WCG Beta Testing GPU Work
- Thread starter doneske
- Start date
GPU tasks should be plentiful for the next couple of days: https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,43386. After that, back to the limited supply.
Hope everyone is doing well!
Hope everyone is doing well!
Not 15 minutes into the stress test and I already can't download or upload any work units.
EDIT: Seems they broke something before the test actually started. https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,43386_offset,30#656616
EDIT: Seems they broke something before the test actually started. https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,43386_offset,30#656616
This stress test has lead to a lot of work units to observe how my systems handle them. Lots of idle time on the GPU with only one task running per GPU leading to much more temperature fluctuation than I care for. I find three task to be enough to keep temperatures pretty stable on a GTX 1060 and RTX 2070.
XML:
<app_config>
<app>
<name>opng</name>
<gpu_versions>
<gpu_usage>0.33</gpu_usage>
<cpu_usage>1.00</cpu_usage>
</gpu_versions>
</app>
</app_config>
Last edited:
I've been getting plenty of work but am having trouble with both up and downloads tonight, mainly on the machine with the wonky networking.. I think the problem is made worse by the short run times, these are taking less than 5 minutes on a 2080 Ti. I haven't bothered to track GPU usage, I'm just sticking with the default.
Running five tasks simultaneously on my RTX 2070 results in an average GPU utilization of 76.4% measured over two hours. If I throw more tasks at it my NVMe thermal throttles. Seems active cooling of my storage or a RAM disk might be necessary to gain more performance for OPNG in its present state.
Last edited:
Now that the server / network hiccups seem to be solved these are running great. As mentioned utilization is poor but I haven't bothered with trying to max out the GPU. I feel that my systems are actually CPU bound which isn't a surprise considering their age. Run times seem to be in the 12-15 minute range on both Windows and Linux which is a little surprising, there's usually a larger difference. I'm running some multithread work on the Linux box though so that could be why performance versus Windoze there is weaker than expected.
It's nice to have another health / biomed BOINC GPU project besides GPUGrid.
It's nice to have another health / biomed BOINC GPU project besides GPUGrid.
Tons of work for Nvidia cards right now with the stress test going.
Placed a fan to keep my NVMe from thermal throttling. Seven tasks maxes out my RTX 2070 with a 3900X. Disk writes average 26.9 MB/s. That's 2.3 TB of writes a day. There is no way I'm going to run this long term if they don't make better use of system memory.
Thank you for the information! I will change my FAH computer (the one with the NVIDIA card) for a while, to see if it can get this OPN/WCG work.
Yeah, save something for the next THOR Challenge (I'm looking forward the 2021 one)!
Already working fine! New badges will come more easily, I guess! Let's go!
I never look at that, so thanks for mentioning it. No wonder the drive was getting hot. Are those figures a total or per task?
Total.
I've decided to put the BOINC/slots folder on a ramdisk on my main system and just not run OPNG on anything else. It's sad that they set up the program to write all those temporary files to disk when they could easily fit in memory. The slots folder never gets larger than one gigabyte with seven tasks going and what we upload is just a few megabytes.
I've decided to put the BOINC/slots folder on a ramdisk on my main system and just not run OPNG on anything else. It's sad that they set up the program to write all those temporary files to disk when they could easily fit in memory. The slots folder never gets larger than one gigabyte with seven tasks going and what we upload is just a few megabytes.