I have SETI set to use GPU only. It is set to NO for CPU. Three computers use a minimial amoung to CPU. One computer is using 100% of one processor on the machine. The other three use less than 1%. The GPU is an 8400 GS. The computer is a AMD 9500 Quad.

---

Uses Excessive CPU. I have an E6750 and 8600GT OC with 256MB on a Gigabyte GA-P35C-DS3R with 4GB Ram. Boinc 6.4.7 and Driver 182.50. When Boinc started it loaded 29 Seti@Home Enhanced 6.08(cuda) WUs.

I have set SETI to not use CPU and to use GPU if available. SETI, however, is using 49% of the two CPU processor times as indicated in Win Task Mgr.

Why so much CPU usage and why so many SETI WUs?

---

Uses Excessive CPU. I have an E6750 and 8600GT OC with 256MB on a Gigabyte GA-P35C-DS3R with 4GB Ram. Boinc 6.4.7 and Driver 182.50. When Boinc started it loaded 29 Seti@Home Enhanced 6.08(cuda) WUs.

I have set SETI to not use CPU and to use GPU if available. SETI, however, is using 49% of the two CPU processor times as indicated in Win Task Mgr.

Why so much CPU usage and why so many SETI WUs?

seti uses a fair amount of cpu for the cuda app during the first 30 seconds or so after loading. after that it sits at 2-3% on my machine.

one suggestion is to try loading 6.5.0 of boinc
32 bit http://boincdl.ssl.berkeley.edu/dl/boinc_6.5.0_windows_intelx86.exe
64 bit http://boinc.berkeley.edu/dl/boinc_6.5.0_windows_x86_64.exe

or try different nvidia drivers.

---

No change after putting in BOINC 6.5.0. Still using 100% of one of the processors. It is still indicating in the Status column (0.12 CPUs, 1 CUDA) after 10 minutes.

---

No change after putting in BOINC 6.5.0. Still using 100% of one of the processors. It is still indicating in the Status column (0.12 CPUs, 1 CUDA) after 10 minutes.

I checked the only task I can see you've completed and saw "Falling back to HOST CPU processing..." so the cpu processed that task and not the gpu. It appears to stem from how much of the cards memory is available.

Are you using the gpu to run a screensaver, more than one monitor, or video games? Also there's info in this thread that may apply.

---

Thanks for the info. I run an MSI 8600 GT OC at factory overclock and no screensaver. I do however run dual 1280x1024 displays. I turned off one when a new SETI WU was started and the 3+ hour task ran in 12 minutes with less than a minute of CPU time. The GPU bumped up 10 degrees. During the WU I restarted the dual displays and the task kept working in the GPU. The next WU also started with only 1 or 2% CPU time. I don't run games but do edit movies and photos on this machine but did not during the CUDA installation and configuration.

It now appears to be working (assuming the work completed is correct), even with the dual displays working.

Thanks

alan

So, am I to understand that if there is not enough RAM for the SETI Cuda, then it switches to CPU?

How much RAM is required? I have two 8600 GTS, with 256 MB RAM each. It was designed to be a SLI machine, but SLI is off so that two GPU's can be used at once for CUDA.

Using BOINC 6.6.20, I was finally able to get 6 tasks running in parallel, but the CUDA tasks don't always seem to be running.

One of them ran for 4 hours without incrementing a single percent. still at 0.000% complete while the other GPU had finished two Work Units in that time.

Later on, I noticed that the two CUDA applications were using two out of 4 CPU's. One process had 1 hour, 39 minutes of CPU time, the other had 30 minutes. The rest of the science applications were splitting the remaining 50% amongst themselves.

I am using NVIDIA 182.50 driver.
Windows XP, 32 bit.

I am not overclocking my CPU or GPU's and have no programs beyond Task Manager monitoring them.

After exiting BOINC and restarting, I am running 6 applications again, one of them is progressing nicely with the Video Card, while another is spinning a full Processor leaving one science project getting 1% cpu time on occasion.

Maybe the solution to this is to tell BOINC to only use one CUDA device, rather than spin the CPU?

---

Maybe the solution to this is to tell BOINC to only use one CUDA device, rather than spin the CPU?

What you described is a stuck task in video memory. At this moment when BOINC stops one task and starts the next it doesn't check if the video memory is available and then it can happen that remnants of a previous task are still in that memory.

Seti tasks use approximately 220MB of video memory.

The BOINC Developers are now busy devising a way to halt CUDA for long enough that the video memory is cleared before the next task is started. The first test for that is in Alpha BOINC 6.6.23, but it seems to backfire, not starting CUDA tasks at all at this moment.

In the mean time, the only way to make sure your video memory is clear, is to do a full power-cycle on the computer, or in other words: reboot.

---

...Using BOINC 6.6.20, I was finally able to get 6 tasks running in parallel, but the CUDA tasks don't always seem to be running...

Perhaps you should set the preference "On multiprocessors, use at most XX% of the processors" to 75% (or less) to maximize the throughput of your machine.

For explanations see the thread Getting maximum credits.

Gruß,
Gundolf

---

Computer sind nicht alles im Leben. (Kleiner Scherz)

SETI@home classic workunits 3,758
SETI@home classic CPU time 66,520 hours