Using INTEL GPUs

[iceman1992]

expect Core_21 again and slightly lower PPD)

Is that just a difference of points assignment, or genuinely the science stuff is done slower on Core_21?

[JimboPalmer]

expect Core_21 again and slightly lower PPD)

Is that just a difference of points assignment, or genuinely the science stuff is done slower on Core_21?

core 21 Available to Windows and Linux for AMD and NVIDIA GPUs using OpenCL. It uses OpenMM 6.2 and fixes the Core 18 AMD/NVIDIA performance issues.

core 22 Available to Windows and Linux for AMD and NVIDIA GPUs using OpenCL. It uses OpenMM 7.4.1 - https://en.wikipedia.org/wiki/List_of_F ... home_cores

So potentially several software revisions separate them.

[HugoNotte]

I would guess that due to the high work load put on the system by F@H any IGPU would cause extreme heat. For the fun of it once tried to run a BOINC project WU on my Macbook Air's IGPU. The temps shot up and the fan almost came out of the case, so stopped it immediately.
I would think that a couple of CPU cores are more valuable than a single IGPU thread, after all CPU WUs are being assigned for a purpose. I might b wrong.

[BobWilliams757]

I would guess that due to the high work load put on the system by F@H any IGPU would cause extreme heat. For the fun of it once tried to run a BOINC project WU on my Macbook Air's IGPU. The temps shot up and the fan almost came out of the case, so stopped it immediately.
I would think that a couple of CPU cores are more valuable than a single IGPU thread, after all CPU WUs are being assigned for a purpose. I might b wrong.

Due to the fact that they don't have much ability, the AMD integrated graphics aren't that bad for heat. On my 2400G system using the onboard graphics without the CPU only draws about 12 watts. Throughput is only 15-17000 PPD (without QRB) but overall it's not much less efficient in terms of heat or PPD than a lot stand alone GPUs of the same age. With my current fan profiles and such, it sends the temp up about 15C, which doesn't even increase my fan speed. I have to check the client or look at a monitor app to see if I'm running a WU.

[SandyG]

Was just migrating a few machine over from Seti@home to Folding@home and figured I had set up something incorrectly as I could not get any utilization of the Intel GPU on a few small Linux machines that I have been setting up to grind data. These don't have any expansion for much as they are the small Gigabyte Brix computers.

Problem solved and I can stop fiddling with trying to fix (not really broken it seems) the OpenCL install!

Sandy

[bruce]

I know nothing is going to change during the Covaid crisis. Everybody who might work on a development project for Intel iGPUs is toally stressed out trying to upsize FAH to deal with the sudden upsurge in the Donor count.

When this crisis is over, I can think of several possibilities:
1) A successor to FAHCore_a7 will allow the CPU threaded client to off-load a portion of the calculations to the GPU and they will be able to run parallel on a single WU.
2) FAHCore_22 will increasingly use the iGPU to off-load selected support functions from the WU being processed by the primary GPU(S).
3) A stand-alone core will be developed that can use the iGPU independently of other FAH activities.
4) Nothing will happen.

Item 1 has been announced by GROMACS so it's a (not so) simple upgrade to the GROMACS code that's currently in use.
Item 2 is already a reality. It's possible that it can be further optimized.
I doubt item 3 will happen. Mainly because it dilutes the overall goals of FAH. Why should it require a separate WU when either 1 or 2 acually focuses the extra processing on speeding up the active WU. That's a better option both from the Points perspective and the Science perspective.