GTX 1080's don't like WU 18601

[WhitehawkEQ]

Like the title says, my GTX 1080 cards don't like WU 18601, I'm getting doodoo for points, like half of what I normally get. So my dally points has dropped some.

 

[WhitehawkEQ]

I need to upgrade my 1080's badly, ok everyone give me all your 3070's, 3080's and 3090's so I can be top dog around here

 

[torin3]

As much as your PPD has dropped, if you weren't having WU problems, I'd almost think you had a system or two offline.

As a team, we've gone form overtaking 3 teams down to 1.

Fingers crossed for you to get a better selection of WUs soon!

 

[HayesK]

My HFM p18601 data has linux cuda 1070 ~1.2M ppd, only slightly below my expected 1.25M ppd target. The project has 308,197 atoms, so should not be much different on windows. Possible you are having issue with windows and/or nvidia driver? Have you checked the GPU clocks and made sure CUDA is being used?

 

[WhitehawkEQ]

It's the WU's, I checked the clocks and all are where they should be, as for the CUDA, I have no idea where to check for that. Now WU 18201 is a low point producer for me as well, but now as low as 18601.

 

[HayesK]

the client log will show use of CUDA or OpenCL at start of wu. P18201 is another typical ppd wu, 302,000 atoms.

20:00:57:WU00:FS02:0x22:********************************************************************************
20:00:57:WU00:FS02:0x22:Project: 17258 (Run 2410, Clone 2, Gen 34)
20:00:57:WU00:FS02:0x22:Reading tar file core.xml
20:00:57:WU00:FS02:0x22:Reading tar file integrator.xml
20:00:57:WU00:FS02:0x22:Reading tar file state.xml.bz2
20:00:57:WU00:FS02:0x22:Reading tar file system.xml.bz2
20:00:57:WU00:FS02:0x22:Digital signatures verified
20:00:57:WU00:FS02:0x22:Folding@home GPU Core22 Folding@home Core
20:00:57:WU00:FS02:0x22:Version 0.0.20
20:00:57:WU00:FS02:0x22:  Checkpoint write interval: 10000 steps (2%) [50 total]
20:00:57:WU00:FS02:0x22:  JSON viewer frame write interval: 5000 steps (1%) [100 total]
20:00:57:WU00:FS02:0x22:  XTC frame write interval: 50000 steps (10%) [10 total]
20:00:57:WU00:FS02:0x22:  Global context and integrator variables write interval: disabled
20:00:57:WU00:FS02:0x22:There are 4 platforms available.
20:00:57:WU00:FS02:0x22:Platform 0: Reference
20:00:57:WU00:FS02:0x22:Platform 1: CPU
20:00:57:WU00:FS02:0x22:Platform 2: OpenCL
20:00:57:WU00:FS02:0x22:  opencl-device 1 specified
20:00:57:WU00:FS02:0x22:Platform 3: CUDA
20:00:57:WU00:FS02:0x22:  cuda-device 1 specified
20:01:27:WU00:FS02:0x22:Attempting to create CUDA context:
20:01:27:WU00:FS02:0x22:  Configuring platform CUDA
20:01:42:WU00:FS02:0x22:  Using CUDA and gpu 1
20:01:42:WU00:FS02:0x22:Completed 0 out of 500000 steps (0%)
20:01:43:WU00:FS02:0x22:Checkpoint completed at step 0

Post magically merged: Mar 13, 2022

When all else fails, reboot.

 

[harlam357]

I'll be adding this detection OpenCL/CUDA to HFM soon. As I'm getting close to the end of the migration of the Work Unit data. I want to capture a few more pieces of data along with each work unit. Driver version(s), at least nvidia, is also on the list.

 

[WhitehawkEQ]

OpenCL

 

[HayesK]

Sounds like the nvidia driver needs to be re-installed.

edit: Note that the checkpoints for CUDA and OpenCL are not compatible, so best to complete the wu in progress to avoid losing the work already completed. Otherwise the wu will re-start from zero.

 

[WhitehawkEQ]

ok, I looked in the core dir (folder for you young kids) and there in OpenMMCPU.dll, OpenMMCUDA.dll and OpenMMOpenCL.dll, so I it seems I'm not getting CUDA WU's right now.
Oh I do have core version 22-0.0.2 and I did reboot.

Oh 1 other thing, do I need GTX1080 drivers as well as RTX3080 drivers if I put a 1080 and a 3080 in same PC or just 1 driver?

 

[HayesK]

Both your cards should be using CUDA on all the current wu and the same driver will support both cards.

Most of my linux hosts are on driver 495.46, but a few are still running 470.86 to provide support for the Kepler gpu. Core22 v0.0.20 does not support the CUDA version used by the Kepler gpu, which fall back to OpenCL. On one of my hosts with mixed cards and driver 470.86, Core22 v0.0.20 runs with CUDA on the Maxwell 750Ti/970 and OpenCL on Kepler 650Ti Boost, all running wu at same time.

When the new core22 was released, there were some issues with cards using old drivers falling back to OpenCL. Recommendation was to use current release drivers downloaded from nvidia. I have not tried the current 510 series, but the linux 470.86/495.46 drivers are working fine for me.

 

[WhitehawkEQ]

I insatalled 511.79 on 1 of my PC's with a 2080 and still using OpenCL even after a reboot. Would it have anything to do with Windows vs Linux for the BarraCUDA?

 

[HayesK]

My only windows folding is a few wu on W10 notebook with nvidia 960m, which was folding with CUDA using driver 496.49 a few weeks ago.

You could try pausing all slots, deleting the core22 folder and allow fresh copy c22 to download.

Are you still on W10?

Some of my hosts are dual boot W10/Ubuntu, so I could try folding in W10 with newest driver and W10 updates, but would need to finish work in progress and get all the updates installed.

 

[WhitehawkEQ]

Win 10 on 6 pc's, Win 7 on 2 pc's

 

[harlam357]

Drivers has always been my issue with OpenCL vs. CUDA. Running 511.79 on CUDA on all my PCs (W10/W11) with no issues.

I've seen where installing the nvidia driver doesn't override the drivers provided by microsoft. Check your device manager and make sure it's actually using 511.79.

 

[WhitehawkEQ]

I wonder if the client has anything to do with using CUDA vs OpenCL? I'm using 7.4.4, should I upgrade it?

 

[HayesK]

I wonder if the client has anything to do with using CUDA vs OpenCL? I'm using 7.4.4, should I upgrade it?

Don't know, but could be the issue. Are any GPU on the 7.44 hosts using CUDA?

 

[WhitehawkEQ]

No, all OpenCL, I'll try installing current client on 1 PC and see what that does.

 

[torin3]

No, all OpenCL, I'll try installing current client on 1 PC and see what that does.

I've still got 3 1080Ti cards and they are all on client 7.6.21 and are running CUDA, not Open CL.

 

[WhitehawkEQ]

any way to force CUDA?