I am running CUDA on an 8600, works fine, but uses what it calls '.03' of one of the four processors on my quad system, leaving only 75 percent of the system in use.

If I suspend the SETI using CUDA and then re-start it, for a few minutes the quad runs four tasks in addition to the CUDA, using 100 percent. After a short time one task will go to 'waiting to run' and only three tasks and the CUDA are running.

Is there a way to run all five tasks all the time?

The current limitation of BOINC is that one CPU "feeds" the GPU (hence why you get 100% for a while as it loads the data into the GPU), then it idles until the GPU task is finished.

There is a fix in the latest alpha build of BOINC that should hopefully be available to the public soon. There is a work-around, but it involves editing some XML files in your BOINC Data directory.

---

I've been looking all through the forums for this topic.

Since starting up with CUDA on a GeForce 8600GT Saturday I very quickly noticed that instead of four tasks being run at the same time I was down to one.
The single running task is using 0.03 CPU's and 1 CUDA.

While Kaylie doesn't have as poor a performance level as I do, we seem to have a similar problem.

The other tasks are from another project. My understanding is that if you have CUDA, Seti will only send CUDA tasks, except Astropulse, currently not being generated.

So, if you have an 'old Pentium D 2.8GiG', you should be better off to run f.i. 2x GT 8800/9800/260/280,(if your MoBo has 2x PCI-EXx16) and use you CPU, to feed the GPU's.

---

That only makes sense, but... Goes beyond my understanding of how BOINC might behave.

At this moment my computer is grinding away on one CUDA task and one astropulse task.
I suppose I'll just have to trust those who say this is up to 10 times faster, because after all, some of these guys really are rocket scientists doing things the rest of us can only dream of.

Three one hundredths of a CPU, Hmmmm…

I’m not sure the system administrators will like this, it is all supposed to run automatically and in the background, but some of us dedicate equipment to these things and want every ounce of computational power out of it.

There is a way to cause SETI to run on a GPU and utilize at least seventy five percent of the other processors in a quad system. Five tasks can be run consistently with participation in a second project.

The method involves only basic setting changes and procedures, preferably performed only during periods of low server traffic, and never on a Tuesday, but it is too lengthy to post here in detail. I’m not keeping it a secret though.

On another note…
I fear this may become more frequent than a slip in the shower. I hope the recent satellite collision does not cause sufficient interference to make more work looking for recognizable signals.

My computer was down for two weeks so I'm waiting for my RAC to build back up and level off.
If it goes higher than before, all is good.
If it is lower, I guess I'll have to do something like disable CUDA.

With AMD Quad idle (Phenom II X4 940 BE @ 4 x 3.0 GHz) and two overclocked GTX260 Core216 55nm.
[Only GPU crunching rig]


I used all stock and had 2 x '0.19* CPUs, 1 CUDA' in the BOINC messages.
* If I remember correct it was at the beginning 0.18, or I had something in my eyes? ;-) Or it's possible that with all stock this can vary?


Then I took Raistmer's V7 mod and had then 2 x '0.04 CPUs, 1 CUDA'.


Then I deleted this entries in app_info.xml:
<plan_class>cuda</plan_class>
<avg_ncpus>0.040000</avg_ncpus>
<max_ncpus>0.040000</max_ncpus>

Since then I have 2 x '1.00 CPUs, 1 CUDA'.



BOINC is little bit 'different'.. so CPUs mean mostly Core of CPU.
So what mean this 0.04, 0.19 or 1.00 CPUs?

0.04 mean 4 % of the Core or the whole CPU?


More infos about my rig:
setiathome.berkeley.edu/forum_thread

---

I don't know what any of that means...I hope I'm not alone in that. LOL

I'm feeling more productive right now as I'm running 3 Astropulse WUs and a regular SETI WU.

My RAC isn't rocketing up there though, so I'm hoping the powers that be are aware.

My post was not only for you.. ;-D

I mean if you look to the BOINC Manager, you see which and how much WUs you are current crunching.
And at the GPU-WU you see something like:
'0.xx CPUs, 1 CUDA'

:-)

---

0.03 CPU apparently means that the CPU runs .03 of the time to feed the CUDA device. You can modify some XML to make BOINC think the computer has one more than the actual CPU's so will run 4 normal tasks and the CUDA's (Thanks for that code OzzFan).

The trick is that you have to maintain both CUDA and non-CUDA tasks in your queue. This can be done by manually controlling downloads using 'no new tasks' and changing the setting at SETI Website to allow CUDA tasks every other download (as necessary to maintain a balance between the two types of tasks). Never on a Tuesday, though.

Perhaps someday this could be programmed to happen automatically. (Please.)

0.03 CPU apparently means that the CPU runs .03 of the time to feed the CUDA device.

0.03 CPU means it is using 3% of the CPU for one CUDA task. If you have two GPUs, it'll be 0.06 = 6%. (or conversely, 0.04 = 4% and 0.08 = 8%)

You can modify some XML to make BOINC think the computer has one more than the actual CPU's so will run 4 normal tasks and the CUDA's

The modified cc_config.xml file is only needed for BOINC 6.4.5; this behaviour is already fixed in any version above it.

---

Where might I find that version of BOINC higher than 6.4.5?

Thanks.

SETI@home MAIN page, under download, you will find a list off BOINC versions.

---

SETI@home MAIN page, under download, you will find a list off BOINC versions.

If you click the All versions link, yes. Else you'll end up with the normal detection page you always end up with.

---

...
0.03 CPU means it is using 3% of the CPU for one CUDA task. If you have two GPUs, it'll be 0.06 = 6%. (or conversely, 0.04 = 4% and 0.08 = 8%)
...

Are you sure?

I ask, because in the SETI@home preferences you can see for example:
On multiprocessors, use at most 4 processors [for Quad-CPU]


So this '0.04 CPUs, 1 CUDA' mean really CPU or Core?


So if it mean CPUs.. then with deleting of the specially entry [in upper post] in Raistmer's app_info.xml I let run my two GPUs with support from the whole CPU.

I have '1.00 CPUs, 1 CUDA'. [two times, because I have two GPUs]


Everything stock I have two times '0.19 CPUs, 1 CUDA'.


If it means CPU.. then I need to insert in the app_info.xml '0.25 CPUs, 1 CUDA'.. for one whole Core..

<plan_class>cuda</plan_class>
<avg_ncpus>0.250000</avg_ncpus>
<max_ncpus>0.250000</max_ncpus>

EDIT:
BTW.
I have per GPU.. ~ 7 % CPU- or ~ 28 % Core- support the whole time.
At the first ~ 25 sec. of every WU I have 25 % CPU or 100 %* Core per one GPU.
[* but mostly jumping over the Cores or divided to all Core.]

EDIT #2:
Is BOINC 'intelligent' enough to follow this entry?

---