory bus that's twice as wide.
Anyway, as a side effect of installing unofficial drivers, I got access to a whole bunch of settings no laptop driver software would generally ever let you touch.
One of these was the desktop overclocking software.
And, that leads to the bottleneck in question;
See, in spite of having double the memory bandwidth of the otherwise identical x1300, (And yes, it IS about twice as fast)...
In examining the overclocking and underclocking results... Well... It got very interesting indeed.
First of all, the clocks could be increased by something like 80% and it still worked (this is without having any kind of voltage control).
Granted, if you overclocked both the core AND memory by that much it got unstable, but you could usually push one or the other nearly that high and push the other by about 20% and get it to run fairly stable at those settings. Of course, the cooling system here means the RAM has nothing at all cooling it and the GPU shares a single cooler with the CPU. (thus CPU load directly affects cooling performance available to the GPU as well.)
So... Anyway, the result that interested me here was the actual performance metrics in games.
Basically, altering the core clock did pretty much nothing in any games I tested. Overclocking it by 80% did nothing. Underclocking it by about 75% (anything less than that and the chip became unstable for some reason) also did nothing.
Nearly double the core clock, or cut it to 1/4 it's stock speed and nothing changes whatsoever in games of the era...
Meanwhile, change the memory clocks and performance in games scales almost linearly with memory speed...
Bottlenecks huh. XD
To be fair, Ati badly misjudged their GPU designs from that era. While the x1950 did really well, most of the rest of their range did not. With the x1600 being especially poor.
The problem is, Ati had assumed games would use a lot of shaders; Procedurally generated textures, vertex shaders, and so on.
As such, they had put a lot into giving these chips (Even the low end ones) a LOT of shader performance.
Unfortunately... What happened in practice was that games of that era doubled down on textures and similar effects that hammer the chip's memory bandwidth and texturing units...
And Ati's chips had not been built with that in mind. In other words they were over-specified for something games didn't really make much use of, and under-specified for things they did. And the low end, but especially the midrange chips were by far the worst off in this regard...
Anyway, just interesting sometimes what overclocking can do in practice when a bottleneck is involved.
You could find yourself with a truly massive overclock and yet the practical benefit is...
Nothing whatsoever...
on curve #2: power increases exponentially with voltage because as you increase voltage, the resistance of silicon falls, increasing the current non-linearly. meaning, at 5% higher voltage you're not getting 5% more electrons flowing per second (current) like a resistor would behave, you're getting say 10% more current, meaning 15.5% more power (U*I)
Your MS Paint videos are always worth listening to, hope for more.
eXXxXxXxXxXcUsEmEiT'sGiMp
Very interesting of point of view, no complicated to undersdand ! 10/10 Dudue :) TY very much!
i love this shit. Nice articulation at 18:15 saying power consumption doesn't go up too bad when the cpu is waiting for ram to respond, since the subsets of cpu logic in question aren't dumping charge into ground writing 0s with calculations.
This was extremely cool to learn and also very helpful.
Thank you young man...
Excellent word choice. Thanks again!
Very interesting as always! Will you be testing your 4 sticks of viper steel 4400 on the unify?
That'd be something I'd really look forward to!
Vega64 - all voltages at 0.900V, with HBM OC over 1GHz -> basically the same performance as stock (no boost), at just 150-160W power draw.
Not true, because the standard clocks at overclocked HBM is 15-20% faster, and you ate 20W in HBM. Besides that flat boost caused by Infinity Fabric behavior (not memory itself), games tend to scale linearly with clock. (At 4k.) So in many games you will get about 5% slower with your settings. (1500 memOC vs 1680 MHz 1.2V. You can run stock boost 1630 at 1.18V for 50W savings and minimal performance drop. Has space for memory OC.)
But about 100W cooler. Presuming the card is stable at 0.9V 1500 MHz, not all are. (Typical is 0.95V, and usually it is capped by IF bus voltage, which has to be increased to 1.025V or 1.05V to run that 1111 MHz memory clock to trigger 1200 MHz IF speed.)
Source: have 2x Vega 64 overclocked. They tend to random brownout and crash in specific titles when undervolted too hard. I can "run" stock settings at 1.135V, but specific games will reliably crash.
By the way, locking clock to P7 gets 5% boost stock. Card is also more stable allowing for more precise undervolt. The cards do feature some sort of LLC.
You've reminded me of one more graph I haven't tested in detail, it was more a small test, gpu sample of one. It was a long(ish) time ago, but I'm pretty sure that lowering the temperature of an R9 380 by 5 degrees gave me about 5W drop in consumption measured from the wall. Really don't remember the fan speeds, also it was undervolted quite a bit. Nothing to write home about, but it was measurable and consistent. Maybe you had more extreme scenarios.
Great video! Thank you and keep up the great work!
U realized us newbies... thank you ooh buildzoid!
And some memory chips won t even gain from voltage increase.Had micron A-die that wouldn't oc any further past 1.35v.
JEDEC specs allow for up to 1.5V to operate. Every stick should be able to run at 1.5V or else it doesn't comply with the spec.
@@dpokor well i havent tried 1.5, but tried 1.4 with loose timings and didn't like it, tried lowering the stock timings with increased voltage, same thing, the memory was corsair led 3000mhz cl15-17-17-35 2n and i could only change 2n to 1n and was stable anything else i couldn't lower them not even by 1 not even trfc by 1...Tho i could oc them to 3333mhz cl15-17-17-35 1n at 1.35v .That was the best i could do.
Second chart is Price (y) vs Performance (x)
Hey man, I'd be interested in a video that explains how different types of calculations or the use of other instruction sets (like AVX) affect the stability and the power consumption of a CPU. The voltage and the frequency remain the same so I wonder why AVX load is so much harder for a CPU to run. What do I have to look out for when overclocking? Do I have to test stability for all instruction sets if I want true 24/7 settings?
YES
You don't need to test all the instruction sets, only AVX.
AVX is special because it adds vector instructions - a normal add does A = A + B, but an AVX add is capable to do 4x-16x of these additions AT THE SAME TIME.
For example Zen 2 CPU with a 256 bit wide vector unit is able to do 8x 32 bit additions. Skylake-X has 512 bits wide vector unit, so it can do 16x 32 bit additions.
You get 4x-16x times the performance, but you also get crazy high power consumption (a lot of work equals a lot of power).
If you want more info, just google AVX or SIMD (Single Instruction Multiple Data).
Some other instruction sets (something like TSX maybe) will increase power draw too, but nearly not as much as AVX/other vector extensions do.
Well my 3700x boosts higher in non avx workloads vs r20. But I'm also not sure how accurate ryzen master is in reporting CPU frequency
Noob in overclockimg entimg.
Great stuff. Thanks for the insight.
some people shun on Downclocking, i call it, Summer Clocking!
Well liquid helium is less hot than liquid nitrogen just much more expensive per oz. So with luiqud helium you can get colder than with liquid nitrogen just it will cost you more money to do so for the chance of getting a little higher.
undervolting even seems to do me favors in overclocking, at least regarding turing cards. my 2060 hit like 1985mhz solid at 950 mv, stock being 1885 mhz at 1000-ish mv. i think that is when temperature comes into play.
I've played with subzero at normal voltages. I find that below a certain voltage depending on architecture it makes very little difference whether you're at 100 or -100 degrees. it's like maybe 50 mhz difference best case. I haven't done it with any recent chips though, so take this with a grain of salt, but I imagine most newer stuff will be similar.
Buildzoid maybe some Sandy bridge dry ice overclocking if they scale well with voltage :D I have a 2500k begging for you to destroy it haha
Thx for knowledge !
I've always kept voltage curves in mind
My mb tries to push my voltage much higher than needed though lol
As we know th left curve is tempture dependent on chip hotspots are ram id's flipchip if so how much die is on top of the transistors if it where to be shaved down would we gain noticable clock speed increases as it is very tempture sensitive. (I believe d8bauer did a video on this with a cpu)
Second are ram ics sensitive to cold tempture and if so is that because ram has tighter timing requirements than a cpu and changes in drift current velocity or the propertys of the transistor materials changes enough to greatly affect timing requirements.
memory is so low power that it wont matter much. you can use 2V or more and get away with a fan an roomtemp air. in a case it might be better to go with watercooling but thats more an exeption. over 1.45 to 1.5V you need a fan or some airflow over it. thats enough.
the other i dont know enough to understand what you mean, but ram tends to like 15-35°C so roomtemp is allmost best you can get. dont know why it wont scale with lower temps, but i think resistors and stuff. On Die Termination is a thing with RAM and a resistor that changes resistance might not be ideal. but again, thats just my idea of it. CPUs and GUPs are a one chip thing. if you had 8 or so on the same board that need to talk in perfect sync, it might behave simular
Nice video man. Something I have noticed is that with a lot of CPUs if you want to run at stock speeds, or even want a 100 mhz oc or something small; You can usually get away with a slight undervolt. At-least with Ryzen. Example being my 1800x. Stock it uses 1.4v, but it runs perfectly stable at 1.25 with a +100 mhz. I can probably get a little more OC at 1.25, but at these settings it uses very little power and hardly gets hot. I would imagine you could do this with any Ryzen. I really want to see how far you can undervolt a 3000 series Ryzen, at stock speeds, and run stable; and how much less power it uses. I think I understand why the stock voltages are higher than they need to be from the factory; It's due to a safe number across the line that guarantees that all chips run stable, because maybe there is a small percentage that will not run stable at 1.25v stock speeds, but they all will at 1.4v.
Yeah and the stock voltages keep the degradation over time in mind; Such a chip _HAS TO_ run even in ten years at the given clock/voltage levels.
Undervolting FTW - same performance, less power, less heat, less noise.
Buildzoid, this was indeed interesting. I have a number of SB i7 2700Ks I run at 5GHz, which for the 2700K isn't really that high (any 2700K on an ASUS M4E will do 5GHz with just a simple TRUE, one fan, not that much voltage and good temps), but today I noticed one of them is starting to need somewhat more voltage to remain stable, even though I know its frequency potential hasn't been pushed that much. I've had to raise vcore to the point where while idle CPU-Z shows vcore as being 1.475V, while a different system (same model of CPU/mbd, same RAM setup) shows only 1.456V. Years ago the degrading unit showed only 1.456V even under load:
valid.x86.fr/euhlxt
It's been used a lot for benching, while the other chip is still my daily tasks system and thus is used somewhat lightly. So what you said about older 32nm CPUs having a degradation point further back on the voltage/frequency curve chimes nicely. The CPU in question normally handles gaming workloads ok, but now it goes splat if I try a CB run or a Handbrake transcode, and today it barfed while running a mere Resident Evil 6 test (a few weeks ago it wouldn't bat an eyelid running FC5, SotTR, Hitman 2, GTA V, etc.)
By contrast, a different config (same model parts except for the RAM) when originally setup back in 2012 only needed 1.4V to remain stable: valid.x86.fr/show_oc.php?id=2334534
Thankfully for the benching system I have another hardly used 2700K I can move onto the mbd, so it should be ok.
So yes, it was useful to see how different architectures, indeed GPUs, could end up being by default on different parts of that behavioural curve. Is it fair to say that the older the arch the further back on that curve the CPU will be? Or is this something unique to SB? What about Nehalem/Bloomfield/Gulftown/Westmere? I have an i7 950, 975 and 990X I've not tested yet, also various XEONs.
Also Buildzoid, have you ever experimented with Lynnfield and Clarkdale? I have an i5 680 which, because of its high base clock, I am hoping to be able to push to at least 5GHz (not tried it yet); see:
ark.intel.com/content/www/us/en/ark/products/48504/intel-core-i5-680-processor-4m-cache-3-60-ghz.html
I have an i3 550 which easily runs at 4.7GHz on even a moderately average mbd (I used an Asrock P55 Deluxe), but I plan on trying the i5 680 on a far more potent mbd, the ASUS P7P55 WS Supercomputer, but what you described makes me wonder if the CPU could indeed scale just fine but may degrade easily even if it's initially perfectly stable and seemingly not that high up the scaling curve. Strange, I don't think I noticed before that the i5 680 and i7 2700K are both 32nm, somehow I regarded the i5 680 as being older.
Hmm, I have some new/unused i3 550s I've not touched yet; if you'd like one to mess about with, let me know and I can post one, though of course you'd need a P55 mbd to use it. They're a fun wee chip for oc'ing, the good old days of bclks, etc. Here's the one I messed about with the most:
valid.x86.fr/show_oc.php?id=2093028
Lynnfield was fun too, reaching clocks X58 owners could only dream of, especially the i5 760, but they were somewhat below Clarkdale or SB. My results:
www.sgidepot.co.uk/misc/tests-jj.txt
*Power consumption goes up by 69%*
Utilities company: Nice!
well im running my 1080Ti on .95 Volts since over a year now and only have 3% less performance from if id overclock it so yeah. Getting nearly 10°C lower as well and running 1900Mhz stable. I guess even most Nvidia cards that get shipped these days are just super inefficient on default than anything else.
I also undervolted my 3900x cause at full load its drawing way to much volts from what the cores are going to need anyways.
Morioh 1820mhz, around 0.9v goin around 200w power draw under 100% load
1080 ti strix
And Im sitting here with my 2200 Mhz GTX 1080ti with 1,250V,people are so diffrent.
how do you undervolt Nvidia since all the software I tried doesn't let me undervolt/ only overvolt usually...
Many people, even regular pc enthusiasts would be quite surprised at how much power their Intel chips would draw at something like the base clock or somewhere in the middle of base and boost depending on what chip /gen with the lowest stable voltage. Especially when the chip maker is already releasing the chips at or near the max of what they're capable pf doing on normal cooling. Like a 9900ks or whatever chip that has the 5ghz turbo.
I kinda like to know more , like some specific like voltages for like architecture , like you know when you o.c like moderate if the voltage like needs a big bump like you described or like it is safe to a point like you don't get like so much performance , like you won't notice more fps like almost without o.c then if I keep o.c like to feel more diff like real fps increase like 5-10 diff then maybe you can say like yeah this is like i like to have it because like i like the performance i get. Do you like get me ? 🤷♂️
Does the increased impedance in the cold metal, due to liquid nitrogen, also effect oc scores?
Love it bro what you think about MSI PRO B450M PRO-VDH MAX closest matx i could find to the mortar in the states would be fine for a 3600x planning on locking at whatever i get the frequency stable and temps good. The case has an absolute load of airflow for the vrms
Random thought on the frequency/voltage curve: could it be that by increasing voltage, the magnetic fields generated by the wires induce strong enough currents in other wires, generating unwanted voltages, and introducing instability?
I know this is old but that is a very interesting thought. I'm guessing not an easy thing to test lol
Btw the, the very telling thing here is that MS released an update to manage core usage in the OS better
Hi Buildzoid,
I have a question about the Frequency/Power scale you listed there. Based on my testing from highly power limited Max-Q GPUs in laptops increasing the frequency just with the slider on f.e. MSI Afterburner never translated to an increase in Power consumption (or heat), I got roughly the OC i dialed in. Where does this come from? Based in my understanding the powerdraw is hugely impacted by the voltage but if you just increase frequency, you will use the OC headroom a given voltage has for your specific chip (based on silicon lottery) and as long as you are not running into instability from too less voltage you do not use more power.
A quick explanation or some sources to read through would be a huge help! Thanks :)
Followed Hardware Unboxed Ryzen mem OC guide last night and got my Royal Bdie 3600 16 16 16 36 up to 14 14 15 14 28 @ 1.45v. On a taichi 3900x. My pc was crazy fast even with me rebooting to bios a few times. But this morning it looped, lost the settings, even the profile I saved in bios...
Next time I will pick calculate save instead of calculate fast...
Safe....
Could you go over a video explaining how/why a cpu wont clock faster when temps are not a problem? I have two 9900k's, both delid, one direct die. I can keep the chips under 80C but I hit a wall with voltages even though I do not get hot. I just dont want to pump voltages even if temps are low.....
Video is great!
definitely at a huge wall with my r9 290, I need +110mv for 65mhz more
and then there is ryzen 3000 and some special cases (reducing ppt instead of undervolting). got a 3700x on a b450i aorus pro wifi and 2x 8GB 3600mhz cl16 ram and when i drop the ppt from 88W to 65W (26% power down), i reduce the Cinebench C20 points by only 2,5% (from 4799 to 4682).
But what happens if you overclock and undervolt at the same time? I managed to get a 7820HK undervolted to -125 Mv while having it overclocked from 3.5 GHz all core to 4.2 GHz all core.
Nice words 👍
Hey dude do you know what NMODE is in the memory timings table? Default sets it to 2 and XMP Profile 1 sets it to 1. I googled NMODE and can't find anything. Thanks.
8:04 Liquid Helium 🤪
Btw hi der8auer
I`d like to see stuff like this about DDR3
BZ, am i crazy to run 9900k direct die cooling with sanded die at 5.2Ghz at 1.42Vcore on z390 aorus master? Is it any "safe" or degradation is goin on nicely? temps are around 85C at prime95 non avx
You'd think they integrate the heatspreader into the LN2 pot and have one layer between the CPU and LN2.
Solder to IHS to paste to LN2 pot works better than LN2 pot paste to die. It's only unsoldered CPUs that are really annoying to deal with on LN2
@@ActuallyHardcoreOverclocking would I be right in assuming that's because stock TIM at sub zero solidifies and kills efficiency?
In gonna start a channel called actually casual overclocking. Every video is "turn on xmp". The end.
Buildzoid I think I have a safer way to do something similar to the liquid metal shunt mod, graphite is conductive and it might work, I’m going to try it when I find a few used cards to test.
The safe way is find a little resistor to attach to the pads
So, is there any good, easy way to know when/where you're OCing into the the point of causing/starting degregation?
well it only speed up, but there are some voltage levels to keep in mind. problem is, that if you higher freq. that also speeds it up. 1.4V might be good for 5ghz cpu and ok for 5.2ghz. but another damaged his 9900k at 1.4v 5.4ghz. depends on architecture a lot. intel can handle more volt than amd. also more people means more "save level". on intel i would kepp the io/sa voltage 1.35V or below. some say 1.3V. older posts state 1.25V or even 1.2V. asus says 1.4V thing is. most people dont run 4000mhz+ on RAM. so they dont need 1.3 V or up. if you dont have 8. or 9. gen, you may cant even run that high. one chip might handle 1.4V good, another degrades after 1 year. if you dont want to risk it, then stay low. and allways stay as low as possible. there is no benefit to run with higher volts than you need
@@mauriceelet6376 you only need 1.15VccIO and 1.20VccSA to run 4000Mhz memory on coffee lake, but motherboard makers just couldn't care less about doing proper testing and lower the voltages: just like with core auto-overclocking they always overshoot the voltage at least 100mV
@Smucky ! There's no easy way to know, but there's some known guidelines:
-on Intel pretty much from Sandy Bridge until nowadays, up to 1.4V is considered safe (with proper cooling, meaning under 90°C with max load).
-on Zen 1st it's 1.4V, 2nd gen is 1.35V and 3rd gen is 1.32V.
Those voltages are for all-core max load with Vdroop, meaning those should be the voltages that are physically going into the CPU, not what you punch in BIOS or what SVID reads.
Hey there, great info, as usual. On the same topic, is the degradation point for zen 2 at the rumored 1.32v? Have you looked into it?
I don't have enough CPUs to gather proper data from but just looking at how my 3950X behaves I wouldn't apply more than 1.25V full AVX load to that CPU.
@@ActuallyHardcoreOverclocking Would you feel safe using higher volts for lower core count parts? Or does their lower binning degree make them leak more, pulling even more current, and therefore having lower number of cores doesn't actually improve that?
NEED MORE GIGAHERTZ DAMNIT.
Alright so is there a graph for the scaling of temperature vs frequency ? Or is that chip dependant ? With ambient cooling my 8600k will not boot at 5.4ghz even with 1.55 volts
that's very chip dependent and I've never tried to model it.
thats part of silicon lottery or as buildzoid said its chip dependent, but if you lower the temps you will get better performance and stability in most of the cases and also increasing voltages only goes so far as he explained in the video and your chip might refuse to give good results with anything above for example 1.5v
So I shunt modded my Titan xp with LM and I have my 9900k running 5.2@1.45v extreme LLC and I'm pushing 230w and 22A. Only once my vrm is over 80 does my chip hit 93 tops but 83c if I aren't running CB for long runs. I think my CPU is not happy running 22A peak.
1,45V BIOS for 5,2 AVX 2 or what?
that current draw reading is wrong unless that's input current not VRM output current
@@ActuallyHardcoreOverclocking sorry just saw this. Yes it is indeed "Current (IIN)" around 20-22A. Output is around 160A and 215-220w output and no avx offset for what it's worth. And it's the aurus pro wifi
What I don't understand is how I undervolt and get 1900 mhz @ 1.000 and when I'm "+110" overclocked my GPU jumps to 2005 at the same 1.000 voltage... but anything higher than 1900 Undervolt is unstable
Are you setting that manually or are you letting the boost algorithm still operate? Because if it's the latter, that's what is doing it.
Re: hotspots, good that AMD puts so many sensors in the GPU, but then messes up reporting. :) (Average rather than max in some firmware revisions.)
This actually helped me notice a bad paste job on my 5700 XT! Hotspot was hitting 110C(throttle point) while edge temp was still only 50-55C. Taking it apart & properly pasting brought it down to 80C hotspot at the same edge temps.
I never would have known this was happening without that hotspot reading, as the edge temps were fine!
Stuff like this actually helps noobs like me learn. reworking my 5.25ghz on my x5690 lol
Hey, i have a memory kit of Hynix MFR or m-die according to thaiphoon. That’s good for OC or not ?
Rahul Mahesh my xmp profile is 2666 16 18 18 36 1.35 volts. But i crank up to like 3200 with 1.4 volts. How can I test stability with 16gb of ram ?
Dram calculator max is 8gb and memtest is like 2.5 gb
It's the worst for overclocking. I have an mfr kit and it doesn't behave anywhere like afr. I'm running mine at 3000mhz 13-15-14-15-21 at 1.5v, and i got very lucky with the silicon lottery. It's literal trash.
I used to have HyperX Predator mfr, was able only to tighten the timings, even +200 mhz was unachievable with reasonable voltages. Needless to say I sold it
When doing LN2 OC how come they do not also put one of those pots on the backside of the motherboard CPU socket? Would not that help with hot spots cooling?
The fact that if the socket and pin or land grid cooling in the metal can shrink them back from each other breaking contact
The skill on handing 2 LN2 pots at the same time and finally "cold bugs"
The back is even further away from the hotspot, so it wouldn't help.
Fun thinking.. I'll guess if all computers (not OC'ed) were undervoltet to the/above *limit at stock speed (*still maintaining max stability) we would shave of 3% power globally. Think... All the datacenters, office/home computeres. That is a lot of kWh "wasted" globally in the world iright now! ;)
Think If each rig can shave of say 10% power consumption average..? New platforms with higher effiency can still (at least 98% of them) be tweaked to better effciency. All "stock" settings are preset to get stability to even the worst silicon batches.
I suspect quite a few datacentres may already be doing this as energy prices increase. It's also why ex-mining GPUs can be great value, as miners very quickly realised that slightly undervolting wouldn't hurt performance, but would drastically reduce power consumption, and thus increase the power/return ratio...
@@JaenEngineering I know. I have been mining a few years. All hardware still working fine (almost 24/7 running 2+ years) after best possible undervolting / efficiency tweaks possible ;)
@@Wushu-viking indeed. And that 24/7 operation also confers another advantage, mainly thermal cycling, or lack thereof. The daily expansion and contraction of parts as they go from ambient to full load (or higher if it's been OC'd) temp and back again is possibly one of the biggest killers, as the various parts expand and contract at ever so slightly different rates, placing stress on any interconnects. The only real downsides are thermal paste degradation as it slowly dries out and that the cooling fans will eventually die as their barings wear out, although neither of those is particularly expensive to fix, so should be easily outweighed by the savings .
@@JaenEngineering Exactly! Generally.. Blower fans are really durable for 24/7 running. (more so than the regular "low quality" Axial fans you find on "aftermarket" graphics cards) All my Vega 64 blower fans are still in good condition. Actually even the paste seems like its still okay regarding temperature. They must have buildt these cards quite good from AMD.
does this work for my printer
Is there a way to unlock power limit on 1650 Super?
I have noticed how voltage required rises really fast above 4ghz on my 1600x. It does 4ghz with 1.23v, 4.1ghz with 1.32v. I have ran it at 4.2ghz but i cant really keep it stable, best stability seems to be at 1.42v, going higher does not seem to help at all.
my 1060 is a bees dick off of falling off that vcore/clock chart, 1.093vcore wont hold 2215 but 1.075v will hold 2202. (edit) is 1.4vcore safe on a devils canyon cpu under a 360mm rad, its second hand so its not the greatest concern
is 85 dollars for 16gb of b-die a good deal
yes. i need to pay 110 for patriot 4000mhz or more 3200/3600 around 150. can go 200 and higher if the bin is not in stock etc
That is either shit bdie, like worst of the worst binning, or it ain't bdie. Can you link the kit?
Pffftt I bought my bdie 2x8gb ddr4 for $30..... MY MICRON BDIE 🙃
the 3rd graph is incompatible with the first two
well, considering when a logic gate is on or off, there is less power bled through the gate FETs than when it's transitioning from one of the two states. the more times a second a logic gate is changing from 0 to 1 or 1 to 0, the more often it's using more power in that state between 1 and 0 (cuz both the high side FET, and the low side FET, are not totally off or on. so the current goes through the high side and low side FETs from the power rail to the ground rail when the logic gate is changing state between 1 or 0). that's why at the same Voltage at different clocks the 3rd graph is so different from the first 2.
The two "Hz" labels do not represent the same thing.
@@fluffypink90 how's so?
@@EasyGameEh the third is powerdraw vs clock at constant voltage, the second powerdraw vs voltage at constant clock
@@effuah oh, sure
Nice video man!, how much voltage would you consider safe for a ryzen 9 3900x? just for games, I'm using 1.387v for 4500mhz
Well he just killed his 3700x today so I would probably ask someone else
@@michaelwerner5049 hahahah LOL xD
I would not push past 1.2625v mayyyybe 1.275v, anything over is simply not that safe. You have to remember that auto oc is only providing voltage in ms time frames. The best bet is to run prime95 small fft for three minutes and see where your sbi tfn2 lands. Thats your cpu's fit voltage then target an oc below that by about 0.0125v.
If you're worried about safe use PBO and consider a negative voltage offset.
I have run systems for over 3 years @1.35v on Intel systems i7 7700k stuff .
Dono about ryzen
Bullzoid soon you will be at 100K probably by Zen 3
👏👏👏👏 you saw the future!
3:08 we are not average human being
if i watch you're video which is about overclocking , trust me we are not
so say all the details i want to KNOW!!
Now i realised how good chip i have. i5 7600k running 4.7Ghz at 1.135V stable
Awesome stuff mate - very interesting - going to try to kill my i5 2500k 😎👌😂
MSI MEG Godlike + 3950x review? :D Love your channel man. Keep it up! Very interesting stuff.
do you see alot of reviews on this channel? your gonna be waiting a while my dude
9:00 lemme help u out bruh. Thermal resistance.
Buildzoid you published this at like 4am go to bed dude get some sleep lmao
Maybe he's not living in your time-zone ?
And even if he is, who cares? I also like being awake at the night. Its silent around you, its dark, nobody interrupts you, and you can simply do your thing :)
ok so we must find a way to decouple power and frequency. :3
why isn't this a feature in all motherboards? :3
I hope you are joking. Otherwise: It's not possible because its not a software problem, but a problem of physics ;)
@@StitchExperiment626 aaah i see the double -> :3 didn't work for everyone as a signal for sarcasm/jokes. i've been testing this a lot lately.
Why does this channel have fewer subscribers than Drachenlord?
hm. yes.
The proper reply would of been: "Verpiss dich etzerdla aus meinem land du spaggne! ich bin derjeniche der hier die macht hat!"
ich schau mir sowas nicht an, ich kenn nur einige hintergrundsachen. aber interessant mal jemanden zu sehen der sich sowas wie das hier gibt und solch ein Zitat kennt. wie hast du hier her gefunden?
A Gumbl hat mir a BehZeh Gombjuda zammengstellt damit ich meine ledsblehs (hadde abbeid) machne kann ferstehst du?
@@sinizzl so halb. aber was haben lets plays mit OC zutun?
i love your voice! trying to not make a gay comment. LOL
First
Goddamn you’re fast
@@avagreen9060 I have le notifications pop up on my phone ;)
@@C0manso That's basically cheating man
@@kommentator1157 thats definitely cheating.
35th
3900x ships with with 1.42v at 3.9Ghz but i managed to run it at 4.2GHz at 1.23V with level 4 LLC.
If you keep your CPU at factory voltage you more likely to kill your CPU early that it should be.