I was actually able to get a single run of the test to give me a total extra +800 points, up to 15.3k, the highest I’ve seen! Granted, that’s not the 10 minute burn in I normally do. It seemed to be boosing to 60W just fine, and it felt like it could hold 55W if it just wanted to. This is all tested on a 240W charger. I’m going to call that a win either way though, I was honestly just hoping it wasn’t worse.
As for the dGPU, I think my GPU is overheating in demanding gaming sessions longer than 25 minutes. My “gpu temp” is in the 70’s, but hotspot is often around 95c and I get power target throttling after a whiel where it’ll jump between 40/60/100W power targets. This is naturally bad for my fps. Any particular reason you went with thermal paste instead of more PTM material (I’ve got a lot left from my LTT order). Also, did it look like they used regular thermal paste rather than the PTM sheet on the gpu when you replaced it?
Yeah framework just uses regular thermal paste of the 7700s. I didn’t feel the need to use the extra PTM7950 I had left over for the dGPU because the thermal paste work well enough as is. I also just did want to use it because of how expensive it was. I already had a ton of one of the best thermal pastes out there so I just used that and will probably use the PTM7950 as extra in case I need to take the heatsink off for the CPU again. Also for the first run of cinebench r23 the CPU is able to do that huge boost in power before dropping down to the 54w limit. That’s why it gives the higher score on the first run over the later ones.
I have spent so much time and money trying to resolve this performance issue in BF6, I’m going to feel very silly if all I had to do was repaste the gpu.
Yeah the other day I noticed that the dGPU was hitting high 90c and was also not using the full 100w but around 95-98w and so I repasted it and now hit a constant 100w on the dGPU. I get like a couple more fps in minecraft with shaders and a bit higher 1% lows in a drone sim but the fans can run much quieter.
Is there a better way to determine whether the heat sink is working or not?
For example, design a test program to load the GPU to 50W (half max).
Then document expected temp range for 50W with fans moving at full speed.
Then, another test that load the GPU to 70W (3/4 max), then document expected temp range with fans moving at full speed.
Then, another test with load the GPU to 100W (max), then document expected temp range with fans moving at full speed.
Then everyone could run the standard tests, and know when its time to replace the thermal paste etc.
With any cooling system, there will be a max heat dissipation rate, whereby it reaches a steady state without maxing out anything. The reason for the 50, 70, 100W tests is to establish what that rate is for a given cooling system. Thus, using that to determine if the cooling system is in or out of spec.
I assume the GPU cooler arrangement is simpler, with maybe it not needing copper shims etc., so the only variable is the thermal paste and fan failure.
Well sadly the heatsink is not well manufactured. When you ptm without the sandwich mod and compare it with the mod, there is quite a difference on average temps and core delta.
The way the original shim is bonded to the vaporchamber it is creating a heat resistance. The actual transfer area is smaller than the chip and the thermal glue used is not the best way in terms of transferabillity.
I had a batch1 dGPU I had to replace the thermal paste to get better performance (~50% increase). Later mine got replaced under warranty, due to the USB-C connector not function properly (usb-2, fine but no display data). The replaced dGPU had good performance out of the box and I have not yet found any need to redo/replace the thermal paste.
It was a simple procedure for the dGPU, much less hassle than the CPU it’s worth to try if you have paste or pads at hand.
Thanks for the info! Yeah, I’m going to repaste mine tonight. Unfortunately, the warranty for my dgpu expired a couple month’s ago, but on the other hand I’ve learned a lot by working through fixing this myself. I’m feeling pretty good about the chances of a repaste resolving my issues.
I’ve never seen a shim on the dGPU heatsink. The copper plate that’s connected to the heat pipe also goes over the VRAM so it’s not something that could be removed without destroying the whole heatsink.
Is Framework aware of this design weakness? I hope in the next board version they can address it, so we don’t have to Frankenstein our machines and risk voiding warranty by attempting to do so. I hope Framework reads these posts and passes on the details to Engineering/IP otherwise how would they know an issue even exists.
Yes they are, but its only a Problem for the 7X4X Generation, the Ryzen AI Heatsink was redesigned and simplified.
@zachary_stout@TechPriestNhyk Yeah sorry for the Confusion, the dGPU Heatsink is great, i was referring to the CPU Heatsink. i PTMed my dGPU over a Year ago and never had Heatproblems with it, but Hotspot Temps in the High 80s to low 90s when running overtdpd at 120W.
