Most X670 motherboard VRMs are MASSIVE OVERKILL
HTML-код
- Опубликовано: 22 июл 2024
- My Patreon: / buildzoid
Teespring: teespring.com/stores/actually...
Bandcamp: machineforscreams.bandcamp.co...
Twitch: / buildzoid
Twitter: / buildzoid1
Facebook: / actuallyhardcoreovercl...
#X670 #ryzen #overclocking #motherboard #am5 
Наука
AMD sent em a cpu? LETS GOOOOO!
😄
The power density issue you talk about around 12:27 is the breakdown of Dennard Scaling, which started ~2006, even before Moore's Law faltered.
The other side of this coin (with temperature getting further disconnected from power consumption) is that a Ryzen 7600X (with a moderate curve optimize value [-15 or so], in gaming can be easily cooled [no throttling] even with a 80mm trash fan 2x heatpipe cooler). As a tester for an SI showed in a reddit post. He Techically showed
Moore's law was a self fulfilling prophecy as the whole industry invested to make it true, with ever increasing investments as VLSI generated industry wide growth and conquered new markets. Dennard scaling was indeed the driver for faster, cheaper, less power as smaller transistors needed less to switch and were quicker.
The problem is leakage became more and more of a problem. There used to be 12V processes that were popular in automotive :)
@@RobBCactive I can't remember where I saw it but I remember seeing someone make the argument that Moore's Law was basically just a manifestation of economics. You could have had more than double transistor count per year on average feasibly, but the cost of achieving that would have been too high every year to be practical.
@@KyleJohnsonVA Jim Keller interview with TechTechPotato?
His idea that Moore's Law was a succession of techs, lifting the bar when older ones exhaust their potential
@@Dogzilla07 Power density has increased, Angstronomics has an article pointing out the Zen4 chiplet challenge.
AMD design uses an IHS and solder to spread heat, but it may be that direct die cooling while impractical for a product will offer lower temps and performance at similar power.
Exotic cooling with a conductive seal over dies and liquid injection cooling may become necessary for some high performance chips
Looks like bz finally got a CPU lol
Nah ryzen 7000 series CPUs are not selling that well
@@Decki777 platform cost with forced DDR5, relatively high initial pricing, expensive X670 motherboards, it will get a lot better but honestly unless you're just looking for top of the line like with a 7950X it doesn't make sense yet. The CPUs are good but nothing minblowing that makes it justify the overall price. Idk what AMD was thinking here, they'll definitely have to drop in price.
@@mechano6505 New platforms are always like this. By the time vcache CPUs are out, mobo+DDR5 will be half of today's price and I can upgrade from my 5800x3d
@@mechano6505 they did $799 to $699. 24k Cinebench R23(5950x) to 38k Cinebench R23(7950x) is very impressive.
@@Decki777 Most people do a platform upgrade every 4-5 years if not longer and A LOT of people upgraded to zen3 during covid lockdowns. Demand for new CPUs in general is low. I reckon that raptor lake wont sell well either
I'm not sure why people still misunderstand the 95C situation. The power management system is configured to drive the CPU temp to 95C to keep the frequency at it's max boost. The extra thick IHS (for Z height) makes it harder for cooling systems to do it's thing. There is the 105w and 65w ECO modes or undervolting and underclocking. It seems ECO mode at 105w gets 90-95% of the performance.
I think AMD did something like this years ago on the stock radeon 290/290x gpus where they ran the temp up to around 90deg. These new CPUs have a stack more complexity with sensors than those old gpus, and can actually work within the temp and power limits much more efficiently in this new PBO mode.
I think they should offer a Windows/ Linux power profile which allows users without messing with bios or OC software a means to run within a more conservative power/ temp limits curve which can still achieve say 90%+ performance, at around half the power draw.
@@TheAzzzzzzzza really? I have a 290x and it runs just fine, yet again, is a sapphire with a ridiculously big cooler, but it didn't ever go over 80c when I used it.
@@TheAzzzzzzzza Almost. The GCN cards just ran WAY more voltage than they needed (sometimes 150-200mV more) to get those last 25 Mhz out of it. They also didn't give them adequate cooling except for some AIB models.You can drop temps by like 20 degrees just undervolting a little. It's actually insane.
Curve optimiser and power limit lets you to have the same performance with much lower temps and power consumption…. Problem solved…
Which is shit, because everyone knows performance doesn't scale linearly with power draw. There is massive fall off at certain points. All this is does is inflating power draw into oblivion and thus completely butchering efficiency. For what? For fucking 5% performance gains at the cost of doubling power draw. Enabling ECO mode or (if possible) undervolting and tweaking PPT, EDC and TDC is almost a must, just because AMD desperately wanted to squeeze every last bit of performance, because as I suspect they wanted to avoid embarrassment of having generation leap in teens. This boost limiting by TjMax allowed them to go above 20% generational performance leap (even if barely) and in public eye, >20% looks so much better than something in teens range (likely only ~15-16% if they used Zen3 level of power limits).
I'm amazed that the CPU is doing 4.725GHz while running Prime95. And "only" using 175A. That is awsome. Love the vid btw BZ. Keep up the good work.
Is that 175 A for the whole CPU package? If so - that's not even 200 W, which is odd, since this CPU has been measured at 250-280 W.
@@VioletGiraffe At 3:59 you can see CPU Package Power at 230w, 173A, which translates to aprox. 1.32v vcore.
@@VioletGiraffe Check your mathematics. VCore fluctuates around 1.2V. Power is voltage * current. 1.2 * 175 is 210. The lowest VCore voltage we see is 1.116V, which is 203W at 175A.
@@HanCurunyr Actually no, because Vcore isn't the voltage of the entire CPU, it's the voltage of only the CPU cores. The SoC voltage (and also a very small additional voltage as a result of the resistance of the CPU substrate) applies separately. The real SoC and Vcore voltages are displayed in the video. SoC voltage is conincidentally about 1.32V, but Vcore is lower.
@@HanCurunyr, thank you. I noticed 1.1 V indicated which led me to the wrong number.
Please do a video on the best efficiency VRM for the B650.
Yup, a B650(E) VRM roundup would be awesome!
All you need to do to hit 200A is delid to directly cool the dies with liquid metal TIM, and likely break your 7950x in the process
You make it sound normal
@@cldpt It might as well be normal considering Der8auer's findings on how much the thicc AM5 IHS is restricting thermal conductivity.
I’d feel safer lapping the IHS down to a couple hundred microns thickness.
I’d want to know how effective this is first. If AMD made the IHS as thick as it is to maintain AM4 cooler compatibility that was a poor decision in hindsight.
Buildzoid is a clever guy, if he uses the right tools, he can probably do it without killing his CPU.
@@BurnsRubber I believe AMD did make this decision on the basis of keeping compatibility with AM4 coolers.
Personally I’m wondering if they will use this surplus of Z height for newer packaging technologies like 3D Vcache.
Can't wait to see how ryzen 7000 handles ram overclocking. Thanks for the vids.
I'm in the B650E + 7800X3D waiting room
I'm still on my i7 3770k from almost 10 years ago and that's what I'm waiting for also but hopefully a 7950x 3d
Not sure if you should pair the CPU with a "low end plattform".
A lot of the cheap boards say "i can do this" on the sticker but in reality it means "yes i can boot but not do anything"
@@TheJazzyjake1234 I'm not sure if they'll do 3d variants of parts with the cores split between two chips since it could introduce more latency in a part designed to reduce latency and that's a workstation oriented product opposed to a gaming one like the 3D.
@@TheNerd are you suggesting there might be some issue pairing a 7800x3d and b650e board? Why would that be the case?
@@ffwast There will be 3 SKU with 3D-Vcache. 7800 , 7900, 7950 .
It's damn annoying that all motherboards particularly the X series now have far more stages than necessary adding unnecessary expense to the costs which translates to higher prices all around.
Realistically 8 phase Vcore with 70A stages should be more than enough for 250W power draws and anything over 12 50A stages is just wasted money. I blame the vrm temp tests done by reviewers which made it a competition for the coolest VRMs being better than ones slightly warmer. So now to ensure getting a reviewer pass it's all overkill on both stages and heatsinks. I like the fin designs with heat pipe links to even out the VRM thermals, but there really isn't any advantage to a VRM that runs at 60°C vs one that runs at 80°C under load but the latter will be placed down the list of winners by reviewers, which will affect end users buying decisions.
It's all good and fine to have your sub ambient motherboards with VRM overkill, but if the boards got 4 memory slots, it's not designed for LN2 and has no need for 16 phase or greater Vcore VRMs. 8-12 Vcore is more than enough.
Let's hope that motherboard manufacturers release more practical high end boards with a focus on balanced features and a range of price points.
👍
A 15 minute video? Where's buildzoid and what have you done with him?? You monster!
00:19 I love how you avoided saying AIO but said All In One Cooler
This is great. It means when they bring out 32 core desktop CPUs in 2025 the x670 boards will still handle them. Remember these boards are supposed to last a long time. So it good they made them overkill
dont forget 6.0GHz base clocks lol
All AM5 CPUs will need to work in the same chipset generation - I'd guess B650 VRMs were also massive overkill and will handle a 32-core 9950X w/PBO no sweat.
Why not 64 cores / 72 threads for 9950x, 16 big core chiplet plus a 48 “C core” chiplet.
der8auer's latest vid has shown the effects of the latest IHS with a de-lidded and on-die cooling CPU. He noted that the IHS is around 1.2mm thicker than necessary for AM4 cooler backwards compatibility.
Don't forget we still have the V-cache models incoming. They might need that extra margin to stack the cache layer on top.
comparind direct die liquid metal to thermal paste on top of an IHS isn't at all fair to the IHS.
@@VTOLfreak that's what I been saying! I doubt that whole cooler compatibility thing tbh
@@VTOLfreak vcache isn’t taller then normal dies. See 5800x3d delids.
@@Ground15 the new 3d v cache is an improved design or 2nd gen
It could be spatially taller
@Actually Hardcore Overclocking, i bought my Thermal Image camera (UTi260B with Macro Lens) directly from manufacturer (UNI-T) on Aliexpress for just 16,3K RUB (with free shipping!). To convert it on today cost in GBP it will be just ~242! It's very afforable and really not bad. There also few reviews on RUclips. Macro lens help to view individual small components. Minus of it - it can't record videos, only take screenshots, but TBH i didn't tested their software (it's possible to connect thermal camera as camera via cable).
🤔
Have fun with that CPU, that curve optimization seems pretty compelling.
Good to see you finally got your 7950x . :-)
Hey, I'm happy you finally get your Ryzen 7000 CPUs :D
Have the motherboard manufacturers recycled designs used on high end Intel boards, which have had to deliver considerably more power in recent releases than the 170w TDP and 230w PPT of AM5? For them reduced design cost and marketing their overkill VRMs to overclockers must be an attractive strategy with higher margins than pairing costs and targetting the value sector, with top tier X670E boards.
Similar recycling of cooler designs lead to 200-220w GPU cards having the same coolers as much higher power consumption designs on the "bigger is better" principal.
So with this board being 16+2+2 which shows the 16 being 8+8 phases parallel, how would it stack up against the MSI PRO x670-P which is 14+2+1?
These boards are built for the 32 core next generation. Well at least 24 cores but 32 makes more sense. Or a big core little core setup
32 core 64 threads?
My man's getting hardware directly from the companies now! This is awesome to see, and I hope you get more and more review support from these companies. If they want their stuff truly reviewed and tested, Buildzoid is the guy they need to talk to.
He's gotten hardware direct from manufacturers for a while now.
Great to see you with a chip to play with Mr Zoid.
Any thoughts on the Asus X670E TUF Gaming and a 7700x? Feature wise it has everything I need at $320, I am just wondering how the 14+2 (70A) will hold up.
overkill for 7950X for sure.
but AM5 should in theory support future Zen CPUs, so if AMD ever deliver 32C Zen5/Zen6 CPUs, it might not be overkill anymore
But that probably wont happen due to socket size, the 7950 is already cramped. That's why thread ripper is such a big socket, to physically fit all the cores.
How you want to cool these CPUs? The won't get over 250W no matter how many Cores ZEN6 will have. 250W for CPU will be the Maximum so this Board will be Overkill forever.
He literally addressed this in the video. He even predicted this very comment and said why it's not relevant.
BZ HAS A 7950X AT LAST
REJOICE
I wonder what amd sends partners when they first sample out chips for a new socket. Is it like just the socket and the chip and some notes about what to power it with? Or is there like some sort of reference or test vrm?
Some reviewers (I think it was Hardware Unboxed) mentioned that they sent them very highend RAM to test with as well.
0:14 7950X provided by AMD?! Thank you Lisa! =)
Didn't you have a bios guide for the Asus ROG x670E? I cant find it
I'm not sure what the Aorus Master means when it says 16+2+2. Why the last +2?
Is it better than the ROG Strix which is 18+2?
Thanks for the info. I was just thinking if I should care about power phases.
This wasn't ramble-ly enough mate, you came in under 15 minutes. Great content as always regardless. Cheers
Thanks for sharing. Now waiting for b series.
I had this cooling and it does not knock you on your knees, did you try to mount efficient fans on this AIO, e.g. delta set them to 100% and check the CPU power consumption again.
I know the chances of significantly higher CPU consumption under these conditions are slim, especially because of this damn thick IHS.
But I know that with direct-die the temperature drops even by ~ 20C down and you can easily exceed 200A there
I wish I could single handedly sponsor you testing the voltage regulation on all AM5 motherboards. I wish that, then I remember I don't have a job... >.>
I have some board and i got bug like in this video. There are no chipset temp in hwinfo.
For me it's not the VRM's anymore but rather the connectivity. I'd like to have 3x M.2 without compromising the 5.0 x16 slot so that leaves few options, and I don't want to use a PCIe card either because I have a vertical GPU in a small case.
👍
What do you mean "that leaves few options"? Most X670E motherboards support this. Unless you need thunderbolt/USB4 as well, it really isn't difficult to find them.
The X670E PG Lightning is the cheapest AM5 motherboard with a PCIe 5.0 x16 slot (at least in the UK) at the moment, and it has 4 M.2 slots, none of which take lanes away from the main PCIe 5.0 x16 slot (one of the slots is 4.0 x2 and one is 3.0 x4, but this really shouldn't affect performance, as your tertiary drive would be bottlenecked by being connected via two chipsets anyway; and you can just populate both of those slots if you're really obsessed with maximising your storage bandwidth)
so now you have cpu. Good, we get to know things.
And you can do a guide on ALL DDR5 timings, since you love that topic ;P
Watch his intel 12th gen ddr5 voltages overclocking vid for ddr5 volts and his DDR4 explained videos to get a good idea of how timings works even for ddr5 to get familiar to tune them. Also for memory stress testing use linpack xtreme or tm5 with that anta777 config file or hci memtest.
Will B650 Aorus Elite be sufficient for 7950X then?
Nice short vid bz!
Any plans on streams, delidding and the stuff ?
Overkill is what we want for AHOC
Unlimited Powaaaah🔥🔥🔥
I have yet to see anyone talking about the MSI ACE. Why?
I was trying to decide between Aorus Elite and TUF Gaming, based on the x570 VRM both were very good VRM wise, so I picked the Elite only cause it shipped (inter) faster. The TUF Gaming had a delivery date of Dec 3th and I find that BS. At least the elite is circa 15th October.
Overbuilt now for the next couple gens of cpus?
What's nice is these overkill vrms will make upgrading the cpu much easier. If 8000 series ryzen requires even more power(doubt) at least the vrm is prepared meaning these motherboards will last users much longer in the lifetime of AM5.
Yeah that inevitably became the case with x370 and some high end B350 boards. It took a hell of a long time for x370 to get those Zen 3 updates, but damn it, it did.
isnt 7950x already hitting socket power limit? I doubt that 8000 will draw more power
Next gen desktop will be called 9000 series probably
Not really. Like BZ says you cannot cram much more into this CPU package. There are limits that you cannot bypass. Thats why Tredrippers are much bigger cpus physically with much bigger socket. Those overkills don’t matter. If 8000 takes more power you could substitute a kettle with it and make tea.
Since when is over capacity bad on a mobo? Glad it's ready for the future. Don't want electronics running on the edge, it leads to failure.
Raptor lake or 7600x…. Hardcore gaming and oc - what to chose?:p
been looking at the GB B650M Aorus Elite, and noticed it has smaller VRM power stages than most other boards, 12+1+1 and a single 8 pin cpu power connector. Is this still enough for 7950x? what about possible future cpu's? I may want to upgrade my cpu in a cpl years to get the latest & greatest and get more life out of my system, but dont want to have to upgrade the board too at that same time because I skimped early on with power stages......
It is more than enough.
How many Phases for up coming Intel 13900k OC?
If you punch in a negative curve optimizer offset that would reduce the voltage but increase the amperage. Would that cause the VRM to heat up more?
well it would if that actually happened. What usually happens is the current and voltage stay pretty much the same and clock goes up a little.
As for the Comment about a 32 Core AM5, I doubt you will see a 32 Core AM5 CPU, but a 24 Core? that could very well be a thing, the 5nm chiplets could house 12 Core version. though you'd think we would have heard about it.
I am holding off till there is a Threadripper 7000 Series CPU Before i upgrade. the 5000 Pro didn't do much for me, i was hoping for a 5000 Non pro series to fit into TRX40 boards, but AMD disappointed
This is good title
5 years ago : this vrm is shit
today: this vrm is stupid
mobo manufacturers: Yes
Now the intel starts comming in. I knew there was a reason im subbed here. If you want details, this is the place.
I was initially planning to never use an AIO again and just use air cooling, now I’m not sure.
I am interested to see how the NH-D15 fares against the 7000 series. (I mean how hard the chip pushes itself in frequency, not if the cooler can drop the temperature of the CPU.)
Several youtubers have already tested it. With a basic $35 dollar "Noctua clone" tower cooler you get 95% the performance at 60% the power draw. It still slams straight to 95C though.
Hardware Unboxed has promised us a 20-cooler benchmark next week, where they go from the most basic Wraith cooler to the biggest AIO.
@@andersjjensen oh nice
@@Ultrajamz Paul's Hardware just uploaded a test of the Wraith Stealth cooler if you want to see.
@@andersjjensen curious if underclocking plus air cooling will be a good option
J2Cs theory on the cooling problem is that it's because they're using a thicker IHS this time around, to maintain the same Z-height as AM4 for cooler compatibility reasons. I don't know how much of it is that vs how much of it is the node density, but I'd be willing to believe it's a mix of both.
No way. Coolers compatibility is the last thing is to do with this. That's noob explanation. Same Z height doesn't mean 100% compatibility.
@@AuDiGo6 check out der8auer's delidding video on the 7950x. They ended up making the IHS thicker to match Z-height but discussed how that really didn't make sense because of other compatibilities problems. J2C did end up using an older mounting solution though so some amount of compatibility was preserved. Der8auer ended up getting -20c temps by cooling the processor directly. However I can't say that solely points to the IHS.
It's also difficult on this chip to know that it only has a -20c difference because of the IHS based off of the fact that the chip tries to reach 95c. Lets say direct cooling contributed -30c to the processor but after being cooled adequately it was able to pull more power and raise another 10c. With J2C seeing the ability to overclock another 400mhz out of the box I'm very curious to see how this all plays out.
the thicker IHS could be to futureproof that height for their 3D V-cache, specially if they expect that tech to be the norm in next gen CPUs.
so right now is a thicker IHS but soon could be a thicker chip and a thinner IHS
100s of engineers at AMD probably though about this but If this becomes such a big problem, In the next gen they can just use thicker PCB with thinner IHS. (Since socket and dimensions will be same until 2025 at least)
@@endless2239 its for am4 cooler compatibility
What are your thoughts on a CPU operating temperature of 95°C
Reads tile. X670 massive overkill. Me: That's a good thing.
How did you attached the EVGA AIO to the new AM5 socket ?
with screws.
serious answer: the EVGA 280 CLC is a generic Asetek AIO and the AM4 Asetek mounting hardware works on AM5(which is the whole reason AMD made the IHS so damn thick).
@@ActuallyHardcoreOverclocking thanks matey, i asked because AM4 EK supermacy block is allegedly not going to fit the AM5 socket without another mounting kit - which is NOT AVAILABLE YET and will be in 2 Bloody weeks time.....sour
If you need a cheap thermal camera get a CAT S60 phone. I have one too an the integrated thermal camera ist very useful. The phone is pretty old so it is not expensive anymore.
AMD seriously needs to send this guy a CPU next time if they want good review of their CPUs
I don't do gaming tests and that's not going to change.
@@ActuallyHardcoreOverclocking Always very honest with us normie students learning the arts of xoc, never change teacher Bzoid!
Hope it is a golden sample :)
good to know my asrock steel ledgend will be more than enough. even though its a bit ugly not sure whats with the white camo look on so many boards this gen black would have been fine
Borderline dislike that idea of prioritizing storage capabilities over price on the x670 chipset
Patiently waiting for B650 Unify-X or Master
Getting the scope after all? 🤔
I have 12 SSDs and 128GB of RAM in my System....
@@TheNerd talking gen5/gen4 but can see you are a file copying enthusiast. Keep up the good work man 😅
scope isn't available in UK yet.
You're not running cooling, but it seems the mobo is out of the case, or at least has the window open. Won't that make a noticeable difference? (definitely not saying you're wrong overall, just that maybe mid to high 70s shouldn't be what people expect)
if it was in a case it would get a lot more airflow.
How much VRM u need? As Much as Possible. Personally, I'm amazed at how well-equipped the VRMs are... In plain language, this means that all X670 boards have at least 6+2 IR P-Stages and some have Phasedubler. Let's see if the B650 boards don't have such oversized P-Stages. Which phase controller do you have there?
I will be looking forward to the AMD GPU breakdowns. I hope you do at least a couple
Laping IHS and Direct-Die cooling is going to become a think again hahahaha
Mi ITX board has 100amps and the power limit is set to like 400+w
and me about to get the ASUS ROG CROSSHAIR X670E Extreme vs the ASUS ROG CROSSHAIR X670E HERO as VRM were almost the only difference... Thanks!
Please do this type of video through the asrock boards
This is good.
Can you use a good air cool for the new amd cpu?
Yes 200W+ xD
@@chovekb ok thanks
Given the extra heat of the thicker IHS, and limited space, I wonder if making AM5 work with AM4 coolers was actually a bad idea. I worry that until AM6 any chip will have a limit because they all run 10-15c hotter and there isnt enough space to really expand past 8 core per CCD. So do we want AM5 to be long lived, or do we want AM6 to come out asap where they built the package around their goals, instead of adjusting goals around the package.
🤔🤔🤔
But what's your ambient temp.
21C
Well in some circumstances, overkill is better than being rightly controlled well calculated especially inside a situation where you can control something flexibly
The problem is money. According to DigiKey, each of those smart power stages cost about $2.00 for the chips alone! Plus passives, doublers, more expensive controllers, and everything else. So, we're easily talking $20 to $50 worth of parts cost, plus the added design cost.
Which means the boards likely retail for $100 more for something that you and I will never use!
Overkill by a few percent is something and it's not really overkill, you can still call that being cautious or being on the safe side or giving headroom. Overkill by 50% to 100% is unecessary and against our interests as consumers.
While I agree that AM5 has overkill power delivery for Zen4, I disagree that there's no way AMD could increase the power needs for future CPUs. And AMD specified the maximum socket power consumption to avoid the situation that it got when going to 16 cores the scaling suffered because it hit the power limit specified by the socket specs and had to reduce the frequencies by default.
There are TSMC technologies that allow the following changes for future products some of them were the Hot Chips papers in 2021 and others are possible with older technologies. 1) The area made for IO die also has the same power density as CPU dies. 2) Change the package design in to cool it more efficiently. 3) Put capacitance inside the package without taking surface area. 4) Increase the size of the CPU chiplet by 25% with a similar increase in CPU chiplet power consumption, maybe by eliminating the gap between CPU chiplet and IO-die.
I don't know what is a reasonable change from the cost efficiency point of view since that would require confidential TSMC information which I don't have, but when there are at least three potential avenues for it AMD needed to specify the socket in a way that they can start using one of those in future products.
As for the core count, there are rumors of Zen5 having an increase in the core count. And that makes sense with Threadripper being squeezed out between two different product lines which each are better to different segments. After the mainstream platform got 16 cores for one segment the reason to go for Threadripper went away, Threadripper PRO on the other hand increased the cost beyond enthusiasts aiming for cheaper Threadripper systems and took the workstation market away. AMD probably calculated there are not enough people to have a high enough volume for that platform. Also, with Intel back in the core count competition, there's a marketing benefit for doing a halo product for the mainstream platform.
It seems that AMD plans to make AM5 replace low-end Threadripper systems, and have Threadripper PRO replace high-end. With Vcache and DDR5, the benefits of having extra memory channels are reduced and with all the IO options of different high-end AM5 boards there would be a really specific IO need to go for Threadripper because of IO, and most of those cost enough themselves that Threadripper PRO would be a real option for that use case. It's no longer the mainstream platform is limited I'll get a bit more IO just in case.
One potential avenue for higher power draw that I put low consideration would be doing APU with just replacing IO-die with something that has a far stronger GPU inside of it while keeping the chiplet design.
In your statement about cheaper boards doing better in VRM performance, Very true. Sometimes its better to get the more stripped down boards in terms of features, because you will find that the stripped down boards concentrate on the VRM rather than how much extra crap can we throw onto the board. You see boards with 10G Ethernet, WiFi 6E, 400 M.2 slots, they have to put alot of development into laying out the board to accommodate these features, which take away from the primary features like VRM quantity and quality. I use a MSI B550-A Pro for one of my systems, it has basically no features, no additional add-ons but for a cheap B550 board, its VRM is spot on, i can run a 5950X on it with PBO set to manual with custom wattage and Amps that the CPU can draw, and the VRM gets warm, like 70-85c, but it never throttles the VRM.
I've also seen cheap boards be absolutely horrific. There's little to no consistency when it comes to hardware.
@@ActuallyHardcoreOverclocking that is true, though if you do take a look at the board you can try and make a bit more of an educated guess on what will be good and what wont. Lets say a B550 board and you look at the VRM and you see what amounts to a 4 phase VRM with no cooling on it? that there is a pretty good sign to stay away from that one. When i was looking for a B550 board i looked at quite a few motherboard images and did the best educated guess i could when it came to the selection. MSI's B550-A Pro i could see what looked like a pretty good VRM (i counted 10 phases) and a hefty heatsink. I've gotten pretty good at picking boards, had like 10 years in Hardware Reviews for Tweaktown, so you pick up a few things lol
@@camjohnson2004 B660 situation proves BZ's take.
When cpus are running 95 degrees, don't you think that's quite concerning in the long run? Would it not be wiser to undervolt it? (Im stupid please be kind)
ryzen 7000 arhitecture is set default to reach 95c it boosts to the max to that, its made to resist 95 because 5nm and amd arhitecture, the 5000 or older and intel cpus fry if used over 80c for long time even if fans started saying but now intel resists 100c too, you can choose ryzen 7000 to run at 60/70/80 and it will be the same moslty because they are easy to cool and if you have the strongest air cooler you can run all 7000 without a problem
@@Noooo23523 thank you for the reassuring answer :)
I was wondering if it is possible that while these x670 boards appear overkill for the current set of chips, that perhaps chips coming out in 2025+ that can be used in these boards might make the VRMs more reasonable? Any thoughts?
Weren't the B150/X170 boards pretty trash? I'd definitely prefer an overbuilt board that can handle the entire AM5 lineage to a trash board that gets scrapped next gen.
@@benjaminoechsli1941 A320/B350/X370 was the initial boards from AMDs AM4 socket, these were not that good of quality compared to later boards that came out for AM4.
He adresses that in the video.
I got a taichi because i thought the massive 24+2+1 power delivery would be useful but now im starting to realize that its pretty unnecessary, even factoring in how long the board will last if its never going to have a hard time. The board has a lot of nice features but i wonder if i should return and go with something cheaper, or just let asrock keep the extra 100-200 dollars and look at it as an investment for future CPUs.
Id take a look at steel legend. Level1techs reviewed both the tai chi and steel legend and I kinda think the cheaper one might actually be better for most people. Unless you need the x8 x8 pcie slot layout. Honestly I'd pick the steel legend if I had a choice between the two, and I'm usually a fan of the tai chi.
Marketing got you😂😂😂😂😂😂
Overkill is the way to go
Grats on getting your preferred chip. 🙂
How much difference does temp make to VRM efficiency? For CPU/GPU efficiency, temp makes a notable difference, but I really don't about VRMs.
Given the global quantity of such components and lack of renewable energy sources, I think we should care more about efficiency than we do (and yes, I am aware of the energy requirements of producing and recycling aluminium, so it is quite the balancing act).
at least for the switching transistors the same cooler= better should apply, probably doesn't matter that much though.
and those working on making the stuff do care about efficiency quite a bit, just look at how efficient current psus are, and even the vrm's are 90+% efficient.
Power sources wise we really need nuclear, that is the only really good source.
@@cj09beira The actual engineers may care, but the execs don't, hence why CPUs and GPUs are being shipped beyond their efficiency peak, in order to have bigger performance numbers than their rivals.
Nuclear's certainly not the only really good source, even including thorium. They require constant consumption of fuel.
Wind, tidal, thermal differential and solar are quite different and use the energy from the sun and moon, which have far greater caps than elements on earth.
Energy (and the disposal of heat) will always be significant limiting factors to our productivity, thus it's important we try to ensure efficient use thereof and minimize waste (which we're pretty bad at, especially in this "more GDP" [un]economic mindset state).
@@ChrispyNut there are lots of execs that care about efficiency, say every single one that has servers energy / cooling costs eating away at their profits, every single power supply manufacturer as its another great marketing tool, oems care quite a bit because the less they need to power the better, the reality is that there are few places where its performance at the cost of everything else like how it kinda is on the consumer diy market, and even here the cpus will clock to the max to finish the workload as fast as possible then they do the inverse they race to sleep, power usage at idle has been falling / improving every gen.
@@cj09beira No, they care about reducing their own costs and that means that they skimp on their products as much as they can.
I've worked for a few corporations over the years (though mostly been small businesses) and the amount of waste is offensive, because they're not interested in the long-term investment because of the capital expenditure required in the here and now.
There's no interest in being a good actor in the world, only whatever's most profitable in the here and now. Heck, most of the long-term investments are simply for tax-writeoff purposes.
This is why things like carbon taxes are essential, MOST businesses only value currency, we need to make "environmental costs", "financial costs" as that's when we'll truly see efficiency matter and have a chance of providing future generations a reasonably hospitable planet.
To use an on-topic example. Intel requires a new motherboard with every [other] generation of CPU, but that's to increase revenue, it's not technically required at all, they could design sockets and chipsets to last several generations, which would be far more efficient, let alone all the segmentation of their product stack, forcing people to buy more expensive products than they should need, just to get a feature or two that's locked out of the lower end products.
Late-stage capitalism is a disease for this planet and our future.
Big thanks to you 😂 I'm returning my 16 phases asrock b650e instead of getting a b650 gaming x none ax version which is on sell right now and 8 phases mobo . I don't give a fuck that if my 7950x is gonna kill my mobo any fucking more
Interesting that there's only one VRM temp sensor exposed, given gigabyte on the x570 boards exposed both the Vcore and VSoc separately, and also had two sensors per, with the VR-Loop1 and VR-Loop2 sensors.
Unless it's just further down in the display that's not visible...
new VRM controller isn't supported by HWinfo yet.
@@ActuallyHardcoreOverclocking Martin is slipping! lol I wonder how many sensors will be exposed from the controller itself, once supported...
Inside case the temps raze 10deg at least.
There is a good chance that many B650-motherboards will be like a proper X570 motherboard, both in regard to quality as price. So let there be a more expensive X670 motherboard at the launch, that price will become more reasonable ($250-350?) for many motherboards and B650 will be fine as long as you don't intend to have a lot of SSD's running at the same time with the highest speed that the fastests SSD's can handle. I expect proper VRM's on B650 (the CPU's use a lot of power out of the box) but also I am convinced that most people will like their Zen4-CPU the most when setting a lower power-target and undervolted. The singlethreaded performance will remain the same and the multithreaded performance will only decrease a little bit when you undervolt a bit and set a powerlimit of let's say 105 W. Overclocking can be fun but for daily usage less heat and noise can be more fun.
The big problem with this launch of Zen4 is that AMD allowed motherboard-companies to only have their more expensive motherboards available. This is the 2nd time, they also did this with X570. I suspect that motherboard companies 'asked' AMD to do this, to sell more of the expensive motherobards. After all, the motherboard-companies had more affordable motherboards with PCIe5 ready for Intel for quite some time.
Everyone is releasing top tier parts first, lol. Intel, AMD, and NVidia. Just wait it out.
for future upgradeability
I can't be the only one who's continually dealing with OCD flareups because manufacturers still can't get cooling right on their own.
It's just asinine at this point that we aren't machining products to have similar thermal performance to if they were lapped and or had proper TIM between the IHS and die.
With the money going into each CPU purchase, is it really that unreasonable to expect $10-20 worth of machining and thermal refinements?
Every IHS and coldplate is slightly different.
Also, $20 of machining is $50 to AMD, and every $ saved counts for a NASDAQ listed company.
@@Wasmachineman That's why we need to stop being ultra 1984 paranoid every time regulation comes into play. Semiconductor manufacturing should be efficient, to the best of the ability of the engineers who design them. No shortcuts to get a quick burst of cash for the investors.
Every manufacturer should play by the same rules.
After all those years of R&D for the actual silicon, I personally think it's downright insulting to encase it in some aluminum that barely makes contact and has a piss poor glue job that you can visibly see from the exterior.
@@skorpers Ethics and morals take a back seat when it comes to earning money, that has been the standard for centuries by now.
@@Wasmachineman Well there's never been a better time in history to vote with the wallet.
Diminishing returns across the board. The only customers that could possibly make a good argument for more would be the server customers buying in bulk for large corporations, and even then, those types tend to want to prolong upgrades as long as possible due to the very nature of major system changes, they would benefit from more careful manufacturing as well.
"Big Money" needs to just collectively stuff it.
@@skorpers This is true. Buy used and vote with your wallet.
tech yes city got the 7950x running at 55 degree's
that's good news :D
Try this on a620 please
Is this with an undervolt? Check out optimums tech video.
Normies do that, not here wrong channel mate.
This comment is provided by the Keep Calm and Raise Vcore gang
@@emini6 Nah. Right channel. This comment is provided by the Keep Calm and Decrease the Vcore gang :) Nice display pic btw.
@@joshc7200 thanks, me & my mom are big peanuts fans, more hardcore & based then anime fans. Feel free to donate to buildzoid's patreon and/or get merch like the keep calm and raise vcore hoodie or t-shirt so BZ could get funds for his new awesome oscilloscope for us Vcore raisers gang are hoping to see epic results.
Also LLC Vdroop + manual voltages on core, soc, io, ddr is the way to go always, say no to auto stock voltage settings.
Plus don't upvote yourself lmao
Its time to bring sanity back to pc parts. That would bring the price of the x670 down to a more reasonable price.
AMD and Intel should just start selling delidded versions of their CPUs
I feel like the current power limits are in some sense “artificial” due to AMDs seemingly deliberate design decision to have a high thermal resistance from the thick IHS and interface. One way of limiting power draw is limiting the ability to cool the chip, and they seem to have done that.
Good, they ought to be
I am waiting for Asus TUF Plus X670E & Asrock PG Lightning X670E (and maybe other 2 sub 300 Asrocks) reviews from buildzoid. Especially on the power side of things.
There are not many other X670E options with PCIe 5 which costs ~300 dollars and lower. Which is already quite high. Many other reviewers by some reason decided to go from top to bottom on their reviews. Which is so useless when top item is 1300$.
The only AM5 boards I would consider now are Asrock B650E. Their VRM is already overkill, PCIe5 futureproof and only cost around 200.
Its MASSIVE OVERKILL for current am5 gen processors, but the next generation may be even more power hungry and more cores, like with ryzen 1000 - 3000 series
Take a look at the world without the future.
How far we have come from complaining over underpowered power delivery to complaining about over engineerd over priced VRMs... well maybe they gona come back to these plastic heat sinks from 2 gens ago. At least this time no one would need to care
Actually modern VRMs require somewhat less engineering since the power stages integrated a lot of the functionality that would be upto the motherboard engineers to design in the past.