Processors

[soupladel]

So i got my new kit yesterday of Asus Crosshair X670E Hero, &950x and 2x 16Gb Dominator Platinum DDR5 6000.

I got it all set up for the most part last night and in some initial testing, i have noticed that CCD2 wasn't reaching the same temperatures as CCD1 when running an all core test like cinebench r23.  Then i noticed, based on what HWInfo was reporting that in addition to the temperatures not being the same despite the same load, the max boost clock on CCD2 was also in the region of 250Mhz slower on every core.

I am using a Corsair H170i 420mm AIO as my cooler, but the version i have uses the existing AMD mounting mechanism of the hook type and i have to say it doesn't inspire me that the mount is particularly even, although it seems to be quite solid and secure.

So my question is, have i potentially lost on on the silicon lottery and just happened to have a CPU that, whilst performing relatively well, isnt living up to expectations on CCD2.  Or, is it likely the aforementioned mount is suspect and causing issues?

I have noticed that when at idle, CCD2 tends to sit around 5.4Ghz and CCD2 sits mostly around 3.6Ghz, occasionally up to 5.4Ghz.  If the mount was the problem, i would expect the hotter CCD to be the slower one when the reverse is true which leads me to believe that maybe its somehow not getting the right level of power?

[cpurpe91]

I have yet to get onto the AM5 platform. Provided I was in your situation, I would first ensure all connections are secure. After that, I would move on to see if the VRM was getting too hot in HWinfo. I have no clue what that would look like on the new platform. If the VRM is fine, I would move on to reseat the CPU and cooler. Now you don't have to worry about ripping out the CPU with the fancy new retention bracket on AM5. If this was still an issue after repasting and mounting the cooler, I would contact AMD directly to see what can be done. 

Currently I am in a somewhat similar situation however my 3700X is not overheating in any regard but 4 of the  cores only boost to 4,350~4,375Mhz, while the other 4 boost to 4,400Mhz. 

I know it's not the same, but I do understand being promised a certain level of power and not getting it with everything assembled and running properly.

Here is a link to email AMD.

Online Service Request | AMD

[soupladel]

Thanks for the response.

I continued doing some experimenting yesterday and one of the things i did was set the PBO Limits to "Motherboard" and that did bring a degree of parity back to the two CCDs in that they were boosting to much closer speeds, but CCD2 was still behind.

I have not had the chance to do anything else with the cooler yet because i had a fairly serious crash during a gaming test run and up until that point i was seeing consistent temps according to what i had seen previously and i was not throttling in any regard.

However i was reasonably sure that some of my stability issues were being caused by the fact i had a Windows 10 and 11 install on the same machine on two different drives so i have wiped both of them and started with a fresh install of Windows 10, i will resume my testing tonight after work.

I know the newer versions of my cooler come with a better mounting mechanism that uses the existing screw holes and back plate once the AMD brackets have been removed and i am hoping i can get a hold of that so i can get a more consistent mount.

[soupladel]

Ok, i have found the time to repaste and reseat the cooler and whilst i wouldnt say i was careless with the first fitting, i was super careful this time round and i am afraid i am getting the exact same results.

Under a full, all core load, CCD1 runs at approx. 5.3Ghz /95c and CCD2 runs at about 5.1Ghz/88c

Edit: I have now logged a call with AMD to see what they make of the situation 

[avendorz]

just linking this here:

https://community.amd.com/t5/drivers-software/wrong-core-affinity-with-the-new-dx11-driver-and-zen-3...

sounds like a similar issue (i just skimmed your post tbf)

[soupladel]

I had a look at this thread and it was more about threads being directed to the slower cores, where as my issues is a whole CCD of cores working consistently slower than the other CCD.

In my case each core on the second CCD is performing at the same speed, where as last gen the cores on each CCD would be almost independent of each other and would boost on a per core basis

[avendorz]

alright, mb then

[soupladel]

So, I have had an initial response and it feels like they didnt read what i had written and have just trotted out the "its by design" they have passing round tech circles as a form of official response to concerns about heat. So i have replied to say my concern is not that its getting hot, its that the second CCD isnt getting hot and therefore doesn't seem to be working as designed, lets see what they come back with.

I have also mentioned the random crashes that keeping happening regardless of what i do, no BSOD, just like someone has tapped the reset switch!!!

[Zolex]

I think it is by design though.. If you have 2 CCD's, they put a fast one and a slower one.
BUT The temperature must be the same so there is something wrong about power delivery
Is there a new bios version available for you motherboard?

[soupladel]

I am on the most recent stable BIOS, there are two beta version there but when i tried to flash either of them, the system crashes at the start.  i haven't tried using BIOS flashback to get one of the beta versions in, but i am not sure its the right thing to do at the moment, it could make things even worse.

[soupladel]

I have now updated to the latest BIOS released today (0705 for Asus Crosshair X670E) and weirdly, when running a single core load, it seems all cored are pegged at 5.5Ghz.

I am still seeing the same behaviour difference between the two CCDs but it looks like the new BIOS is limiting the max boost for some reason and it can't be power related because this motherboard has a pretty good power delivery, some might say an over engineered power delivery and i am running on a 1000 Watt PSU.

There is nothing i can see in BIOS that would limit the max boost in this way, any ideas?

[avendorz]

afaik agesa 1.2.0.6 or newer limits amds pbo which in return might result in less overall performance.

[soupladel]

there is now an option under PBO to set the temperature target of either 90, 85 or 80 i have it set at 90 and on the good CCD it hits 90, the slow CCD it only hits 80

[cpurpe91]

Do you think maybe you have a dud?

[soupladel]

Well thats my worry, i have a support case raised with AMD and their opening gambit was the party line of "hitting 95c is by design" but they missed the point that my concern was almost entirely about the fact the second CCD isnt hitting 95c and therefore isnt working be design 

edit: i forgot to mention the case with AMD has been escalated but havent heard anything for a couple days

[avendorz]

where can you even properly report issues like this?

the bug report tool doesnt allow for responses afaik.

[cpurpe91]

You can report issues via the support email. 

[soupladel]

the first reply to my post had a link to an online service request, they seem to have taken it on but i dont know how urgent they are treating it.

for clarity, here is the link again: https://www.amd.com/en/support/contact-email-form

[skyward737]

The H170i cooler has an adapter bracket for the AM4 that you use on the AM5, as well.  Do not use the hook  mount.  Change it out.  Also, since you have already made contact with the cooling block to the CPU, I would clean off and replace the thermal paste.

[soupladel]

the H170i unfortunately doesn't have the bracket you mention, yes if you look at the instructions on corsairs website it suggests there is, but that bracket did not come with my h170i which i have had in use for about a year at this point

I appreciate the suggestion none the less but if the problem was cooling related, it doesn't explain why the second ccd is not htting 95c given its designed to that and fit the performance into that temp.  My expectation would be that it would hit 95c but before even worse than it is if cooling was an issue

[skyward737]

That's strange that your 1 yr old H170i didn't come with an AM4 bracket.  AM4s have been around for quite a while.

[battyus]

My 7950x seems to behave similarly, but I'm not sure if it is really a problem or by design.

It would be interesting to see, how many of us experiences the same, and if there are any CPUs out there, which have the two CCDs reaching the same frequencies.

[Necromancin]

I am using the same cooler h170i, my ccd 2 is garbage at best, 5.4 all core no problem ccd1  5.2 is limit for ccd2.

[Thunderball]

Is there a problem with this cooler and AM5? I'm still waiting on my motherboard to arrive, but this cooler is the one I picked out and its sitting in box next to me.  I'm out of the loop (so to speak) on any problems with this and the bracket.  but it seems like maybe its not making contact correctly.

[battyus]

I'm using EK AIO 360 Elite RGB with 6 fans (push-pull), and my CCDs still behave the same as the OP's CCDs, so I would not think that the h170i cooler would be the problem.

I start to believe that this CCD behavior is by design, since I have not seen any posts that would confirm that somebody has a 7950x with both CCDs operating on the same frequency.

(If there are posts like that, I would like to read them, so please provide link)

[soupladel]

When you say your CCDs behave the same, can you confirm your temps on both CCDs under full load such as a all core load of Cinebench r23 or the likes?

[soupladel]

There has been a bit of confusion and misinformation around AM4 coolers on AM5, but the rule of thumb is this - if it doesn't need a custom backplate, it should be absolutely fine.

Now, as far as the H170i goes, it is an excellent cooler and performs really well but it seems it has two different versions of mounting to AM4.  There is the version i have which uses the existing AM4 mounts where the cooler "clips" down onto them, or more specifically has a clip either side which attaches to the existing mounts and is tightened via two screws.

Looks like this:

Then there is what i presume is the newer version where you unscrew those two exisiting mounting brackets and replace them with 4 stand-offs which allows the pump blocked to be screwed into place by a screws on each corner, looks like this:

Either of these two mechanisms will work just fine.  However, the general consensus is that this newer mechanism with the 4 stand-offs and screws is better because it provide a much more even level of pressure to the mount.

in all honesty, i suspect you will have this newer mechanism because i bought mine about a year ago, so unless you have been unlucky and got one thats been sat on a shelf for years, you should be ok

[soupladel]

What are your temps like under a full load for both CCDs?

[battyus]

I've found these two articles, so it really seems this common, and not just some CPUs:

https://www.techpowerup.com/review/amd-ryzen-9-7950x/26.html

https://www.techpowerup.com/forums/threads/amd-ryzen-9-7950x.299234/page-13#post-4845102

[battyus]

Using Cinebench r23, these are my results:

Multi Core test running for 10 minutes:
CCD1: 5275 MHz, 94.8C
CCD2: 5025 MHz, 87.6C
CPU Die (Average): 94.6C

Single Core test running for 10 minutes:
Active Core Clock: 5500 MHz, 50.1C
CPU Die (Average): 63.0C

Attaching some screenshots as well:

ryzen-7950x-cinebench23-multi-core-clocks-temps-average-speedryzen-7950x-cinebench23-multi-core-clocks-temps-fastest-speedryzen-7950x-cinebench23-single-core-clock-speedryzen-7950x-cinebench23-single-core-tempsryzen-7950x-cinebench23-results

I showed you mine, now show me yours

[soupladel]

Ok, here are my screenshots

My numbers are going to be a little different because i am on bios 0705 for the Crosshair X670e Hero which seems to be capping the max boost speed.  Also, because of this new bios, there is a feature to set the max temp target under PBO which is set at 90c and lastly, because my memory stability issues seems to have been solved, i have been able to set a -30 against all cores in curve optimiser but i am yet to do any proper stability testing such as corecycler but things mostly seem stable.

But even thought the numbers are different, you will se the same behaviour in that CCD2 is lagging behind and never hits the temp target of 90c

at idle:

During Multicore:

And Single Core:

[battyus]

Thank you for the results.
So, at this point to me it does seem like that this behavior is not because these processors are defective, but rather it is by design. What do you think?

[soupladel]

actually, i have just watched a der bauer video on youtube where is talking about using an offset cooler to see if it improves thermals and funnily enough, his CPU is doing the exact same as ours

[soupladel]

I would really like to see further evidence of this happening to others, perhaps even some sort of clarification from AMD itself would also help.  But, i would also like to understand why the latest BIOS from Asus has my chip limited in its boost speeds at 5.5, this isn't a beta version i am on and the thought has crossed my mind that perhaps the limitation has come into force because of a fault detected by the BIOS.

On a previous version of the BIOS i was boosting to 5.75 on all cores but was unstable because of the memory compatibility issues, so i would like to see my boost speeds returned to the advertised  before i will be convinced there isn't a fault on these chips

[ajlueke]

Too be fair, the maximum boost of 5.7 GHz is the maximum single core boost.  AMD doesn't guarantee that boost will be maintained in an all core load.

[soupladel]

That’s right and I would never expect it to be maintained in an all core load but the original post was concerned with the fact the second CCD wasn’t hitting the 95c that I thought both CCDs should be doing, also the second CCD was lagging behind the first CCD by a significant 200Mhz, but crucially all cores in each of the CCDs were doing the same thing

[ajlueke]

As others have stated, the CPU die temperature is what is important.  And as in the Ryzen 3000 and 5000 series, when there are multiple CCDs, one will be more performant than the other.  CCD1 supports higher clocks at the same voltage, and thus is being boosted preferentially over CCD2.  CCD2 isn't hitting 95C because the die is already maxed out, so nothing can be boosted any further.  

If you instead had CCD2 at 96C and CCD1at 89 with the same temp of 95 on the die your performance would likely be lower.

[soupladel]

My experience of 3000 and 5000 doesn’t agree with that, I didn’t have a CCD which performed better than the other, there were 2 core on each CCD which were the preferred core and this boosted the highest but all other cores outwith those four were general the same.

however, this 7950x has the cores on each CCD performing exactly the same as one another which is unexpected accordingly to my experience previously with a 3900x and a 5950x

[ajlueke]

In terms of the "Max boost", I only ever had two cores on CCD1 that could reach the max boost clock in lightly threaded scenarios.  All other cores, including those on CCD2 could not.

It is true, that in heavily threaded workloads, I was typically PPT/TDC bound and all the cores generally showed the same clockspeed.  What seems to be different in the 7000 series is the fact that the processor is getting very close to the "max boost" clock even on all core loads.  Which was certainly not the case with previous generations.  This is likely why you are seeing the core differences manifest even in all core workloads.

[battyus]

In this review there are some 7950x MHz numbers, which seem to line up with our findings (Look at 20:28) :
https://www.youtube.com/watch?v=_BeYPcJOQPc

His CCD2 seems to be behind CCD1 as well, when they are boosting. Interestingly, when they go below ~5GHz, they do show same MHZ, like 4000MHz all 16 cores.