NVIDIA GPU 0: GeForce GT 430 (driver version unknown, CUDA version 4000, compute capability 2.1, 1024MB, 179 GFLOPS peak)

I couldn't finish the Long Run task (6.13) in time :(

A similar problem has been reported in the following message:
Message 20532

NVIDIA GPU 0: GeForce GT 430 (driver version unknown, CUDA version 4000, compute capability 2.1, 1024MB, 179 GFLOPS peak)

I couldn't finish the Long Run task (6.13) in time :(

Have you changed your preferences to no longer ask for the long WUs?
http://www.gpugrid.net/prefs.php?subset=project

---

Thanks - Steve

Have you changed your preferences to no longer ask for the long WUs?
http://www.gpugrid.net/prefs.php?subset=project

Yes, I have ... but, I was awarded with the following message several times ...

Út 26. duben 2011, 07:31:01 CEST GPUGRID Message from server: No work sent
Út 26. duben 2011, 07:31:01 CEST GPUGRID Message from server: No work is available for ACEMD2: GPU molecular dynamics
Út 26. duben 2011, 07:31:01 CEST GPUGRID Message from server: No work is available for ACEMD beta version

It seems that I just bought a totally useless card :|

Were those from when BOINC was asking for CPU workunits or from when it was asking for GPU workunits? For GPUGRID, expect all requests for CPU workunits to fail with error messages similar to those; requests for GPU workunits should work most of the time.

BOINC normally asks all enabled BOINC projects for one type at some times and the other type at other times.

Don't get a GeForce GT 430, it's not powerful enough to crunch for GPUGrid.

http://www.gpugrid.net/forum_thread.php?id=2427&nowrap=true#20352

http://www.gpugrid.net/forum_thread.php?id=2383&nowrap=true#19886

Actually, it can finish Long run tasks in time, when running nonstop ... but if You compare the performance with the most powerful cards, then Your contribution is less then 10%. The question is, if that's enough for this project.

Does the time limit have to be really so strict?
Can higher delays affect the scheduler negatively?
Who decided to set the timeout to 6 days?
10 days should be sufficient for GT430 owners to finish the tasks in time.

Actually, it can finish Long run tasks in time, when running nonstop ... but if You compare the performance with the most powerful cards, then Your contribution is less then 10%. The question is, if that's enough for this project.

Does the time limit have to be really so strict?
Can higher delays affect the scheduler negatively?
Who decided to set the timeout to 6 days?
10 days should be sufficient for GT430 owners to finish the tasks in time.

OK, the GT430 works but it's not recommended because its so slow.

The time limit does have to be so strict, in fact it could probably do with being more strict.
If you dont return a task within 2 days it gets resent to another cruncher, usually a faster card so it gets returned asap. Sometimes it can take about half a day to send it, but basically if you cant finish a task within 3days the task will be run in duplicate, making your efforts worthless and possibly slowing down the project.
The time limit is 5days, not 6. The Scientists decide.
By 10days your task will have been run by another card, returned and the results used to build new tasks, which in turn would have already been crunched and returned and used to produce more tasks. Time is key.

Stick to the normal length tasks only. If these can finish within 2 or 3days then the card under your setup is doing something positive. If not there is no point crunching with it at GPUGrid.

By 10days your task will have been run by another card, returned and the results used to build new tasks, which in turn would have already been crunched and returned and used to produce more tasks. Time is key.

Ok ... that's the important thing I had no idea about.

If the unfinished tasks block creation of new ones, then slow cards can affect the whole project negatively.

Unfortunately the default settings for this forum lead to old messages being displayed first and that's why I decided to buy GT430.
According to the Message 7753, the GT430 is recommended.

That's why the list/table of recommended and not recommended cards should have a separate page and shouldn't be a part of the forum. It forces people to read the whole thread and that can sometimes lead to mistakes.

I cannot change the web site (front page links, or default thread ordering), and I did say in that thread that the GT430 is not recommended. I'm a bit surprised why you went by a 2year old post from a cruncher that has not crunched for some time rather than by what I suggested, but New members do struggle with many aspects of crunching and using the forums.

I started another thread (locked) with just what I think are the recommended cards.
Hopefully the webmaster will edit the front page to link to this thread,
http://www.gpugrid.net/forum_thread.php?id=2507
CA's can edit this info to keep it up to date.

Just budget for a $200 or so card brand new and you can have confidence it'll do the project some good. ;)

---

I'm a bit surprised why you went by a 2year old post from a cruncher that has not crunched for some time rather than by what I suggested

Because that 2 years old thread was linked from the main page / System requirements section and I was simply not wary enough and too rash.
That was the only thread I had read before buying the card. That's it.

Now the thread is hidden for administrative purposes and that's good!

Fortunately I found different projects, where GT430 is still considered a very powerful card. In fact it seems to be 16 times faster than my 2.33GHz Core2Duo CPU for the offered tasks!

The GT430 was never a powerful card, more of an entry level card. Glad it's of use elsewhere.

The GT430 was never a powerful card, more of an entry level card. Glad it's of use elsewhere.

Well ... now I know it isn't powerful if we compare it to other graphics cards, but it seems to be faster than available CPUs for some kind of tasks.
It's a pity, that so many projects offer CPU tasks only, whilst low performance GPUs could do the job better :]

It's true that the GT430 is probably twice as fast as a top CPU (i7) for some things, and much less expensive, so I share your concerns. As yet, the great unrealized potential of GPU based research remains far from being achieved.
This project is one of but a few trailblazers. With limited resources and manpower and specific research objectives it is just not capable of utilizing all GPU classes from several generations. To meet the research objectives it naturally has to cater for the most powerful cards. At the present time these are mainly CC2.0 Fermi cards and high end 200 series cards.

The problem with GPUs is that while they pack an impressive amount of raw horse power, it's much more difficult to actually use it. That's why most current GPU projects are the ones using relatively simple algorithms, with the exceptions being GPU-Grid, Einstein, SETI and Folding@home (non-BOINC). None of these manage to extract as much work from the cards as the "simple" maths projects (Collatz, DNETC, PrimeGrid). Other projects would fare even worse.. without substantial effort.

MrS

---

Scanning for our furry friends since Jan 2002