Have installed CUDA and the latest Boinc sw into 4pc both 32bit and 64bit Windows. Mobos is Gigabyte X-45 & X38, and one MS-7365 all with quad CPU (all OC).

All is equipped with "Gigabyte 9500GT OverClock 512Mb". Installed the included driver and upgraded to v.180.48 in 2 of the PC and the other two I installed v180.40 directly (after removed in safe mode).

Installed Boinc 6.4.5 on top of previously installation (2cPC) and also clean BOINC installation in new folder (the other 2). BOINC finds CUDA device ok.
When running I got "Running (0.03 CPUs, 1 CUDA)" on all and one PC halts after 17sec (always), the other 53sec (always) the third indicates halt after 43 or 52 sec but is still running (counting down %).
None of the PC indicates any calculations on the 4 cores, just on GPU.

I am running Gigabyte Gamer Hud Lite (tried both default and OC), CUDA-Z 0.4.74 and EVGA Precision v.1.4.0.

GPU temp 44-46C
GPU-Z indicates multiprocessor=4
GPU Single precision 127KMflop/s unloaded (approx half when GPU-crunching)

Tried double 9500GT (no SLi) on one PC and BOINC found 2 CUDA but only used one (according to CUDA-Z)

Reinstalled AK_v8.0_Win32_SSSE3x and now all 4 cores are running again. Game Hud indicates 100% GPU activity (goes down to 0 when running bench) but CUDA-Z indicates same as unloaded so I suppose GPU is not crunching.

Believe I read (?) somewhere that GPU and CPU can not crunch same thing (WU or AP) - is that correct? Could that be my problem that only GPU was crunching and not CPU?

If graphicscard has 4 GPU (due to GPU-Z) why just crunching one?

Must the SLi connector be on to work both cards?

So to my very question :) How do I get 4 cores and all GPU's working?

Merry X-mas!

(Sorry for the long story...)

/Peter

---

//Peter

Hi, you can't run AK_v8 and CUDA at the same time.

You can however do CUDA and AP at the same time by going to this thread and downloading the modified CUDA MB & AP app. You'll also need this info to get all the cores and 1 CUDA running at the same time.

Edit: I also read over at Beta that you should be able to get 2 gpu tasks going with two cards. The posts I saw over there didn't go into how they got them installed, other than loading the drivers for the cards, but there may be someone else that has more info on that. here's the link to the thread on it at Beta.

---

Believe I read (?) somewhere that GPU and CPU can not crunch same thing (WU or AP) - is that correct? Could that be my problem that only GPU was crunching and not CPU?

http://setiathome.berkeley.edu/cuda_faq.php

Q) Does SETI@home run GPU and CPU versions simultaneously?
No. If BOINC determines your CPU is capable of running the CUDA version, only the CUDA version of SETI@home will run. One copy will run on each GPU you have installed. If you want to keep your CPUs occupied at the same time, you can join another BOINC project.

The BOINC developers expect the capability to run CPU and GPU jobs simultaneously for the same application will be added in the next couple months.

If graphicscard has 4 GPU (due to GPU-Z) why just crunching one?

How many tasks did you have at the time available for CUDA?

Must the SLi connector be on to work both cards?

As far as I know it doesn't have to be on.

So to my very question :) How do I get 4 cores and all GPU's working?

See above. Not for some time yet if you want to crunch Seti only.
Else you will have to attach to another project. Then 3 of your CPUs will be used on that project and one CPU + the GPU will be used on Seti. The GPU needs to use part of that one CPU at all times.

---

Thanks Byron and Ageless

Started up with 2pcs 9500GT (no SLi) in one X-45 mobo with Q6600 @ 3,6GHz as before. Copied in modified-seti-mb-cuda-opt-ap-package-for-full-gpu-utilize.0

I got approx 30 WU's at 12:53 (GMT+1) and all was estimated 1:46.
With 2 graphics card there was 2 WU started with msg "Running (0.04 CPUs, 1 CUDA)" each. (Later I removed 1 card and only 1 WU was active.) First bunch of WU's had finished almost exactly 2 hours after.

All WU CPU-time freezed differently at 17sec up to 1,52, some WU still running (increasing %) and some freezing (increasing TimeToFinish) wich I had to abort.

First 2 WU run through on 17sec, next 2 got error after 17sec(!) and then some finished very quickly on approx 8-20sec.
The first WU wich passed without error was:
03dc08ai.3646.19295.12.8.175_0
03dc08ai.3646.19295.12.8.178_1
each took 12min to complete (estimated 1:46 at start).
As you can see approx 50% was in error.

GPU temp rised from 46 to 54C (default speed whole test)

In GamerHud I noticed that GPU utilisation was 7-20% (thought it should be higher) and all 4 CPU-cores was 15-25%.

At last I created cc_config.xml and rebooted but still just running 2 CUDA and no CPU-core.

Now I have received a new set of WU's and till now none seems to have errored out. "CPU-time" has freezed on all from 18 to 52 sec as before.

I'm leaving for the weekend so I because of all errors I just restore AK_v8.0_Win32_SSSE3x and start CUDA up again on monday.

Have a nice weekend!

/Peter

---

//Peter

Hi,

I've looked at some of the tasks at the time you sent your last message, such as this one and it seems like it was doing good.

All WU CPU-time freezed differently at 17sec up to 1,52, some WU still running (increasing %) and some freezing (increasing TimeToFinish) wich I had to abort.

This is actually normal for the cpu time not to move very much. Your not using the cpu as much to do the CUDA. As long as the progress % is going up you're in good shape. The % to completion does go up in the begining of the task but will start to go down after a while. That's normal too.

When you get back we can try again if you like, but it seems you're on the right track. Both cards are being seen by Boinc as you can see from task example so hopfully we can get both gpu doing tasks and all the cores too.

---

Hmm it appears to me that BOINC is utilizing my GPU and CPU cores at the same time. There's 5 processes, of which one (and the fastest one :P) is used by CUDA, and four others presumably by my CPU.