Question Why does TDP and PPT differ, on consumer CPUs? And what role does Core Performance Boost and Turbo Clocks have on TDP and wattage?

[VirtualLarry]

Serious question. I've got a 65W-rated TDP Ryzen R5 1600, in a rig, on a 65W-rated AMD stock heatsink. It's blue-screening, crashing, and the CPU temps just keep going up and up.

I updated HWMonitor, and it's showing a "Package Power" for the entire chip, at 82W or so. No wonder it's constantly overheating and crashing. 82W TDP CPU > 65W TDP Heatsink.

The worst part is, this is AFTER limiting the number of PrimeGrid threads, down from 12 to 9. That's right, I'm not even running the CPU at a full thread load.

Edit: Yes, I know that the obvious answer, is to "get a better heatsink", and that the "stock heatsink" for the 1600 was the 95W TDP model. Which, at the time, was stated that AMD wanted to give users the ability to OC on the stock heatsink. Now I know that was a lie, it's because AMD CPUs (at least, the 1600), are NOT able to stay within their stated rated specs.

Edit: A slight update, very important, actually. My original premise for this thread, was that I *thought* I was using a 65W TDP-rated AMD stock Wraith Stealth cooler with my Ryzen R5 1600 CPU, and it was crashing, at "stock BIOS" settings, which includes "Core Performance Boost" on "Auto", which defaults to enabled, to allow "Turbo Clocks" (the 1600 has an ACT of 3.4Ghz). I was initially placing the blame on AMD for the fact that HWMonitor reported the "Package Power" as something like 82W, which I thought was overcoming the 65W-rated heatsink. As it turned out, I actually was using a 95W Wraith Stealth (copper-cored) in the first place. Yet, it was still crashing due to overheating of the CPU. Part of this was due to the heat load of dual GPUs mining, and part of it was due to using a case that had NO vents on top, no fan mounts, no rad mounts, nothing but a solid steel top, and only a single 120mm exhaust out the rear, combined with the fact that my PCs are in desk cubbies. They are open to the front, and have dual 120mm intakes and vented fronts, but that still wasn't enough to prevent the CPUs from slowly creeping up in temp, passing 95C, and crashing/restarting.

Thus far, I have split the two GPUs up, one per PC (same case, same type cubby, same EVGA 650W G1+ 80Plus Gold PSUs), and disabled CPB on both of them (one has a 3600 CPU, one has a 1600 CPU), and then also in Wattman, set the Power Limit for the RX 5600XT (which was a refurb, both of them) to -20%. Thus far, overnight, they seem to have stabilized at under 90C on the CPU, and haven't crashed.

 

[VirtualLarry]

That all sounds great, until I look in BIOS, and find no static voltage options. :|

Maybe this thread should be titled, "Gigabyte sucks".

 

[ksosx86]

I'll look for "ECO mode", although I would think setting the cTDP to 45 would be the same.

I'm done going through the whole Gigabyte BIOS and yeah no joy. Sorry. Hmm.

• Stock settings for the B450 AORUS PRO WIFI - then enabled everything below except for Core Performance Boost (disable)
• Try enabling ERP & CEC 2019 Ready
• Leave everything else at stock - CPU cooler on CPU_PWR header (that way a fan profile won't matter)
• EDIT: Yeah I recall having one way back before swapping to another brand and being bummed the best I could do was create a negative offset for my cpu in GIGABYTE BIOS and it would only go so far like there was a cap on it.

 

[ksosx86]

That all sounds great, until I look in BIOS, and find no static voltage options. :|

Maybe this thread should be titled, "Gigabyte sucks".

Nothing showing this? Should be CPU VCORE you can set to something?

 

[VirtualLarry]

Nothing showing this? Should be CPU VCORE you can set to something?

Nope. Only "Dynamic VID (vcore)". Which is, sadly, also known as "offset mode".

I did uncover, in the advanced options, AMD PState Overclocking, so maybe I could use that instead, but I'm not well-versed in PStates, I would rather just punch in a frequency and voltage (fixed).

 

[DrMrLordX]

That all sounds great, until I look in BIOS, and find no static voltage options. :|

Maybe this thread should be titled, "Gigabyte sucks".

Wow, that's just bad. Can you at least set static clocks?

 

[yeshua]

Well, the thing is by their definition they do run within their rated specs. All of AMD's 65W desktop CPUs are designed to hold 88W under long term load. All of their 95W CPUs are designed for 132W under long term load, and their 105W CPUs do 114W. There's a GN article that explains AMD's TDP definition, but for the most part, you're best off ignoring it because it's extremely convoluted and instead looking for the PPT (power pulled form the socket) definition.

That's the definition of lying. 65W stated, 88W actual power dissipation. I've been calling AMD out for years for that but no one cares.

 

[coercitiv]

That's the definition of lying. 65W stated, 88W actual power dissipation. I've been calling AMD out for years for that but no one cares.

Put a 65W cooler on the AMD 65W CPU and - at stock - the average power dissipation will converge towards 65W for long-term loads.

I'm getting really tired of this charade, so let's get one thing straight: the OP is not using the CPU at stock. This needs to get engraved in people's heads before even more judgements come out flying without proper context awareness. Enabling XMP on any board - be it AMD or Intel based - has a number of intended and unintended consequences on CPU voltages and possibly even power & thermal limits. XMP is not just enabling custom memory clocks and timings, it is a shortcut to overclocking for beginners.

Expect the unexpected when using it, on both Intel and AMD platforms.

 

[StefanR5R]

This seems to be the specification which I asked for:
Source: https://www.evga.com/articles/01185/evga-g-plus-power-supplies/
View attachment 32156
So there is only a single +12V rail. It is IMO unlikely that the two GPUs manage to trip the PSU's over-current protection by simultaneous spikes. But I am not an expert with GPUs and PSUs.

Here are high-frequency measurements of the power draw of an RX 5600 XT, BIOS at 150 W TDP, gaming load:
review at igorslab.de
(Caution, this article looks as if it was machine translated.)
Based on these measurements, the author recommends to size the PSU such that 200...250 W are set aside for this GPU.
I don't know if short-period power draw looks very different with mining workloads.
(Power draw probably looks much different if the card is tuned for best efficiency under mining workload.)

 

[ksosx86]

Yeah, I've been repeatedly trying to drive home that point - you'll only start to figure out what's causing the issues with overheating through component isolation.
Isolate all of your hardware piece by piece - this includes having to make your setup as vanilla as possible first and starting from there. The point being if your setup cannot handle the physical space and environment of your setup at vanilla don't expect -redacted- any further than that without further hardware modification(s).
@VirtualLarry Many people have pointed this out already but you haven't responded in the affirmative yet as to if you've actually tried that, what's going on over there, why the insistence on avoiding starting with the basics, and going from square one?
You should keep a troubleshooting log everyone is like "where's he at in the process" just keep editing your first post.

Profanity is not allowed in the tech forums.

Daveybrat
AT Moderator

 

[StefanR5R]

What about my 2970wx, by 3 2990wx's, and my 3 3900x's ?

With Threadripper 2970WX and 2990WX, AMD's default PPT limit is at the same value as the default TDP, which is 250 W.

With Ryzen 3900X, AMD's default PPT limit is 142 W, bigger than their default TDP of 105 W.

 

[Velgen]

The offset is most likely variable based on SKU. From what I understand some of the higher SKUs had a 20C offset, while my 1600X had a 10C offset based on HWInfo readings.

Ahh was not aware of that I only have the higher SKUs of first generation Ryzen so all of them have the 20C offset. Have to say was not a fan of that offset it made the default fan curve more aggressive than I would care for.

As for the BIOS it looks like the Gigabyte BIOS is in Easy Mode that some people mentioned previously. Looking at this post from Gigabyte should be F2 to switch between them (may need to be on front page) and it looks like at least from that post if you switched to advanced you should be able to edit Vcore, but not 100% sure since I do not have a Gigabyte board. Hopefully all the screenshots there are of the AMD bios since the first image in that post is with a Ryzen 2600, but it is possible they were showing both Intel and AMD BIOSes. Oh and the post was of the new design as of October 2019

 

[Gikaseixas]

Dear Larry,

Please buy a decent cooler when planning to use a CPU outside of the manufacturer approved parameters

Regards,

Anandtech Community

 

[Ken g6]

Nope. Only "Dynamic VID (vcore)". Which is, sadly, also known as "offset mode".

I did uncover, in the advanced options, AMD PState Overclocking, so maybe I could use that instead, but I'm not well-versed in PStates, I would rather just punch in a frequency and voltage (fixed).

I'm not using Windows, but if you are, you could try ThrottleStop to at least set frequency manually. I do something similar on Linux, but it's Intel-specific, and I don't have a Ryzen.

 

[ksosx86]

I'm not using Windows, but if you are, you could try ThrottleStop to at least set frequency manually. I do something similar on Linux, but it's Intel-specific, and I don't have a Ryzen.

If this has been brought up my bad... regarding Windows-specific: AMD Ryzen Master

 

[biodoc]

Anyways, I don't know if I have the PBO section in the BIOS, these Gigabyte BIOSes are WAY stripped-down as far as options go, as compared to my Asus and ASRock boards. I'll look for "ECO mode", although I would think setting the cTDP to 45 would be the same.

I have the PPT in the bios set at the processor TDP on all my 4 of my Zen 2 processors. PBO needs to be on or on auto for it to work properly. They all run 24/7 and run cool on air. This article is a good read for "undervolting" Zen processors and according to the author, reducing PPT is the preferred method.

If you can't find the section in the bios for controlling PPT and you are a windows users, then reset the bios to default settings and run ryzen master and explore your options. For ECO mode there should be a "button" to select that option. For manually setting the PPT, I found this image:

 

[UsandThem]

I've been calling AMD out for years for that but no one cares.

Yup.

I know I don't.

 

[Velgen]

I have the PPT in the bios set at the processor TDP on all my 4 of my Zen 2 processors. PBO needs to be on or on auto for it to work properly.

PBO or Precision Boost Overdrive is only for anything from Ryzen 3000/Zen 2 onwards I do not believe they added backwards compatibility for PBO to the older Zen CPUs (yes I know PBO existed on Threadripper before the consumer Zen CPUs). So unfortunately that specific way of doing it won't work for him most likely he will just have to do it the old fashioned way of adjusting vcore and such.

 

[blckgrffn]

Ever more helpful advice.

Or...

You know...

Avoid running multi GPU mining rigs in cubbies and then crushing the CPUs with distributed computing products at the same time. Also factor in a thunderstorm outage or two and UPS failures.

This whole thread is a study in how to:

A) Explore exactly why refurb hardware is likely refurb​

B) Heatsoak your components until they die​

And yet this thread title remains:

Why does AMD lie about CPU TDPs? (AMD TDP not covering "worst-case" AVX software loads accurately.)

If this was someone with 20 posts and a 2019-2020 account creation date I'd say it was paid flamebait.

But Larry, gosh darn it, you keep cherry picking advice from the thread. One step forward, two steps back. I think I am coming around on it just be an amusing pass time while waiting for RNDA2 launch but really you seem to be investing a lot of time into this and it pains me to watch it unfold

 

[VirtualLarry]

FWIW, neither rig has rebooted or crashed, after going into BIOS, setting "Core Performance Boost" to "Disabled" (After setting XMP), and then in Windows, setting Wattman on the RX 5600XT to -20% Power Limit. So far, so good.

It seems that the problem with AMD CPUs not keeping within TDP, is largely down to Core Performance Boost, being "Enabled" by default, after a BIOS "Load Defaults" operation.

It would be interesting, if:
1) All mobo mfgs do this, or only a few, and
2) If this TDP-busting behavior, is in fact, AMD-specified (BIOS writer's guidelines, or optimization manual)

 

[coercitiv]

It seems that the problem with AMD CPUs not keeping within TDP, is largely down to Core Performance Boost, being "Enabled" by default, after a BIOS "Load Defaults" operation.

So your solution for AMD CPUs is for them to have turbo boost disabled by default?

 

[StefanR5R]

@VirtualLarry,
it should be well-known by now that "PPT limit" is the power draw limit¹, not "TDP".

The default PPT limit of Ryzen 1600 is 88 W.
(It should say so on the box, but it does not.)

Ryzen 1600 is an 88 W processor (by default²).

It is also well known that it is *not* a good idea to use an overclocked processor at PrimeGrid. This'd be too likely to produce incorrect results. Ditto, don't overclock memory at PrimeGrid. Each and every PrimeGrid challenge primer posting says so, and it has been proven true again and again.

_____________
¹) Except when some throttling mechanism hits, notably thermal throttling.
²) It makes a lot of sense do deviate from this default: I.e., set the PPT limit lower. *Much* lower.

 

[moinmoin]

With all Zen gens PPT is 1.35 * TDP. At stock it will never go beyond PPT. There is no "lie" or any "TDP-busting" involved.

 

[StefanR5R]

PS,
there is no "TDP-busting behavior". Why? Because TDP is a practically meaningless number. It is a certain control point of very low practical significance. You can't "bust" that which does not have any meaning in the first place.

There is one meaningful parameter for Ryzen's power consumption: PPT limit. Try to bust that one.

The thread title should not say "Why does AMD lie about CPU TDPs?". It should say: "Why does AMD print a totally insignificant parameter on the box, but can't be bothered to prominently document the really important parameter, PPT limit?" And the answer to that is: Most likely because marketing, competition, and those at AMD who decide things like these believe their customers are idiots.

Edit, ninja'd by moinmoin.

 

[blckgrffn]

PS,
there is no "TDP-busting behavior". Why? Because TDP is a practically meaningless number. It is a certain control point of very low practical significance. You can't "bust" that which does not have any meaning in the first place.

There is one meaningful parameter for Ryzen's power consumption: PPT limit. Try to bust that one.

The thread title should not say "Why does AMD lie about CPU TDPs?". It should say: "Why does AMD print a totally insignificant parameter on the box, but can't be bothered to prominently document the really important parameter, PPT limit?" And the answer to that is: Most likely because marketing, competition, and those at AMD who decide things like these believe their customers are idiots.

Edit, ninja'd by moinmoin.

I think you have a fair point on the thread title IF maybe more than a 1/3rd of the posts in this thread dealt with that topic. You have a great, well thought out post here. Arguably, both CPU vendors pretend we are idiots in their own special ways. I agree with any of the opinions in your post and your specifics seem to be dead on as far as I know.

That said, the thread title should be "Why does my mining rig keep freezing up?" as the TDP/PPT values seem have a very tenuous relationship with the basis of this thread.

 

[Velgen]

Well the issue with TDP is that all brands AMD, Intel, and Nvidia use STOCK clock as TDP not turbo. Not a single one of the brands to my knowledge guarantee anything past the stock clocks they do not guarantee you will ever see those max turbo/boost clocks. There is a reason some heatsink manufacturers have stopped saying what TDP they rate their heatsinks for these days (the most notable of which is Noctua).

@blckgrffn Thanks I work in retail I've long learned it's best not to try and dissuade people from doing something that seems stupid, even if it is actually stupid, because generally there is a reason behind it. Try to give them the best advice you can to avoid the customer having problems and make it go as smoothly as possible.

In this case though I really feel like Larry should stop trying to rush getting these things running and troubleshoot the components one by one. Then work on undervolting/downclocking both the CPU and GPU since it won't affect mining performance much if any as another poster stated. That will reduce the heat output of it significantly and only once that is stable throwing them back in the cubbies. Maybe he's got limited space and doesn't want them cluttering up the limited space he has who knows.

Oh and a new case and heatsink for those things wouldn't hurt either