Earlier in this thread Zoltan said he had problems on one of his systems (GTX670's I think). Moving from 320 to 307.9 (on XP) seems to have resolved the issues.

There was/is several different problems. One problem was GDDR limitations on some cards. I still had problem on a 2GB GTX660Ti and a 2GB GTX660, as have others, but I didn't have problems with a GTX650TiBoost on Linux.

I've had fewer errors on 314 than 320, but none on 304.88. Hardly conclusive, and may only reduce some problems but generally it appears that earlier drivers are better.

---

FAQ's

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

I'm now about a month experimenting with my new GTX660 and to me it also depends on the set up of the system. In some systems (could be low PSU, wrong MOBO or wrong MOBO settings) it did not great an another they do.
I got a GTX770 yesterday and that is now highly under-performing in an i7. That could not be the card. I will find that out to swap it in another system.
I still believe that the GTX660 is good for this project. There a few things going here with the WU's so currently not a good period the value or de-value a GPU.

---

Greetings from TJ

Looked at from a purely PPD point of view on my GTX 660s, I just got 20,500 points from a I654-SANTI_baxbim1 in the Short que, which took 3 hours 33 minutes, or about 139k PPD. In contrast, the last Noelia Long to complete was a 2-NOELIA_2HRUN which yielded 112,500.00 points in just over 24 hours. So with this card I may be slightly better off in the Shorts, though there will be some Nathans in the Long que that would probably complete without incident in the usual times.

However, the main point is that you could get hung up even longer with the new work units; there is presently a 44x1-NOELIA_1MG that has been running for 28 1/2 hours and is at 87%, or about 33 hours to complete, assuming that it completes at all. I think that if GPUGrid wants the support of the mid-range cards for the Long que, they will need to adjust the work units accordingly.

But they may be happy enough with the higher-end cards for the Longs, in order to get new science done, and this is one way of achieving it. So it is not clear if it is a bug or a feature; they haven't told us one way or the other.

Earlier in this thread Zoltan said he had problems on one of his systems (GTX670's I think). Moving from 320 to 307.9 (on XP) seems to have resolved the issues.

It did. I'm crunching NOELIA tasks error free.
I have 4 active hosts at the moment, every one has different drivers and OS.
1. WinXPx64, v310.33, 2xGTX680: 2 errors (in short time)
2. WinXPx64, v307.90, 2xGTX670: 11 errors due to the previous driver, and the experiments with the v320.49
3. WinXPx64, v314.22, GTX680+GTX670: 1 error: NATHAN_KIDKIXc22 :))
4. WinXPx86, v314.07, GTX680: 0 errors
Not active host:
5. Win7x64, v311.06, GTX480: 6 errors due to low GPU voltage (1000mV)
BTW: it seems that the long queue nearly run out of NOELIA workunits, as my hosts have 8 NATHAN's and 5 NOELIA's in their queue.

Reporting back on the NOELIA_xMG_RUN I was crunching on my 1GB 650Ti: it took long but went fine! Here is the WU info.

Since I'm running on Linux, I can't tell how GPU and MC load was, all I know is that its temperature was normal (~62C). Right from the start, estimated time was at ~44h and the unit completed in 42.5h, about 1.5h earlier. So, this tells me that I didn't hit the "MC load at 1%" issue. Or, if I did, I did right from the very start! The WU consumed 622MB on the GPU.

137.5K of credit for 2 days of crunching is kind of lame, but I'm glad I can crunch these beasts! :D

I even stopped / started BOINC and suspended / resumed the WU a couple of times without issue :O Well, the joys of Linux, I guess!

---

I even stopped / started BOINC and suspended / resumed the WU a couple of times without issue :O Well, the joys of Linux, I guess!

No, no I did that too in Vista x86 and it worked as well without failing the WU. Even booting the system. I tried everything I knew to got more than the 1% MCU load but nothing helped. I guess you had not the 1% load, because when I had it, the estimated time to finish was wrong every time. it was several times updated by BOINC, though wrong. Happy crunching![/quote]

---

Greetings from TJ

Thanks Zoltan, that helps paint a picture.

Vagelis Giannadakis, 42.5h for a GTX650Ti (1023) is likely down to the GDDR limitations. Typical WU's for that sort of credit and on that card take just under 24h.

---

FAQ's

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

I had 4 of these units fail simultaneously, on my windows 7 computer. I did a routine reboot, without first suspending the units, and they all crashed.

http://www.gpugrid.net/result.php?resultid=7060754

http://www.gpugrid.net/result.php?resultid=7057895

http://www.gpugrid.net/result.php?resultid=7057887

http://www.gpugrid.net/result.php?resultid=7057605


This doesn't happen with xp, but on windows 7, I guess you have to suspend the units before rebooting. This is a major issue!

alax117-NOELIA_UBQ1-0-1-RND6675_0: 16.3h on GTX 650Ti / Linux, 142.5K credit! Mmm, yummy!! :D

---

I have got a new type of Noelia: leux12-NOELIA_UBQ1-0-1-RND9216, it took almost 83000 seconds to complete. I saw its MCU load was only 15%.
I have now a new one running: prox80-NOELIA_UBQ1-0-1-RND00515_0, expected run time: 23 hours. This is better than the 1% load but still not great for fast return times. GPU load is around 85% and temperature is low: 53°C.

---

Greetings from TJ

You have got to be kidding me...

12x6-NOELIA_2HRUN-3-5-RND4863_0 Completed and validated 86,791.58 2,841.42 112,500.00

That's 24.11 hours run time. Not even zero cache would have helped... Grrrrr. :-(

---

Well thanks, than its not my hardware :)

---

Greetings from TJ

24.11h on a GTX570 does sound bad.
What else were you running, did the MCU drop to 1% and did the GPU downclock?

---

FAQ's

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

Do the 2hrun units had the same issue about gpu ram? Because his win7 desktop needs more of that memory while my xp with 570's (or perhAps the desktopless and vram empty second card) needed "only" between 60-62k secs.

---

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

24.11h on a GTX570 does sound bad.
What else were you running, did the MCU drop to 1% and did the GPU downclock?

Right now, I'm running CPDN, POGS, WCG, Einstein CPU tasks, nothing unusual that I haven't run before alongside GPUGrid. I didn't note any MCU reading.

I did notice a little while ago that the NATHAN_KIDKIX that's been running since the lengthy NOELIA_2HRUN completed was taking unusually long and the GPU load was maybe a little low. I rebooted and it seems to be back on track to finish in 15 hours.

Also, I see that I completed another NOELIA_2HRUN several days ago in a time that I would have expected (61.7 Ksec).

So, it seems I probably was a victim of downclocking. Must remember to check that! :D Thanks!

---

If BOINC suspends the Noelia (for whatever reason) you may get a driver crash, which may lead the card in some strange state. I've seen memory downclocking (which actually increase Memory controller load, so quite the opposite of what SK and others are seeing) and just downclocking the chip. In the 1st case it's enough to set proper clocks again, in the 2nd only a reboot helps.

MrS

---

Scanning for our furry friends since Jan 2002

I saw just that the clock was down to 50% and MCU load of 17% doing a Santi LR.
Checking my tasks showed a Santi LR had errorred out before. That should than have result in the clock to down clock as ETA says.
A reboot did help. This is not nice, if I am home and can not reach the rigs at my office at work or vice verse.
And I had an very long running Nathan just like ritterm.
Thus NOELIA's are not the only WU's that behave strange lately.

---

Greetings from TJ

The reason was likely a driver reset triggered by some error happening in the GPU. It should be quite hot now in your attic, isn't it? Maybe GPU clocks of -13 MHz are in order.

MrS

---

Scanning for our furry friends since Jan 2002

Yeah way to hot 33.7°C. I have taken one PC downstairs but my girlfriend is not enjoying the noise. So I guess a bit less crunching in the next days.
I still have my heat problems as you said ;-) If its not the PC then the ambient temperature. Crunchers like winter...

Edit: by the way ETA, I see your RAC is very low for you, also under pressure by the heat wave?

---

Greetings from TJ

Happy you :P we have here 38 degress ^^ tommorow it should get 39 or above :( :( :(

---

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