We are uploading a version with cuda2.3.
If you have new drivers you should receive it instead of the old one. It seems to fix some problems with older cards.

gdf

Let me know if you suddenly start to have problems.

gdf

started about 4 hours ago. it crashes my computer randomly and i reached my daily quota. i am using drivers 195.62, win7 64 bit, gtx 260 192.

Just had a TONI-HERG fail after 20Ksec, using the new cuda23 app - host 43404 (ignore the run of 10-second errors, a reboot cured those).

Let's hope the replacement GIANNI-BIND runs better.

Your clock seems way to high.

gdf

Your clock seems way too high.

gdf

It's a Zotac 'AMP' 9800GTX+ at factory settings. Any visible overclock was done at the factory (and given a 5-year warranty), not by me.

What's your opinion of host 45218 - also Zotac, a 9800GT, but not an 'AMP' overclocked edition?

This is the sort of host which should start to crunch without problems when the cuda23 application is used.

gdf

This is the sort of host which should start to crunch without problems when the cuda23 application is used.

gdf

Please have a look at what 45218 did to an IBUCH_TRYP last night - using the cuda23 app.

This is due to another bug of Nvidia drivers on older cards for which we have sent a reproducer but they are never going to fix.

gdf

I never fully understood why David Anderson created separate sample BOINC code for <plan_class>cuda</plan_class> and <plan_class>cuda23</plan_class> (when I asked, that's what he said it was - sample code).

I proposed on the boinc_alpha mailing list (05 August, "CUDA version control") a more general approach to concurrently handle drivers elderly, current, and yet to be developed: they didn't take it up. But I still feel it was arbitrary to single out 2.3 for special treatment.

But from what you're saying, it would make sense to develop <plan_class>cuda_G92</plan_class>, <plan_class>cuda_G200</plan_class>. How about it?

PS - the overclocked one is doing just nicely on a GIANNI_BIND - passed 85%, famous last words.

Something that works for G92 usually works for G200 cores. So, it is not that useful. The problem is that pace of development is so high that Nvidia does not keep up with it. G92 cores are two generations old now and probably they receive zero support at nvidia.

Some older drivers were working just fine with older cards, then they modified something and here is the problem. There is no single application that works with all drivers on all cards. We will need to provide separate apps and let the user try which one is best for him depending on the driver he installs.

gdf

The overclocked one has now completed and validated task 1578512. Perhaps in future initial responses you could distinguish between "overclocked" and "insanely overclocked"? :-)

That's bad news about the drivers. Because I test extensively at SETI, I was one of the very first to upgrade to 190.38 after they received WHQL release: I've been running them since 22 July. It feels as if the increased error rate started much later than that.

I'm not prepared to roll back the drivers before 190.38, because the 2.3 runtime+fft makes a dramatic improvement to SETI runtimes. I would be happy to upgrade to a later driver if one can be found which suits both projects on these cards.

Failing that, you seem to be implying that there is nothing to be done to improve the GPUGrid success rate of three cards which all still have over four years' warrantied life left in them. In that case, the best policy would appear to be to shoot on sight (via the abort button) any tasks from IBUCH_TRYP or OTTO_HERG. My apologies in advance to the scientists responsible for those two research projects: but they will actually get their results more quickly from an early abort/resend, rather than waiting for 12 hours cache and 6 hours wasted crunching before erroring and triggering the same resend.

The solution would be to remove all together the CUDA FFT and use another one.
Something that we are seriously thinking.

gdf

My humble GTS 250's seem to be working fine with the old driver (285.85?) so should I bother updating driver?

Don't wanna fix what isn't broken.

Any overwhelming reason to upgrade?

---

Join Here
Team Forums

Failing that, you seem to be implying that there is nothing to be done to improve the GPUGrid success rate of three cards which all still have over four years' warrantied life left in them. In that case, the best policy would appear to be to shoot on sight (via the abort button) any tasks from IBUCH_TRYP or OTTO_HERG. My apologies in advance to the scientists responsible for those two research projects: but they will actually get their results more quickly from an early abort/resend, rather than waiting for 12 hours cache and 6 hours wasted crunching before erroring and triggering the same resend.

Don't forget the TONI_HERG WUs...

Don't need to change anything if it is working.

[By the way I had a Dalmatian as well]

gdf

Don't forget the TONI_HERG WUs...

Thank you. Just aborted two of them in a row.

The solution would be to remove all together the CUDA FFT and use another one.
Something that we are seriously thinking.

gdf

What about the opposite, that is try the cuda 3.0 FFT libraries? Does that improve things (I assume you've tried this in-house)?

---

BOINC blog

We are uploading a version with cuda2.3.
If you have new drivers you should receive it instead of the old one. It seems to fix some problems with older cards.

gdf

1. I'm running ubuntu 9.10. Will I get new app?
2. In terms of speed - does is run faster?

right now I'm using 185.18.35 and CUDA 2.2

---

Hello to all Cruncher.

I have been for some time, the problem that my Quad PC no longer works with the CPU. Previously it was but. There were still working +3 CPDN Wu's on.

AuthenticAMD
AMD Athlon (tm) 9950 Quad-Core Processor [AMD64 Family 16 Model 2 Stepping 3]
(4 processors) [6] NVIDIA GeForce GTX 295 (895MB) driver: 19562 Microsoft Windows XP
Professional x64 Edition Service Pack 2, (05.02.3790.00)

My Preferences - here I can not take OFF CPU

Use Graphics Processing Unit (GPU), if available yes
Use Central Processing Unit (CPU) yes

After save is currently ON!

I like only work with the cpu - CPDN
And with the GPU - GPUGRID!

what's the problem?

Thanks for your help
RELCLAG