Zdenek, It might be a driver issue. Others have reported similar problems with 319.x on Linux.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Zdenek, It might be a driver issue. Others have reported similar problems with 319.x on Linux.

Yes, you are right. Moved back to 310 and all is ok.

I have problem with my own distrrtgen app also. IMHO It stucks on synchronizing between CPU and GPU in cudaDeviceSynchronize().

I think there has generally been issues with this since about CUDA 4.2 dev.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

I have found that 6xx and Titan have problems. 5xx looks ok.

Two more NOELIA failures yesterday: one after 4 seconds, the other after 20,927 seconds. I will continue with 'short' tasks.


6895000 4475926 25 May 2013 | 16:07:19 UTC 25 May 2013 | 16:18:17 UTC Error while computing 4.10 3.21 --- Long runs (8-12 hours on fastest card) v6.18 (cuda42)
6894999 4475925 25 May 2013 | 16:07:19 UTC 25 May 2013 | 22:08:10 UTC Error while computing 20,927.20 9,396.42 --- Long runs (8-12 hours on fastest card) v6.18 (cuda42)
6894967 4475903 25 May 2013 | 16:18:17 UTC 25 May 2013 | 16:21:06 UTC Aborted by user 0.00 0.00 --- Long runs (8-12 hours on fastest card) v6.18 (cuda42)

Hi

Just completed what looks like a brand new Noelia

http://www.gpugrid.net/result.php?resultid=6992175

87k sec on a gtx 650 ti, poor gpu utilisation despite a reboot half way thinking there was a problem :(

Hope its a one off?

Just completed what looks like a brand new Noelia

http://www.gpugrid.net/result.php?resultid=6992175

87k sec on a gtx 650 ti, poor gpu utilisation despite a reboot half way thinking there was a problem :(

Hope its a one off?

I got one of these too. The first guy aborted it and it took my OCed 650 Ti well over 24 hours (92,919.91 seconds) to run it in Win7-64 (vs yours in XP). GPU utilization was OK, but these are TOO LONG and to add insult to injury only give out about 1/2 the credits they should.

http://www.gpugrid.net/workunit.php?wuid=4527468

I had one, too - http://www.gpugrid.net/result.php?resultid=6992593

I aborted it when I noticed it at 16+ hours, 54% completed.
The GPU load was at 95%, but Mem Controller load was only 4% on a 660ti.

Hmm, Does not sound promising - don't suppose anyone has noticed what the gpu mem utilisation is?
Was wondering if it went over 1gb as seen a 680 complete one in a third of the time.
Agree these are too long, in my opinion they should be in a separate 'bucket' with a clear min hardware spec requirement; The long and short descriptions are far too vague.

There's one thing I'd like to say though, Nathan sure did a bang-up job on those NATHAN_KIDKIXc22's, I'm getting 98% GPU load and 35-38% memory controller utilization on my GTX680's. This buds for you, Nathan! You should have named them KIDKIX_BUTTc22, you really should give a clinic for the fellow researchers (I'm sure they'll get it sorted, not complaining).

Hmm, Does not sound promising - don't suppose anyone has noticed what the gpu mem utilisation is?

Mine was around 1045 MB.

Wow does that mean this units dont run on 1GB VRAM Hardware? (didnt tried)

---

DSKAG Austria Research Team: http://www.research.dskag.at

Thanks, at least we have a plausible explanation, not so sure ruling out the mainstream will be good for GpuGrid. Pity we can't isolate WUs as I can think of an addition to Flashawks naming convention - but lets not go there.

Are you asking about GPU utilization or the size of the work units? I take it that these wu's are from the short queue and if the work unit size is larger than the amount of GDDR memory on the video card, that would not only cause a massive slow down in crunching times it will also make your computer almost unresponsive (mouse, keyboard and such).

Beyond knows what I'm talking about and if that were the case, I'm sure he would have mentioned it. I am confused by petebe's response, is he talking about the work unit size? I haven't done any short queue tasks in sometime and I do know that Noelia's work units are setup differently than Nathans and her wu's typically have a lower CPU and GPU utilization.

That's what I understood too. All scientists are working on different projects/amino acids and use different algorithms. Thus WU's differ. The latest one from Nathan seems almost optimal as far we can see in error-free and rather fast cycles on the fastest cards.
But I also like to mention that I had very little problems with Noelia's WU's as well, only one beta failed and one because Windows though to update itself (this is now not longer possible).

---

Greetings from TJ

Flashhawk, I was referring to the "Memory Used" figure as reported by GPU-Z.
Memory Used was 1045 mb and Memory Controller Load was 4%.
In contrast, NATHANs usually run around 250-450 mb Memory Used and 35% Mem Controller Load.

I don't know how this relates to a WU size - sorry if this was confusing.

This particular 660ti does GPUGrid crunching only - it's not connected to a monitor. One HT CPU reserved per GPU.

Are you asking about GPU utilization or the size of the work units? I take it that these wu's are from the short queue and if the work unit size is larger than the amount of GDDR memory on the video card, that would not only cause a massive slow down in crunching times it will also make your computer almost unresponsive (mouse, keyboard and such).

Beyond knows what I'm talking about and if that were the case, I'm sure he would have mentioned it. I am confused by petebe's response, is he talking about the work unit size? I haven't done any short queue tasks in sometime and I do know that Noelia's work units are setup differently than Nathans and her wu's typically have a lower CPU and GPU utilization.

They're long queue WUs yet they credit like the short queue. If I see any more I'll make like a Dalek: EXTERMINATE, EXTERMINATE!!!

BTW, like you mentioned: kudos to Nathan on the new KIX WUs. Nathan, give the other WU generators a class in WU design. Please?

NOELIA_Mg WU are Long runs. Most of Noelia's work has used >1GB GDDR and taken longer than other work.
207850 credits would be about right in my opinion (including the 50% bonus).

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

NOELIA_Mg WU are Long runs. Most of Noelia's work has used >1GB GDDR and taken longer than other work.
207850 credits would be about right in my opinion (including the 50% bonus).

The ones listed above scored just 69,875. Including only 25% bonus though since they're SO LONG :-(

Flashhawk, I was referring to the "Memory Used" figure as reported by GPU-Z.
Memory Used was 1045 mb and Memory Controller Load was 4%.
In contrast, NATHANs usually run around 250-450 mb Memory Used and 35% Mem Controller Load.

I don't know how this relates to a WU size - sorry if this was confusing.

This particular 660ti does GPUGrid crunching only - it's not connected to a monitor. One HT CPU reserved per GPU.

No petebe, I wasn't confused by anything on your part, I was confused because more people aren't complaining about unresponsive computers. If someone is using an older card with only 1GB of onboard GDDR, then the system RAM or swap file would be used slowing computers to a crawl.

No, you're fine buddy, sorry for the confusion, I should have been a little clearer. Frankly, I'm shocked I haven't seen more of this in the forum, that's a huge wu for not much credit. I guess I'll have to turn on the short queue and check them out, that is odd they aren't in the long queue.

Edit: I understand now (I'm pretty slow sometimes), there coming through the long queue, I haven't seen one yet.