The CPU usage of the GPU application on my machine is running at 0.92 of a CPU.

This is rather high and taking up a almost a full CPU as well as the GPU.

I run other project on the CPUs and would prefer that the CPU usage was down to under 0.1 of a CPU (0.05 would be better) as most other GPU projects that I run.

Using an Appinfo file would this be possible or are the work units set to use this much CPU and it wont change?

Conan

The client which is running under-the-hood is the distributed.net client.
I've tried running that original client and it seems this client is taking so much CPU (50% of this machine's CPU)

I'm sure there's a way to decrease that value anyhow for us, because the default is way too high.

Stay tuned.

For the time being I am going to stop running this project as it is making my CPU projects run so slow.
Normally eon2 would only take 7 minutes per work unit, now it is over 40 minutes.
No idea how much longer it will make my Yoyo work run but I have noticed a slow down adding many hours onto a work unit.

I will check back but it is killing my CPU output.
I don't intend to try and allocate 1 core to a GPU project, I want them for my CPU projects.

I will check back a little later.

Conan.

---

@Admin:
You can ask the admin of DNETC@home to setup their applications here.

I just ran into the same type of problem. Workunits running for 45 minutes with no GPU activity and CPU was getting hammered hard. Did a reboot and started fresh WU, same thing. Was working fine until an hour ago or so.

Ati RAdeon 5970 and CPU usage 50% (2 GPU core = 100% load 2 CPU core) :(

I'm seeing 0% CPU usage here (real one, not the wrong setting in the boinc client which should be drastically reduced), both with the official distributed.net client and the wrapper ... (5770 on XP with 11.3 drivers)

What is your setup ?

When I check Task Manager, I also notice what appears to be very little CPU usage. However this does not improve things on my machine. Work units on my CPUs still run extra long times when Moo!Wrapper is running and the GPU is woefully slow. So I believe that it is still taking a lot of resources I just can't see what they are.

Computer AMD Opteron 285 with 2GB RAM and single ATI 5870 graphics card and ATI 11.1 drivers.
(This setup would run a DNETC@Home WU in about 19 minutes, dual cards half that time but one card went faulty).

Run time for Moo!Wrapper is over 2 Hours per WU.

My AMD Phenom II X4 955 with 4 GB RAM and 2x ATI 5870 cards in it with ATI 11.1 drivers.

Run time is 9 Minutes per WU.

So I can not run this project on a single card computer.

Conan.

---

I've noticed the dnetc-1.00.ini (same as d.net's dnetc.ini) is initialized as:

[processor-usage]
priority=4

By default, it's be set to 0, is there any reason why this is initialized as 4?

I've noticed the dnetc-1.00.ini (same as d.net's dnetc.ini) is initialized as:

[processor-usage]
priority=4

By default, it's be set to 0, is there any reason why this is initialized as 4?

If you set it to 0, cpu usage went down to 12% but gpu usage also to around 63%.

By default, it's be set to 0, is there any reason why this is initialized as 4?

Yes, there is. That's the recommended setting for GPU crunchers. This way the client gets CPU cycles to keep feeding GPU and thus keeping them busy. Normally it shouldn't matter since the clients are supposed to only use fraction of CPU time.

I've seen this same high CPU on my CUDA system. (You can watch it here -> http://moowrap.net/results.php?hostid=3.) I've actually managed to get it down to reasonable levels by adjusting the core used by the client. (This was suggested by some posters around here, a big thanks goes to them for suggesting this!)

You can try this yourself too, there's now a preference for this (see http://moowrap.net/prefs.php?subset=project) and the v1.01 application should feed it to the client. It seems that by default GPU crunches select core 0, which is not optimal for neither Stream or CUDA. (There's bugs open at http://bugs.distribute.net to change their defaults.)

I'd think core 3 could be best for Stream and core 10 for CUDA clients. Although, you should experiment since this can be different for your setup.

-w

PS. It seems to core selection is broken on Linux wrappers but a temporary workaround to set the core is to use anonymous platform and feed -c via job.xml to the client. (Or wait until I get that fixed.)

I'd think core 3 could be best for Stream and core 10 for CUDA clients. Although, you should experiment since this can be different for your setup.

YES and no.

9 uses a full CPU core and pushes the GPU to the limit and is fastest mostly.
10 uses less than 1% CPU and utilizes the GPU above 90%.
11 and your milage may vary.

maybe 2, 5 or 8 will be faster on cards with a high number of cores.

but everyone can test it by running the app from commandline with --benchmark

Against DNETC apps all prefs and current apps are inefficient.