I am seeing a wide variation in CPU times from 26,000 to 45,000s per WU.
Are all WU's the same size, ie all things being equal should complete in the same amount of time?
If not the same size what variation is there?

They're all the same. To see the run time of your WUs you have to check the ouput of the individual tasks, unfortunately. E.g. here it's saying:
"# Approximate elapsed time for entire WU: 64636.438 s"

There's less variation in these times and ideally they would all very similar. The actual time / performance dependes on how much CPU time the GPU-app gets / needs, which is influenced by the other tasks running on the machine. Actually the most critical part is how fast the GPU-app gets cpu time..

MrS

---

Scanning for our furry friends since Jan 2002

How big of an effect does running multiple projects have?
Better to have GPUgrid and one other project (dedicated 24/7 crunchers so I want to use all 4 cores-quad core,XP home, (1) 8800GT/9800GT, 2 x 1GB RAM) vs GPUgrid and 2 or 3 other projects?
Thanks!

Default is to run 4 CPU-WUs along one GPU-Grid-WU. That's a good performance balance, although under windows the 4th cpu task doesn't get much time. You can increase the GPU speed a bit by setting your BOINC to use at max 3 CPUs, so you get 3 CPU + 1 GPU task. Depending on the speed of your CPU and GPU this will give you more credits / day than 4+1. With your 8800GT/9800GT 3+1 is probably the best choice.

Edit: "multiple projects2 .. BOINC really doesn't care if the cpu WUs are from one or different projects. Only the number of concurrent WUs is important here.

MrS

---

Scanning for our furry friends since Jan 2002

We are actually starting to send new WUs for testing. These are not of the same size nor they should have the same time/step.
WUs with similar names though should be the same.

gdf

We are actually starting to send new WUs for testing. These are not of the same size nor they should have the same time/step.

Interesting to hear. :-)
How will the credits look like then?

---

Member of BOINC@Heidelberg and ATA!

We are actually starting to send new WUs for testing. These are not of the same size nor they should have the same time/step.
WUs with similar names though should be the same.

gdf

Well thank you for that info- that may explain things a bit... and I second the question regarding how credit will be assigned for varying size WU's...
Thanks!

I am seeing a wide variation in CPU times from 26,000 to 45,000s per WU.
Are all WU's the same size, ie all things being equal should complete in the same amount of time?
If not the same size what variation is there?

What other projects do you run ?

Some projects use the wrapper application, and I have found some of those do not behave well, they hog cpu time, although they show they are running at low system priority, they are not because that does not show spawned threads. JAVA is the worst. I had really bad times when running the 'optimzer' application of malaria control, the gpugrid task would sit at ZERO cpu useage while those were running. I've heard reports of others too. Now I run 1 less core for CPU projects and gpugrid almost always gets its own core and runs more consistantly, mostly.

What you nned to do is observe cpu usage in task manager for windows or whatever for linux. I would start with no other tasks. Then allow one project at a time and see if any have an effct on acemd. You probably will find one project that has a large effect than others, that will be the one slowing down you cpu usage and increasing the elapsed time.

Hello:
I run optimized SETI and optimized Milkyway with GPUgrid as 3 +1... Task Manager shows SETI and MW running steady 24-25% per and GPUgrid at 19% +/- 1%... I suspended SETI and ran MW only with GPUgrid and the same suspended MW and ran only SETI with GPUgrid...XP Home, (1)9800GT>> core clock 602, shader clock 1512, memory clock 900<<
I guess based on Task Manager this is the best I can do. I am running this Q6600 OC'd @3.5 (389 x 9) on water but CPU temps are always great... GA P35 DS3L and 2x 1gb Adata DDR2 800 under clocked slightly @778
WU wV22478-GPUTEST-4-20-acemd_0 is running now and shows CPU time of 5:24:04 @ 43.294%, I dunno whats up...

Credits will be assigned based on flops computed by the wu. Credits/hour should be more or less equivalent regardless of the WUs, but not exactly identical.

gdf

@GDF: What is changed of current test WUs (except length)? Any news about these existing errors with .cu?

JAMC, everything you report looks perfectly normal :)

MrS

---

Scanning for our furry friends since Jan 2002

JAMC, everything you report looks perfectly normal :)

MrS

OK, thanks for the feedback... I'll add another cruncher to the project later in the week :)