Gravitational Wave search O2 Multi-Directional 2.09 (GWnew)
18 Jun 2021 13:14:44 UTC
Topic 225571
(moderation:
Hello,
my computer does not be able to compute any "Gravitational Wave search O2 Multi-Directional 2.09 (GWnew)" task. These tasks consumes abnormal RAM 48GB+ and hundreds GB SSD from the pagefile. Task will never be finished and freezes on integer value of percentage. I need to manualy kill these tasks because they are blocked my logical CPUs. Sometimes my computer due this issue freeze completely and then restart is needed.
Please help.
Computer:
Motherboard: ASUSTeK COMPUTER INC. (ROG STRIX Z390-F GAMING; version: Rev 1.xx)
BIOS: American Megatrends Inc. (version: 0805; date: 29.01.2019)
Processor: Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz 3.60 GHz
Installed RAM: 48,0 GB (usable: 47,8 GB)
System type: 64bit platform x64
SSD: Samsung SSD 970 EVO Plus 2TB (1,8 TB)
Operating system: Microsoft Windows 10 Pro (version 10.0.19043; build 19043)
I also have another computer which computes the "Gravitational Wave search O2 Multi-Directional 2.09 (GWnew)" tasks without problems but it is slow and old.
Magman wrote: Hello, my
)
You are trying to run gpu tasks on an "INTEL Intel(R) UHD Graphics 630 (19595MB)" gpu which is essentially your cpu and that just doesn't work which is why all the errors, turn the gpu tasks off and only get cpu tasks and you should be able to get thru a whole bunch of tasks pretty quickly with that pc. GW gpu tasks require a stand alone gpu with at least 4gb of ram.
mikey wrote:You are trying to
)
Mentioning GW GPU tasks when he's not actually running any is just confusing the issue.
Also, saying that using the Intel GPU "just doesn't work" isn't correct because he has validated tasks for both GRP and BRP4 for that GPU.
The real problem (most likely) is that he is trying to run too many tasks (of whatever flavour) on that machine. It only has 8 real cores (16 threads) so if he tries to use the GPU plus as many CPU threads as possible, the performance is likely to be woeful.
The only way to get the most efficient performance is to start low and gradually increase the number of concurrent tasks until the performance starts to suffer. If it were my machine (and if I wanted to use the Intel GPU) I'd start with just a single GPU task (GRP and NOT BRP4) and get the figure for the fastest GPU crunch time. I'd then start allowing CPU tasks (say 4 to start with) and get those times. Hopefully there would be little to no impact on the GPU crunch time. I'd then gradually allow extra CPU tasks until performance worsened enough for overall output to decline. The number is likely to be well short of the full 16 threads.
Cheers,
Gary.
Gary Roberts wrote: mikey
)
You are of course correct and I stand corrected!!