Hello,

Yesterday I have joined GPUGRID.
I have an i920 CPU with 4 (8 virtual) cores and have set to use only 6 of them.
The graphics card is a GTX285.
A message I have got from GPUGRID in the BOINC manager is:
26/06/2009 20:30:47|GPUGRID|Message from server: (reached per-CPU limit of 1 tasks)

However I set it to use 6 on this particular pc.
How is this happend and how can I change it? Does anyone know where these settings have to be made?

This is the BOINC manager message:
26/06/2009 20:28:57||Preferences limit # CPUs to 6

Thanks for your help.

Greetings from TJ

GPU-Grid is called like that because it runs on the GPU ;)
You have one of them, so one WU can be processed at any time. The message you are quoting is related to the maximum number of WUs which you're allowed to queue. Due to limitations in the BOINC software this is currently tied to 1 WU per cpu core and is completely unrelated to processing the WUs. Does this answer your question?

MrS

---

Scanning for our furry friends since Jan 2002

Yes it does, thank you MrS.

That also explanes why the CPU is doing almost nothing.

---

Greetings from TJ

Well I thought it answers the question but I still have one.

The GTX 285 has 240 stream processors. BOINC manager says:
GPUGRID Running (0.26 CPUs, 1 CUDA).
Do I understand correctly as I say it is using about 1/4 of the CPU and all the GPUs of the graphics card?
A 'simple' (not expensive Pentium 4 processor) would be okay then to use with a GTX card?

I have read a lot about CUDA, CPUs and nVidia GPUs. But English is not my native language and it is very likley that I did not understand all very well.

A TESLA card has 960 processors but than will be processed one WU at one time as well? These cards are very expensive and out of reach for me. But as they run one WU at a time they are 3 times faster I think?

Can someone explain a bit for me to understand better?
Thanks again.

---

Greetings from TJ

The GT200 chips (which the GTX 285 uses) have 240 stream processors, but BOINC does not care about such low level details. For BOINC one CUDA device is what the driver reports as one. That's why cards like the GTX 295 with 2 chip can only use one of them if SLI is enabled. In SLI mode the driver reports both chips as one device and distributes the work between both chips behind the scenes. For games this works well, but for GPU-Grid you have to disable SLI to use both of them and run 2 WUs in parallel. The Tesla you mentioned is basically 4 of these chips with more memory, so you'd run 4 WUs in parallel on them. In the end you'd get the same as running 2 GTX 295 (which cost way less).

The 0.26 cpu is just a meaningless guideline for BOINC. If you take a look at the task manager you'll likely find that GPU-Grid uses even less cpu power under windows. So, yes, any slower cpu which can be plugged into a board with PCIe slots will keep a GPU busy. However, I wouldn't use a P4 as these are horribly energy inefficient. There are muc better CPUs available which also cost ~30$.

MrS

---

Scanning for our furry friends since Jan 2002