BitBOINC

I have a host with multiple GPUs for testing in here. The strange thing i was wondering is....GPU0 shows flatlined 100%, GPU 1 and 2 flatline zero. All WU have the same %..and thats not possible because the GPU0 is faster then 1 and 1 faster then 2.

But there are 3 WUs running. IN BOINC manager it seems to run every one GPU, 1 piece of WU, but technically said..it seems all 3 running on GPU0...because at home..on single card i dont have anywhere near 100%...

Sooo the conclussion is, perhaps any WUs are redirected to GPU0.

Im not sure if this problem existed on another project too.... was it WCG in GPU beta development? dont know..

PS: I will try in few hours when the WUs are done, to redirect a WU to GPU2 only. Just to see if it POSSIBLY could run on any other GPU.

---

---------- 24/7 Crunching since 2011 -----------
DSKAG Austria: http://www.dskag.at

Sooo i tested it, even when i exclude gpu0 and 1 and start it on GPU2 (BOINC Manager shows 2 too) it got reroutet to the hardwares GPU0..i see it in afterburner.

As far as i know is this a missbehaviour between the app and boinc.

---

---------- 24/7 Crunching since 2011 -----------
DSKAG Austria: http://www.dskag.at

It seems that tasks at checkpoint locations are being assigned to a single GPU for some reason; I will fix this.

---

I don't really understand what is happening. According to BOINC one task is running on GPU0 and one on GPU1. Yet they are both going at exactly the same speed, despite the fact that the discrete video card has about 15 times the processing power of the onboard graphics.

A single GPU is likely handling the entire workload, and BOINC is misinterpreting this. I am researching and working on how to fix these issues.

---

And looking at my computer's page, the 32GB of RAM it is showing I think means it is the on board graphics card using up to that amount of system RAM.

Edit: I have found the issue in my machine I believe.
In cc_config.xml I had

      <use_all_gpus>0</use_all_gpus>

rather than

<use_all_gpus>1</use_all_gpus>

I have now changed this but because I have an older CPU task still running that will start again if I exit BOINC I won't test it till this finishes. The discrete card only has 8GB RAM, but the extra speed should more than make up for that.

Believe me thats not the issue ^^ all multigpu machines should enable this by default, so also i have that enabled by many years ^^

---

---------- 24/7 Crunching since 2011 -----------
DSKAG Austria: http://www.dskag.at

Believe me thats not the issue ^^ all multigpu machines should enable this by default, so also i have that enabled by many years ^^

They certainly weren't in my case. I am new to GPU computing so won't argue on it being the issue. It will probably be tomorrow morning before I get the chance to see if enabling them all makes a difference in my case.

Even after restarting BOINC both for allowing both cards to be used and making the discrete card the priority one, it is still the on board graphics that is being used for tasks.

If you're running Windows, that can also be an issue.
I've got a way better Multi-CPU workload and balancing after switching to Debian. Maybe that could work with multi-GPU setup as well )

Just as guess! You can try in BIOS to look what GPU it looks first iGPU or PEG...perhaps it is set to igpu and you dont have too (dont know your desktop setting) and rotating it..perhaps..dont know..could turn the numbers in BOINC around..so that the intel is shown as GPU1 and it uses your discrete card as GPU0 then. Like i said, just something you can look. I dont know that myself, but i would try for testing purpose ^^

---

---------- 24/7 Crunching since 2011 -----------
DSKAG Austria: http://www.dskag.at

I will have a look at that next time I reboot. Both GPUs in my case are AMD/ATI

Sat 03 Oct 2026 07:53:10 AM BST |  | OpenCL: AMD/ATI GPU 0: AMD Radeon RX 6600 (driver version 3614.0 (HSA1.1,LC), device version OpenCL 2.0, 7.98GB, 7.98GB available, 9.86 TFLOPS peak)
Sat 03 Oct 2026 07:53:10 AM BST |  | OpenCL: AMD/ATI GPU 1: AMD Radeon Graphics (driver version 3614.0 (HSA1.1,LC), device version OpenCL 2.0, 31.01GB, 31.01GB available, 563.20 GFLOPS peak)

It looks very much like it is your original issue of tasks running on just the first GPU that BOINC finds.
BOINC8.3.0 XUbuntu26.04

Next experiment here, with 2 tasks running and just one in the queue, I am going to see if a single task completes in close to half the time on the GPU.

In reply to Dave's message of 4 Oct 2026:

Next experiment here, with 2 tasks running and just one in the queue, I am going to see if a single task completes in close to half the time on the GPU.

Yes, it does run twice as fast, confirming two tasks do run on the same GPU.