AMDGPU Linux Driver Seeing A Lot Of Power Saving Optimization Work

[Guest]

Be careful doing this because it overrides the automatic fan control. For example, A couple months ago I had set the fan to 5%, did some light browsing, and then fired up game. Performance was terrible and stuttering nonstop. Why? The card had got up to 82 degrees and was thermal throttling.

So it's fine to do if you have a start_gaming (sets fan to say 35%) and stop_gaming script (sets fan to say 5%), but not as a set it and forget it.

cl333r did you apply the patches to 5.10-rc4? I can confirm it works on Polaris (via the rocm-smi and visually inspecting the fans)

You could also just read the hwmon fan sensor, that works too.

[perpetually high]

You could also just read the hwmon fan sensor, that works too.

Yeah I had to check the fans to be certain :P it idles at a much higher temperature now, which should've been a dead giveaway but still. Used to idle at like 39-41, now it's in the 50s. Doesn't go over 66-70 degrees at about 24% fan speed (auto set by amdgpu) when gaming, though.

[darkbasic]

So would it work or not? I didn't quite understand.

No idea, sorry.

[Guest]

Yeah I had to check the fans to be certain :P it idles at a much higher temperature now, which should've been a dead giveaway but still. Used to idle at like 39-41, now it's in the 50s. Doesn't go over 66-70 degrees at about 24% fan speed (auto set by amdgpu) when gaming, though.

Sounds like some kind of fan control daemon would help. That way you don't need special scripts, and it's more temperature based so should work in all scenarios.

By the way, undervolting can help reduce temperatures, and so does using a different thermal paste like liquid metal. IMO, best solution is a liquid cooler for the GPU. The current heatsink + fan approach is ridiculously bad, when compared to CPU cooling.

[perpetually high]

Sounds like some kind of fan control daemon would help. That way you don't need special scripts, and it's more temperature based so should work in all scenarios.

By the way, undervolting can help reduce temperatures, and so does using a different thermal paste like liquid metal. IMO, best solution is a liquid cooler for the GPU. The current heatsink + fan approach is ridiculously bad, when compared to CPU cooling.

Yeah I was mentioning the fan information just for context on the temperatures. It's all handled automatically by the driver without me messing with it thankfully. Barely audible also which is great.

Undervolting has been extremely helpful on my RX 480 (MSI Gaming X 8GB). I'm still learning more about this card every day.

I just discovered the card came unstable from MSI with too low of voltage for its default core clock (1303 MHz @ 1156 mV). As a result this whole time I've been using it underclocked and undervolted at 1236 MHz @ 1075 mV which solved all my crashing issues. I now realized that bumping the voltage from 1156 to 1165 has solved all my issues at 1303 MHz. I didn't realize such little voltage was needed to make it stable. The max allowed is 1175 mV. Just frustrating that MSI didn't add that slight difference which would've made the card perfect from the factory.

On the bright side- I'm now running it overclocked at 1310 MHz @ 1165 mV and temps are great. Feels like a new upgrade with some of the improvements in FPS from 1236 MHz. I'm running the memory at 2050 MHz @ 1075 mV (stock is 2000 @ 975 mV). After lots of stress testing and games, no crashes at all. Power cap left at stock 180W as well. (PSU is also 700W so no concerns there, and the RX 480 uses a single 8-pin connector). Anyways, time to game finally