Hi,

my WUs are going wrong? I've this problems for weeks. (But I haven't not crunch much in this time in common) In this days, I aborted a WU cause the computer was turned off most time and I could not finish not nearly before deadline. Two other WUs started and were running for a time, but failed after a fraction of whole run time.

I know that my 9800GTX+ is not the newest/fastest GPU but it should be able to finish WUs ..

I checked the temps of GPU while computing, it was round about 45°C.

Thanks a lot for your suggestions.

Best regards

Sebastian

Hmm, maybe i've coosed the wrong forum. Please move thread if this is the case. Sorry and thanks.

ACEMDlong takes about 600 MB VRAM, do not crunch it on 511 MB VRAM GPU, set your preferences to acemd2. I mean skgiven is going to send you PM what to do overall to be succesfull, if not post msg., i'll send you some recomendations.

Hi , clockspeeds, can have a big impact, on every nVIDIA GPU running GPUgrid
WU's.
I run my X9650(3GHz)@3.51GHz and the GTX480 on 1.6GHz, temps are (below)normal
50C for CPU and 80C for the 480, atm.(Air-cooled, open, no case, PSU is @ 35C).
Room temp, can cause the 480, when running 3 SETI at a time, running the fan at 80% doesn't get hotter, bcs, fan setting is based on GPU+RAM temps.
I still use 1 250, but the noise it makes, will force me to take it out and
seek another suitable (GTX470) card. Also have 1 side off and a extern fan blows
air in this side.

Compaired to a lower clocked 480 0.8GHz, speed isn't a lineair function of the WU crunch time.
(Impact on SETI WU's are much bigger, if using the anonymous LUNATICs app. x32f,
FERMI compatible, also runs on 200 series and even lower 8000 &9000 series).
The 'older' 8000 & 9000 series run a high clock-frequency, as do some of their
'newer'(G92)GTS250 (8800GT;9800GTX+)

Runtime/load-unload time of the CPU are less, probably through the overclock.
MFLOPS and frequency are alomost lineair, @3GHz 3000MFLOPS, @ 3.51GHz 3520MFLOPS! (10965M Integers)

Those Longruns work OK, 2 faults, but do give a nice credit!
I don't even know, if the 480 have a 'default setting', probably a factory setting.

---

Knight Who Says Ni N!

Hi cosmicfate. You would have little chance of completing Long run tasks on a 9800 GTX (CC1.1 card). As nenym suggests you should only run the short tasks.
If the short tasks do not fair any better then it’s likely your card is not up to the job of running the complex GPUGrid tasks, and you would be better off using it for another project. However you could try to use an earlier driver and run the 6.12app; this app is generally faster for the earlier cards.
See the FAQ: Best configurations for GPUGRID thread for general configuration tips.

Fred, the reference (NVidia default) frequency for the GTX 480 is 700MHz/1400MHz.
It’s worth remembering that if you get one failure in ten tasks this will undo a 10% overclock and that some overclocks actually result in a poorer performance.
The GTS250 as with the 9800GTX will probably struggle, especially with the 6.13app. So you would be doing the right thing by replacing it with a GTX470. Avoid the temptations of the 500 series Ti cards (CC2.0 good, CC2.1 not so special), but wait until after the 22nd March before buying one.

Hi skgiven, so you are saying that 9800gtx+ is probably not able to handle long runs correctly because of low computing power or because of Compute Capability 1.1? What is weird that standard tasks are always completed successfully on my Asus 9800gtx+ Dark Knight however each long run WU fail saying: SWAN: FATAL : swanMemcpyDtoH sync failed. very strange because I never saw such error message while computing standard WUs. I had some issues when I started to compute gpugrid because of faulty memory modelues (now replaced with new modules which were tested with memtest and prime95). Low card memory is not an issue neither because WUs do not use more than ~230MB for computing gpugird (checked with gpu-z). GPU temperature never goes higher than 56-58 C (depends on ambient temp). Card is slightly OC (6% gpu and shaders) but I already tried to use default frequencies however this didn't help. I use swansync 0 parameter, shall I try to disable this and use only GPU computing power? Or it is not worth and I should stick with standard WUs only?

Thanks for help or some hints.

Yes, there would be less chance of completing the long WU's because of CC1.1 but also because these tasks are built differently.

As for GPU memory, this depends on the task and the GPU; the long IBUCH task I am running on my GTX260 is using 511MB GDDR, but on my GTX470's 787MB is being used.
GPU Memory also varies during the run to some extent.
I'm not sure if this is impacting upon your 9800GTX+ with respect to the use of swan_sync or not but you could try running without using swan_sync (restart system after changing). If you get nowhere you should probably stick to the normal length tasks.

The other thing is, you might just be running out of system memory (RAM); 2GB is not much for W7, especially if the system is allocating some of this to the GPU.