Request to bump species of AMD Radeon 780M

[arisu]

Oh okay. When I downloaded the new gpus.json they weren't 9 yet.

[muziqaz]

They should have been. Ah sorry, misread the numbers. I'll leave iGPUs at 7

[arisu]

Sounds good, especially since they may be getting thermally limited if folding on the CPU at the same time. I get more PPD from folding on both than from folding on just GPU or just CPU (not nearly as much as the sum of the two, but still pretty good).

I'm glad I made this request! Now my little laptop is hot to the touch again, just the way I like it.

[muziqaz]

Let's see how you'll talk, when your laptop bursts in flames

[foxpy]

I think with modern "thermally optimized" hardware the only concern around laptops is the battery. It might not like high temperatures, otherwise I don't really see any problems with laptop folding.

[muziqaz]

I'll stick to my cautious approach ;P

[Joe_H]

On old, but still good idea with many laptops is to elevate it a bit off whatever surface it is on. That allows some airflow over the bottom case, helps especially with metal cases.

[arisu]

The battery might not like it but it will survive. It's replaceable. I'm more concerned about the electrolytic capacitors. Those are the components that are harder to replace in modern laptops and they are almost always the first components to fail. And heat accelerates their failure.

But this is a small laptop. I don't need it to have multi-decade reliability.

[arisu]

As an update, the 780M is doing great so far. Even with the bigger WUs, it never gets even close to the timeout, even when the CPU is folding at the same time. So I think it's a success.

[arisu]

One thing that I have noticed is that the VRAM is a limiting factor. Most projects complete well within the time limit, but some of them use almost all of the video memory. Right now, 12922 is using 504 out of 512 MB. It folds but it causes the graphical desktop environment to crash. This isn't a problem for me because I am comfortable using the terminal (I'm writing this in a text-based terminal browser right now).

The 780M uses system memory as VRAM. The BIOS can configure this limit between 64 MB and 512 MB but it cannot go higher. So far only the 12922-12925 series and that series with the IUP in 1.2M water molecules (I don't remember the project numbers) have caused memory problems. But the GPU itself is fast enough to make the timeout.

[foxpy]

wait, don't you have GTT to extend the video memory? you can use radeontop tool to read the amount of available GTT. If there is none, you could specify more via amdgpu.gttsize module parameter: https://www.kernel.org/doc/html/v6.14/g ... eters.html

[arisu]

wait, don't you have GTT to extend the video memory? you can use radeontop tool to read the amount of available GTT. If there is none, you could specify more via amdgpu.gttsize module parameter: https://www.kernel.org/doc/html/v6.14/g ... eters.html

I do, but for some reason the framebuffer does not want to use it. Once I exceed 512 MB, framebuffer allocations will fail (amdgpu's concept of a framebuffer is slightly different than Linux's concept i.e. DRM still uses the framebuffer by its definition). They fail with error -12 (ENOMEM), which causes the screen to go black until I switch back to a virtual terminal. If that happens too many times, the amdgpu driver can reset itself, which causes other problems.

Radeontop is what I've been using. The GTT is available and slightly in use, but it still doesn't stop framebuffer allocations from failing. Which is frustrating because the GTT is just as fast as the "dedicated" VRAM because they're both just system memory on an APU...

[foxpy]

This concept is really stupid and I don't understand why they made it so complicated

[muziqaz]

This concept is really stupid and I don't understand why they made it so complicated

AMD is aware of it now. Not sure if they will do anything about it.

[muziqaz]

What distro is this?
Kernel?
ROCm version (if used)
Thanks