Advanced search

Message boards : Number crunching : Too much errors with NVIDIA's 200 series and older types.

Author Message
Profile Fred J. Verster
Send message
Joined: 1 Apr 09
Posts: 58
Credit: 35,833,978
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18998 - Posted: 21 Oct 2010 | 21:14:06 UTC

Looking at my results and one thing, that I see all the time, all my wingmen have invalid
results.
I.m.o. a waiste of time, has this been looked into?
It's been going on for some time now.
(Sometimes even an GTX470, running a old CUDA version)
When I tried to sign up for GPUgrid, I was told, that my hardware wasn't useable.(Running a 9800GTX+GTS250)
Changed that for a GTX480 (and trying a 470 along, not working)


____________

Knight Who Says Ni N!

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18999 - Posted: 21 Oct 2010 | 22:55:15 UTC

Actually Fred the GTX 2xx GPUs are the very best for running GPUGRID.

Profile liveonc
Avatar
Send message
Joined: 1 Jan 10
Posts: 292
Credit: 41,567,650
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwat
Message 19000 - Posted: 21 Oct 2010 | 23:04:38 UTC - in response to Message 18999.

Sad but "still" true. At this pace Fermi won't be good until Kepler comes out, & Kepler won't be good until Maxwell comes out. Can't see how this will help NVIDIA sell next generation GPUs when they are launched.
____________

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19009 - Posted: 22 Oct 2010 | 16:42:55 UTC - in response to Message 19000.

Fred, I think your GTX480 is doing quite well, http://www.gpugrid.net/results.php?hostid=78963.

It is averaging about 65K to 70K per day, with only one mid run error. We are all getting some immediate KKi errors (due to the automated task generation system).

Profile robertmiles
Send message
Joined: 16 Apr 09
Posts: 503
Credit: 727,920,933
RAC: 155,858
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19013 - Posted: 22 Oct 2010 | 21:36:05 UTC - in response to Message 18999.

Actually Fred the GTX 2xx GPUs are the very best for running GPUGRID.


But not all of the GTS 2xx series (notice the one letter difference) - some of them were built with older graphics chips that will not run GPUGRID properly. And some, but not all, of the GTX 260 model.


Three of my recent ACEMD2: GPU molecular dynamics v6.05 (cuda) workunits have terminated by telling Windows Vista, but not BOINC, that they are halting. Should I take this is meaning that these workunit types are moving away from being able to run on a 9800 GT?

http://www.gpugrid.net/result.php?resultid=3171660
http://www.gpugrid.net/result.php?resultid=3171447
http://www.gpugrid.net/result.php?resultid=3158571

The system cannot find the path specified. (0x3) - exit code 3 (0x3)

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19014 - Posted: 22 Oct 2010 | 22:37:39 UTC - in response to Message 19013.

Actually Fred the GTX 2xx GPUs are the very best for running GPUGRID.

But not all of the GTS 2xx series (notice the one letter difference) - some of them were built with older graphics chips that will not run GPUGRID properly. And some, but not all, of the GTX 260 model.

None of the GTS series cards run GPUGRID very well. All of the GTX 2xx cards run extremely well except for the early 260 models that used the flawed 65nm chip (260/192/65nm & 260/216/65nm). The GTX 260/216/55nm runs excellently, as does the GTX 275, GTX 285 & GTX 295.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19017 - Posted: 23 Oct 2010 | 9:34:07 UTC - in response to Message 19014.
Last modified: 23 Oct 2010 | 9:39:53 UTC

Actually, the GTS 250 sports a 55nm G92b Fab. However, it is identical to a 9800 GTX+ which makes it immediately derived from the G80 architecture and therefore Compute Capable 1.1.

It should never have been placed anywhere near the GeForce 200 range, but NVidia chose to stick it in there to confuse people into buying it. It worked, but that cheap sales trick which they still employ has mainly served to diminish their credibility. Now people aren’t sure which NVidia card to buy, know they can’t trust NVidias labelling system, so they get an ATI.

Unfortunately people see the 250. Those that can count Know this is higher than 240, so presume it is a better card. They think their presumption is confirmed when they see it has an S, higher clock speeds and more shaders. What they don’t understand is that they are actually looking at a GPU from a previous generation. CC1.1 cards are now grandparents.

When it comes to crunching here there is NO reason to buy one; no matter how many or how fast the shaders are. The fact is at GPUGrid a GT240 does way more work than a GTS 250, is far more reliable, costs a lot less to buy, is smaller, quieter and cheaper to run.

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1576
Credit: 5,606,811,851
RAC: 8,643,637
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19018 - Posted: 23 Oct 2010 | 10:30:45 UTC

Although I agree with all of skgiven's reasons for not buying an old-architecture card for crunching here, people like me who already have the older cards can still make use of them.

I did have some bad experiences recently with the KKi4 tasks (documented in that thread), I'm now having more success with my 9800GT-series cards.

The trick seemed to be some TLC attention to the software on the host:

* installed latest Windows XP security patches
* installed NVidia 260.89 WHQL driver
* reboot

In particular, rebooting after a failed task seems to help the next one.

I haven't touched the hardware in any way.

I still can't get HIVPR_n1 to run on anything less than a Fermi, but the KK14 are now running on 9800GT. The project may have changed the tasks since I hit problems, of course.

Post to thread

Message boards : Number crunching : Too much errors with NVIDIA's 200 series and older types.

//