Message boards :
Number crunching :
Wu Size
Message board moderation
Author | Message |
---|---|
Yakk Send message Joined: 22 Feb 10 Posts: 6 Credit: 5,488,476 RAC: 0 ![]() ![]() |
My 5870 has been burning through WU's in +/- 80-85 seconds per WU. However between WU's I can see a drop in GPU activity, it is very short, but it is a drop. This drop is it lost time as the WU changes? Would longer WU's increase effeciancy? Or it is code being unloaded & reloaded into the graphics card? I'm only asking this questions because the WU's fly by so quickly this drop happens very often. Y. |
TJ Send message Joined: 12 Aug 09 Posts: 262 Credit: 92,631,041 RAC: 0 ![]() ![]() |
This drop is it lost time as the WU changes? Would longer WU's increase effeciancy? Or it is code being unloaded & reloaded into the graphics card? I'm only asking this questions because the WU's fly by so quickly this drop happens very often. Hi Yakk, This normal. When a WU is finished the GPU has less to do and its activity degreases. When a few secs later a new WU starts it increases again. And so on. You can see the same when looking at its temperature curve. Greetings from, TJ |
Zydor![]() Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 ![]() ![]() ![]() |
There is an unload / load time of 2 secs a go, as the old one is pulled out by the cpu, and the new one is loaded. On some slower cards that can be about 3 seconds, nothing you can do, its a pure load/unload process |
Vid Vidmar*![]() Send message Joined: 29 Aug 07 Posts: 81 Credit: 60,360,858 RAC: 0 ![]() ![]() |
That's why some of us use app_info.xml files to get 2 WUs running concurrently on single GPU. If their start time is offset by more than 2s then there is no dip in GPU activity. BR, ![]() |
Emanuel Send message Joined: 18 Nov 07 Posts: 280 Credit: 2,442,757 RAC: 0 ![]() ![]() |
So yes, having longer WUs would technically increase GPU utilization slightly. But there's no simple way to artificially increase WU size - the WUs would actually have to do more calculations, or do the same calculations at a higher precision. That has already been done several times, but at this point I think you'd have to develop a more complex model to benefit from increased precision or wider strips of the sky. |
©2023 Astroinformatics Group