A tale of two 5950x

[john terragon]

Hi.
So, I have to 5950x, I'll refer to them as the old one and the new one.
They are installed on computers with roughly the same config. Same memory,
same AIO cooler, almost same MBs (the old is on a MSI x570 meg ace, the new
one is on an MSI x570s meg ace max), same config in the bios...

The old one behaves under load as, I think, we're used to see a 5950x do:
with a 32 threads load under linux it boosts at about 4.65 ghz on all cores and
temps goes up to 55C over ambient.

The new one behaves under load in, I think, a strange way for a 5950x:
with 32 threads load under linux it boosts at about 4.3 on all cores BUT temps
never go above 30C over ambient.

(don't compare the 32 threads load under linux with a multi core cb20 load
under windows. Both of them will probably boost higher with CB20
under windows but the comparison of their behavior would be the same.)

Now, I thought that a ryzen 5000 would, roughly speaking, boost as high as it can
as long as it does not pass a certain thermal threshold.
So, if this was the normal situation in which the old is a decent chip whereas
the new one not so much, I would see these low boost on the new one but coupled
with high temps. Instead I get this low boost AND low temps and it seems
like the new one is holding back or is being held back (maybe by the agesa code?
but they are both under 1.2.0.3c so...I know it's not the slightly different MB,
because the new cpu was on an Asus x470 MB for a little while and it behaved
exactly the same).

So, two questions:

1) Has anyone seen/known of a behavior like this?
2) What should I do to try to give the new one a "nudge" into boosting higher?
Maybe try to put manual PPT/TDC/EDC (and what values exactly? I'm not much
of an overclocker). Or maybe something else, I don't know.

Thanks

 

[john terragon]

I thought total package power (142W for a 5950X) was the limiting factor if not thermally constrained? You'd probably want to try the curve optimiser to get it boosting a bit higher.

Exactly. But it does not seem to be thermally constrained. It's as if the cpu was operating with one of those profiles that you find in the "amd overcloking" part in the bios like eco or 80w (I don't remember the names). Except that is not the bios instructing it to do it. It seems to be doing it by itself. So maybe just telling it directly what the PPT/TDC/EDC should be might help.
I don't know, it's strange.

 

[john terragon]

Has one of them been repasted?

Well, the thermal paste is a few weeks older on the old cpu than the new cpu.

But even if there actually was a relevant difference of performance between the two "pastings" it would show in the other way around.
That is, I would see low boosts on the cpu with higher temperatures and high boosts on the one with lower temperatures.

What happen here is the opposite: the one with the higher temperature gets 350Mhz+ higher boosts than the one with lower temperatures.
IT really does not make sense unless is the cpu itself that uses a lower PPT or maybe the agesa that recognizes that cpu and enforces a lower PPT on it.

Also consider this: both cpus on single core loads boosts easily above 5 ghz. That would be coherent with the different PPT hypothesis because
I think single core load does not even get close to the nominal 142w PPT that both cpu should hit "voluntarily". So on single core load even the new cpu does not pass its hypothetically lower PPT and it boosts as much as the old one.

 

[john terragon]

Exactly the same linux installation (not only distribution), exactly the same governor (schedutil, but its the same with the others).

And about swapping cpu:
The new cpu (the slow but cool one) has been in two very different MBs (the msi x570s and the asus x470) and it behaved in the same exactly strange way.
Now that would leave the possibility that the msi x570 (the one with the old cpu) was the crazy mb that for some reason was pushing the old cpu.
The problem is that it's the old cpu that is behaving like a 5950x with pbo enabled should behave. So I see normality in the pair x570+old 5950x and I see anormality wherever the new 5950x is installed.

To reiterate: 30C max over ambient with pbo enabled under 32 threads load, all core boost to 4.3 ghz.
To me it seems like its not even trying.

Is there a way to get the power consumption of the cpu? Maybe ryzen master under windows? Or hwinfo? Sorry but I use windows only to play videogames so I'm not really up to date with the various utilities.

 

[john terragon]

I just tried to put the values for PPT TDC EDC manually at 142 95 and 140 (that I think are the base value) et voila! the new cpu suddenly boosts at 4.55 ghz on all cores. And the temps are still good at 35 over ambient.
WTF is going on here!

 

[john terragon]

So, does it make any sense to set the PPT TDC EDC higher than default? And what value should I put in?
I want to see if it pushes the boost freq even more by just setting those values and without using the curve optimizer.

 

[john terragon]

I tried 215/150/170 and it does worse than 142/95/140 (which I had to put in myself because who knows what "auto" was doing).
Now that would not be a problem if there was any trace of temperatures getting higher. But temps stay around 25 over ambient (45C absolute) with 32 threads loads.
So, it seems like there is some other limiting factor that the cpu and/or the mb take into account, because temps are ridiculously low.
At this point I want to try manual overclocking because if temps don't get to 50 or 60 over ambient (which in absolute terms would be 70 or 80 C) then there is something wrong.
The last time I overclocked a cpu it was a 4790k...so if anyone who has a stable manual overclock could just tell me what values to modify I'd be great.

 

[john terragon]

Is the 'new one' one of the B2 stepping chips? As this is the only time I have witnessed similar behaviour.

Where do I look to find that out? (I hope not on the heat spreader...)

 

[john terragon]

OK.
I'll look as soon as I'm back on windows on that machine. But I think it's probably somewhere in the bios too because I've just read that it seems that the mb has to have specific support for the B2 stepping.