18113 extremely low PPD (25% usual) and low GPU utilization

[psaam0001]

I ended up putting my AMD iGPU back online, as my Alienware laptop which was running CPU only work is D/O/A (cause of death: GPU cooling fan failure).

So.... It's time to try to save for upgrade parts, and get them when the "supply chain issues" (read: the issue backlogged ships needing to be offloaded at a US port) is resolved.

Paul

[bruce]

Especially since the latest versions of the integrated GPUs from Intel have dropped double precision support in the hardware. I am waiting to see what the discrete cards are like, they may be more suitable.

The FAHCore (actually OpenMM) is written to REQUIRE hardware DP. Running emulated DP cuts Everybody's speed, not just those who don't have the DP hardware.

@John: is there any hope for some server logic that assigns a very limited set of projects that run a single precision FAHCore? ... or is that too much work for the net benefit to global FAH? After all, the iGPU is a low performance device. I can envision the iGPU to be inadequately cooled if you run too much DP

[muziqaz]

Especially since the latest versions of the integrated GPUs from Intel have dropped double precision support in the hardware. I am waiting to see what the discrete cards are like, they may be more suitable.

The FAHCore (actually OpenMM) is written to REQUIRE hardware DP. Running emulated DP cuts Everybody's speed, not just those who don't have the DP hardware.

@John: is there any hope for some server logic that assigns a very limited set of projects that run a single precision FAHCore? ... or is that too much work for the net benefit to global FAH? After all, the iGPU is a low performance device. I can envision the iGPU to be inadequately cooled if you run too much DP

Bruce, I don't believe Intel GPUs are getting any work at all at the moment. Just some random testing projects here or there.