Well, I definitely am running an official BOINC release, not some hacked version. There is 1 out of 3 computers with gpu(s) that I have that is getting extra work units. It has 15 at the moment. It is a 4core/8thread i5 with a intel uhd 620 igpu. I have 2 of that series laptop, but the other one has a 4core/8thread i7 with the same igpu. The second one is not getting extra units.


The 1/3 of my computers with an iGPU has extra. The other 1 with an iGPU did not get extra. The 1 with an RTX 2060 and an iGPU did not either.

Thanks for all the updates Keith. I'm roughly on the other side of the world, which is why I tend to talk during a different part of the day.

Studying Intel iGPU results. Interesting case:

https://www.worldcommunitygrid.org/ms/device/...s.do?workunitId=590484762

Five of us have tried it, four have failed.
Mine was replication _3. It got to 37.5% progress (16 AutoDock runs), but then stalled. The workunit display shows 0.01 hours CPU, but the host page shows 8.21 hours elapsed.

The next two (Windows 10) machines completed fewer tasks - 4 and 1 respectively - but recorded ~ 3 hours CPU time. I can't see the elapsed time. We were all killed off by the timeout error.

The final user - replication _5 - completed the tasks in 157 seconds CPU, 31 minutes elapsed, but has status 'Too Late'.

I have three other iGPU tasks still shown as 'Running', but in practice doing nothing. They have been active for around 12 hours, and I expect them to time out after about another four hours. I'll gather what further information I can, and write them up later.

I ended up getting 11 WUs on the ATI Radeon hd 5770, 1 errored out with the exceeded elapsed time limit. https://www.worldcommunitygrid.org/ms/device/...og.do?resultId=1588786117
and 1 was server aborted. https://www.worldcommunitygrid.org/ms/device/...og.do?resultId=1588882096

The GT 1030 got 4 WUs all validated fine.

That is too bad. It was so close to finishing too! This is different than the other timeout ones with 7.28 that I have seen. This one has the timeout exception only 4min and 23s after starting the final job.

Thanks for picking that up. I've preserved the two big output files, in case Keith wants to have a look at them, but no error messages are obvious to my eye.
Sadly, I already know they're going to fail. Here's the complete stderr output from an example:

projects/www.worldcommunitygrid.org/wcgrid_beta29_autodockgpu_7.28_windows_x86_64__opencl_intel_gpu_102 -jobs OPNG_0000089_00015.job -input OPNG_0000089_00015.zip -seed 1482125208 -wcgruns 1700 -wcgdpf 34 
INFO: Using gpu device from app init data 0
INFO:[20:18:03] Start AutoGrid...

autogrid4: Successful Completion.
INFO:[20:18:08] End AutoGrid...
INFO:[20:18:09] Start AutoDock for ZINC000904329721_2-ACR2.14_RX1--fr2266benz_001--CYS114.dpf(Job #0)...
OpenCL device:                           Intel(R) HD Graphics 4600

It's been pegged at the placebo limit of 100% since I went to bed last night. The last sign of life was

<active_task>
    <project_master_url>http://www.worldcommunitygrid.org/</project_master_url>
    <result_name>BETA_OPNG_0000089_00015_1</result_name>
    <checkpoint_cpu_time>5.335234</checkpoint_cpu_time>
    <checkpoint_elapsed_time>8.230011</checkpoint_elapsed_time>
    <fraction_done>0.000000</fraction_done>
    <peak_working_set_size>93335552</peak_working_set_size>
    <peak_swap_size>90198016</peak_swap_size>
    <peak_disk_usage>245937</peak_disk_usage>
</active_task>

Process Explorer shows something's happening, but not a lot. The highlighted line flashes back to zero every other second.


For comparison, here's what a similar machine looks like when it's working on its day job at Einstein@Home:


And I think that's about all I can say. I've preserved (I think) enough of the entrails to construct an offline test run, but I think for that to be useful we would need a binary with better debug instrumentation.