Glad that I came back to the forums, 7940 getting consistent 13.7-8k CB23 10 min runs. What is the best way back to full performance, swapping with PTM7950? Any options to rectify via support?
Prior runs were ~13.7k
Prior runs were ~13.7k
Batch 1. So, I decided to quickly just check my core temp spread using s-tui. Setting power to “performance”…
7/8 cores spread: 32.8°C - 46.8°C (this is already a weirdly large spread)
1× core (Acpitz,3) is sitting pretty at: 99.8°C (
)
Total package power is at ~29W.
On idle the temp spread ranges from 39.8°C → 46.8°C with core 3 randomly spiking to >60°C for a few seconds as it gets work to do.
These values are so bad I’m even wondering if I read them right.
Ugh, I signed up for the thermal pad replacement. I really hope it ships soon, else I might have to RMA, but this is my only system I have right now for work.
framework is moving to ptm. there are plans to provide conversion supplies ‘early’ 2025. there’s a form to request them, found the link in this post
1 Like
What OS?
If linux and using lm-sensors, per core is not reported. Just an FYI
For per-core reporting on Linux,
Use “amdgpu_top”. e.g. from: GitHub - Umio-Yasuno/amdgpu_top: Tool to display AMDGPU usage
Although it says gpu in the title, it shows the 8 CPU cores.
3 Likes
Yup, thats what I use.
the way you’ve written the last two messages suggests maybe you weren’t aware, but in amdgpu_top if you select the 780m then scroll down a ways, it includes 8 cpu core temperatures
I simply told the other guy that his “large temp variance” he was seeing was likely from using lm-sensors.
Ive been using AMDGPU_TOP since about November when I did my PTM swap.
my mistake
1 Like
I confirmed that the acpitz sensors report through lm-sensors.
I now re-looked at it using amdgpu-top whilst running the same load test.
My CPU core temp spread ranges from 81°C to 100°C, so about 19°C under load. This seems more sensible. Still bad, but within the realm of what other people here are talking about.
It’s still only one core that is at 100 and the rest is in the range of 81-92°C, so there is still clearly some kind of unbalanced cooling.
The thing that worries me more, is that under all-core load, I only get 3.3Ghz all-core speed, and the wattage peaks at 29W with performance mode being set.
I remember getting mid-40 watt and ~3.8Ghz all core when it was new, so I feel this is a regression.
Sorry for a bit off topic, but how do you get 110fps on Helldivers 2? Most I can ever get is about 70fps. Also, how are you seeing cpu at 35-45w? I never see mine go more than 31w.
I changed the liquid metal for some ThermalGrizzly PhaseSheet. See Uneven CPU thermals! for the details!
Yeah, I’m currently in the “escalation” stage with support. But dang, 70fps to 110fps is a massive boost to just have the thermals working properly, and I have the 7940 so it hurts even more.
The TM Phasesheet is very cheap on Amazon. If you don’t want to wait support and you’ve disassembled PCs/laptops before the procedure is pretty easy, good luck!
Wasn’t there a shim required to be added though? I any case, I might just do that depending on how support works out, or doesn’t. My biggest concern is cleaning off the liquid metal without ruining it.
framwork will prsumably publish a guid onc thy ar rady to start snding out ptm7950 sampls to affctd customrs
1 Like
The shim is required for the full potential of above 55w tdp. But with just replacing the lm to ptm you get to around 45w sustained tdp.
But thats not the main topic and it doesnt fully affect gaming performance with the dgpu.
Are you using AFMF?
On linux, without the framegen of AFMF in windows, I get about 70 as well.
I think his is much higher thanks to framegen.
Yes. As stated here, I was using RSR 1920x1200 + AFMF2 @ Quality/High.
eeeeeeeeeee! grate!