Hey Guys,
I know this has been asked before, but could not find the thread. I started running dual 460's. In the manager only two WU's are running, but two are waiting, and six more are waiting. What is the corrrect setup to run two WU's per GPU?

Thanks in advance

---

In your app_info change every instance of <count1count> to .5 (or .33 if you want to try three at a time.) If you are running 64 bit there will be 6 instances, 32 bit will only be 3.

---

PROUD MEMBER OF Team Starfire World BOINC

I have change the setting <count1count> to .5, but The app is only using one of my GPU's the other is idle. I am running Win7 64-bit, if that makes a difference.

Chris

---

Hey Guys,
I know this has been asked before, but could not find the thread. I started running dual 460's. In the manager only two WU's are running, but two are waiting, and six more are waiting. What is the corrrect setup to run two WU's per GPU?

Thanks in advance

Is BOINC recognizing both cards? According to your computers list on here, the machine with a 460 is only showing 1 card. Post the first 20-30 lines or so of your startup log.

---

also, your details page is not displaying two 460's.

Do you have "<use_all_gpus>1</use_all_gpus>" in your cc_config file under options?

Like so:

<cc_config>
<options>

<use_all_gpus>1</use_all_gpus>

</options>
</cc_config>

Lt


edit: Blake posted while I was still pecking away...(too slow again...sigh...)

He shouldn't need use_all_gpus with 2 identical cards.

---

Maybe not, but BOINC doesn't appear to be recognizing both of them according to his details page.

Lt

He shouldn't need use_all_gpus with 2 identical cards.

Correct. I have 4 GTX460's in one machine, 3 being ASUS cards and the 4th being a PNY Factory OC'd card. I do not need to use that setting on mine.

Maybe not, but BOINC doesn't appear to be recognizing both of them according to his details page.

Lt

This. Something is keeping BOINC from using and reporting the card.

---

Also, even within recently reported results only 1 device is displayed (at which point it doesn't even care what Boinc 'thinks', as it's just Cuda device enumeration)

setiathome_CUDA: Found 1 CUDA device(s):
Device 1: GeForce GTX 460, 1024 MiB, regsPerBlock 32768
computeCap 2.1, multiProcs 7
clockRate = 1526000
In cudaAcc_initializeDevice(): Boinc passed DevPref 1
setiathome_CUDA: CUDA Device 1 specified, checking...
Device 1: GeForce GTX 460 is okay

This probably means the other card is AWOL at a lower level that Boinc, such as power or [nVidia] control panel settings [For Cuda devices) etc.

Jason

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

There are several tasks that DO show 2 cards, however.

http://setiathome.berkeley.edu/result.php?resultid=2296968228

Sooo, it did work at one point in time. Something changed.

---

Good spot. As far as Boinc's end of things goes, they are not identical. It will choose the higher clockrate one unless the use_all_gpus cc_config.xml option is specified. That doesn't explain why it's gone (from driver enumeration) though. I suspect it may have simply gone to sleep, or similar.

Jason

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Good spot. As far as Boinc's end of things goes, they are not identical. It will choose the higher clockrate one unless the use_all_gpus cc_config.xml option is specified. That doesn't explain why it's gone (from driver enumeration) though. I suspect it may have simply gone to sleep, or similar.

Jason

I respectfully disagree sir! My machine has a PNY card in the top slow factory clocked at 765mhz, while the others are NVidia spec at 675mhz. Boinc uses them all with no cc_config options :)

---

Yes, the cards are both GTX 460's. but one is an OC card, the other is just a regular card. anyway to get both to run WU's at the same time?

Chris

---

Everything is fine. I just needed to restart the app.

---

Good spot. As far as Boinc's end of things goes, they are not identical. It will choose the higher clockrate one unless the use_all_gpus cc_config.xml option is specified. That doesn't explain why it's gone (from driver enumeration) though. I suspect it may have simply gone to sleep, or similar.

Jason

I respectfully disagree sir! My machine has a PNY card in the top slow factory clocked at 765mhz, while the others are NVidia spec at 675mhz. Boinc uses them all with no cc_config options :)

BOINC doesn't look at the clock speeds directly when comparing cards. However, it does do a strict comparison of the assessed peak_flops rating of the cards, and should ignore the slower card(s). What are the peak_flops values of the different cards displayed as in your startup messages?

Ah - edit: there's a 'return 0' on line 769 of coproc_detect.cpp. Intentionally or not (I suspect not, going by the comments in lines 750-755), I don't think the speed check is working.

Good spot. As far as Boinc's end of things goes, they are not identical. It will choose the higher clockrate one unless the use_all_gpus cc_config.xml option is specified. That doesn't explain why it's gone (from driver enumeration) though. I suspect it may have simply gone to sleep, or similar.

Jason

I respectfully disagree sir! My machine has a PNY card in the top slow factory clocked at 765mhz, while the others are NVidia spec at 675mhz. Boinc uses them all with no cc_config options :)

BOINC doesn't look at the clock speeds directly when comparing cards. However, it does do a strict comparison of the assessed peak_flops rating of the cards, and should ignore the slower card(s). What are the peak_flops values of the different cards displayed as in your startup messages?

Ah - edit: there's a 'return 0' on line 769 of coproc_detect.cpp. Intentionally or not (I suspect not, going by the comments in lines 750-755), I don't think the speed check is working.

It must not be, because the OC'd card does report a higher GFlops than the others

---

Bah. We're all wrong, it's intentional.

757 // If "loose", ignore FLOPS and tolerate small memory diff 

BOINC uses a 'tight' comparison to find absolutely the best card. Then it uses 'loose' comparisons to decide whether to use other, similar, cards. Provided all the cards in the system have the same Compute_Capability (and cuda_version - but I think that comes with the driver, and since all the cards will be using the same driver, I don't think that comes into play), all will be used, irrespective of speed.

LoL, Boinc operation aside, still doesn't explain why the card's gone AWOL at driver level according to the app. (I guess Chris must've fixed it with a restart.. hope that doesn't happen again, that's downright strange ;) )

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

LoL, Boinc operation aside, still doesn't explain why the card's gone AWOL at driver level according to the app. (I guess Chris must've fixed it with a restart.. hope that doesn't happen again, that's downright strange ;) )

So, I was right the first time, he just needed to restart?

---

PROUD MEMBER OF Team Starfire World BOINC