Has anyone else had BOINC/CUDA running a different numbers of jobs at different times?
Some times BOINC runs 3 jobs plus a CUDA job on my quad core, and others it runs 4 jobs + CUDA. My cc_config is set to 4 cpus, and not 5 (to force 5 jobs).

Most of the time BOINC runs 3 CPU + 1 GPU, but occasionally it runs 4 cpu jobs +1 gpu. Anyone else get this?

---

Bambi

BM version?

---

My guess is that when it is running 3+1, you are running 3 SETI MultiBeam and 1 CUDA. Then when you are running 4+1, you are running any four AP tasks or a different project and 1 CUDA.

The limitation is that when CUDA is running with MultiBeam, only 3 CPUs can be used (unless the cc_config is modified to allow 5 CPUs). Whereas with any other app such as AstroPulse or any other project, this limitation does not apply.

This is being looked into for the next BOINC release. The plan is to support all CPUs + GPU regardless of what app is running.

---

BM version?

I dont ask withaout a reason. I heared rumoring BM 6.5.x handels cpu+gpu without cc_config. Maybe its a bug (or you confuse the BM) if you are using 6.5.x and have set <ncpus> in cc_config ...

---

My guess is that when it is running 3+1, you are running 3 SETI MultiBeam and 1 CUDA. Then when you are running 4+1, you are running any four AP tasks or a different project and 1 CUDA.

The limitation is that when CUDA is running with MultiBeam, only 3 CPUs can be used (unless the cc_config is modified to allow 5 CPUs). Whereas with any other app such as AstroPulse or any other project, this limitation does not apply.

This is being looked into for the next BOINC release. The plan is to support all CPUs + GPU regardless of what app is running.

Don't think that this can be the case, OzzFan, since that host has not returned an MB crunched on the CPU since 29th Dec.

Anyway, surely once Boinc recognises CUDA capability, all MB's will be tagged on download to crunch with CUDA so once the pre-CUDA cache has been used up one can't crunch MB on *both* the CPU and GPU? The modified cc_config is to allow a fourth AP to crunch on the 99.6% of the core that otherwise goes to waste.

F.

---

Has anyone else had BOINC/CUDA running a different numbers of jobs at different times?
Some times BOINC runs 3 jobs plus a CUDA job on my quad core, and others it runs 4 jobs + CUDA. My cc_config is set to 4 cpus, and not 5 (to force 5 jobs).

Most of the time BOINC runs 3 CPU + 1 GPU, but occasionally it runs 4 cpu jobs +1 gpu. Anyone else get this?

I saw that only when the other tasks were high priority. I fixed that with 3+1 on my quad Q6700 Vista 64

I have seen this when running MB-CUDA on my 9600 GT and Einstein + Climate on CPU.
It usually starts with 3+1 and after a few CUDA is done it starts up a 4th CPU-task so it then runs 4+1, when this happens the GPU-task does slow down considerably.

This is with BM 6.4.5 and no cc_config

---

It usually starts with 3+1 and after a few CUDA is done it starts up a 4th CPU-task so it then runs 4+1, when this happens the GPU-task does slow down considerably.

Same here, as I reported in another thread. My only S&H application is CUDA Multibeam, but I run other projects on my CPUs.

Most of the times BoincManager starts by running 3 CPU tasks, but sometimes after a while it switches to 4. I was not able to figure out under which condition it decides to do so.

I did not mess with ncpu setting, either.

---

I posted something similar to this resently; Crunching 6 WU on 5 cores...?

I was asked if I was running the cc_config override, which I am not. I only experienced the 5 CPU + 1 GPU with Einstein.

But for me, most of the time it seem that it depend on what CPU WU's I am crunching at the time. And this happens on either 6.4.5 or 6.5.0.

Check my Signature to see my current basic system config.

---

Q6600 @ 2.40GHz
Vista Ultimate x64 SP1
Mem: 4GB
8800GT 181.22
BOINC 6.6.3 Win64

ACTIVE PROJECTS