I've started using BOINC TASKS .44 recently and am wondering about some settings for SETI. In the APP_INFO.XML file, I have upped the avg cpu from .04 to .14 based on the fact that BOINC TASKS shows that running a CUDA task takes on an average 20-22% of a cpu. What are the implications of this update of the avg parameter?


From app_info.xml file, I'm now running with:
<avg_ncpus>0.140000</avg_ncpus> - this used to be .04
<max_ncpus>0.250000</max_ncpus> - this used to be .04

From BOINC TASKS screen, it shows:

On a CUDA task: USE 0.14C + 1NV

For a CUDA task: 21.55% CPU

Should I change the parameter back? What difference does it make?
Alex

---

yes change it back. they created it to work correctly. The majority of the CPU usage is to load the WU into the GPU. after that its all GPU. so you can hold your CPU back by do this

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

What difference does it make?

None for the execution of the task, as long as the CPU part is lower than 1.0. If it is, another CPU-only task is allowed to run concurrently.

Gruß,
Gundolf

---

Computer sind nicht alles im Leben. (Kleiner Scherz)

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

I guess my question is why is BOINC TASKS showing that a CUDA task is taking 20-30% of a single cpu (on an 8 cpu system)? My GPU monitor says that it is using 55-60% of the gpu, so the numbers don't jive.

I'll change it back to .04 but just doesn't make sense.

Alex

---

I guess my question is why is BOINC TASKS showing that a CUDA task is taking 20-30% of a single cpu (on an 8 cpu system)? My GPU monitor says that it is using 55-60% of the gpu, so the numbers don't jive.

I'll change it back to .04 but just doesn't make sense.

Alex

For a multi-CPU system with a single GPU, the setting matters not at all as long as it's less than 1.0. It is used only by the BOINC core client to decide what to run.

IIRC, the actual CPU usage does tend to be high with 196.xx drivers, without a corresponding reduction in elapsed time.

                                                               Joe

[ADDED]If you set CPU to < 1, it's considered 'useless'and/or use GPU instead?

And probably you'll see, a shorter CPU time, if you 'give' it 0.14 (14%).
You can be sure that this is tested, before using a standard percentage CPU usage.
That's probably why one of my host's, run 6 task's, at a time.(X9650@3196MHz)
4 CPU's (25%+24%+20%+21%)& 2x G92 (GTS250/9800GTX+). During the GPU loading by CPU, after this, 4 WU's run on the CPU and 2 on each GPU.
Appears to be possible. Sometimes those annoying triplet found twice, shmem/shmget faults, found about 10 when on 03-23-10 at 14:04, about 400WU's
had been uploaded and validated or errored out.

---