High amount of CPU WU discarded

Started by Caprio, September 28, 2012, 05:09:40 am

Previous topic - Next topic

Caprio

I have 100 recent CPU WU invalidated:

http://setiathome.berkeley.edu/results.php?hostid=3299266&offset=0&show_names=0&state=5&appid=

What is the cause of this? Does BoinC have trouble with High amount of GPU processing power compared to CPU?
Id scheduler try to send somehow the WU to CPU instead of GPU?

Claggy

September 28, 2012, 07:58:18 am #1 Last Edit: September 28, 2012, 08:00:29 am by Claggy
Read the numerous threads in Number Crunching, or the FAQ entry at lunatics:

http://lunatics.kwsn.net/1-discussion-forum/faq-read-only.msg40265.html#msg40265

QuoteQ: I have tasks in my task list, that timed out with an impossibly short deadline!

A: Those tasks are VLAR and therefore won't be sent to a NVidia GPU.

The following happened:
Your host asked for work. The CPU got assigned some VLAR tasks (marked as .vlar in the task name). Due to a communication hiccup your host never received the server reply with the tasks. After a while it asks again for tasks. It also sends a complete list of what tasks it already has. The server compares this to what you got assigned. It finds those .vlar tasks you never got and - with 'resend lost results' active - wants to send them again. Only this time your host is asking for NVidia GPU tasks. As the .vlar tasks can't be send to NVidia GPU they expire instead - showing as time out with impossibly short deadline.

NB This can also happen to AP tasks, if there is no GPU app.



It also happens if you aren't running the same app on the GPU, ie you have AP on the CPU, but not on the GPU, or MB on the GPU, but not on the CPU.

Claggy

Powered by EzPortal