Hello, I have been experimenting with app_config.xml and Im wondering if anyone else has tried 2 WUs on 1 GPU? I have it running now and it seems stable, my setup:
i7-3770k @ 4.2Ghz
GTX 780Ti
GTX 670
Win7 64bit
SWAN_SYNC=0
WDM=Disabled

Am trying this because of the less than 100% GPU utilisation when running 1 WU on a GPU, curiously running 2 WUs at 50% each has not resulted in 100%, but is still higher. I read about Win7 not doing a good utilisation job on Win7, trying to overcome it.

Anyone think this will produce more points?

Hello: I have personally experienced in Linux, GTX770+FX8350 there is a small improvement in GPUGRID (in other projects with lower GPU load if it works great) but have two problems.

If tasks are no longer in danger of exceeding the 24H and lose the bonus and long process increases the possibility of error in the short tasks is better, always you will have the bonus but the danger with longer processing time always exists. Greetings.

There was a thread in which Jacob Klein describes his experiences running 2 WUs. The short answer is that it leads to more WU errors and isn't a good idea. Look up the thread and you can read the whole epistle.

Thanks for the replies, I could not find that thread by Jacob but I will look out for issues.
Re the long running WU and lack of bonus am pretty sure that wont be an issue on the 780Ti, it seems to tear through even long tasks, with a single WU I was seein 5 hours. But it might be a problem on the 670.
I may use cc_config to lock the 670 to short tasks only and if running 2 WU together on that card is still a problem then make the short tasks run one at a time.
Ultimately Im in this for the science (not points) so if I see errors or invalid results I will be back to 1WU per GPU quite quickly.
Thanks for the advice.

Thanks for the replies, I could not find that thread by Jacob but I will look out for issues.
Re the long running WU and lack of bonus am pretty sure that wont be an issue on the 780Ti, it seems to tear through even long tasks, with a single WU I was seein 5 hours. But it might be a problem on the 670.
I may use cc_config to lock the 670 to short tasks only and if running 2 WU together on that card is still a problem then make the short tasks run one at a time.
Ultimately Im in this for the science (not points) so if I see errors or invalid results I will be back to 1WU per GPU quite quickly.
Thanks for the advice.

If the project requested we don't do it, then I am not doing it, good enough for me, thanks.

Wow. Such misinformation. Please be sure to read that WHOLE thread, if you want the WHOLE story.

The invalid results were not caused at all by running multiple tasks on the same GPU. The invalid results were caused by a BOINC installation problem that resulted in registry entries, such that part of BOINC's data directory was stored in the VirtualStore directory of Windows, and was not being properly cleaned by BOINC. And then GPUGrid was not properly running results through validation, erroneously granting them full credit even when they failed after a couple seconds on my botched installation. Here's the post of my solution, after weeks and weeks of research. I'm obviously a bit miffed that you erroneously attributed the problem to running multiple tasks on the same GPU.
http://www.gpugrid.net/forum_thread.php?id=3332&nowrap=true#29894
Sheesh.

You CAN run 2x on a single GPU, and as far as I know, the admins do not have any problems with it. If you are going to attempt it, I recommend a GPU that has at least 2 GB of VRAM, alongside a high-speed internet connection. More details might be found here:
http://www.gpugrid.net/forum_thread.php?id=3331
Good luck.

Looks like I forgot the end result of that excruciatingly long year old thread and didn't wade through the entire epistle. Sorry.
Are you running 2x now? Is anyone currently running 2x? Did the admins ever change their stance and say it was OK?

Regards

The admins never had a stance against it. So, any user is free to attempt it. My testing indicated a possible performance gain by running multiple, but others had a performance decrease. I only recommend attempting it if all of your GPUGRID GPUs in your system each have 2GB or more of VRAM, since tasks will immediately fail if they cannot allocate enough VRAM. Since I have a 1GB GTX460 in the mix, I only run 1 GPUGRID task per GPU.

Thanks for the clarification Jacob. On another topic, I swapped in a 750 Ti in place of my last GTX 460. In doing so gained around a 60% increase in speed (at the least, for some WUs the speed increase is almost 100%) at a decease in power usage of 100 Watts. In addition the 750 Ti is 2GB. The 750 Ti isn't expensive on sale and at my power rates should take about 1.5 years to pay for in electricity savings.

WDM=Disabled

So im a noob and through reading these forums i came to believe (be it true or not) that Windows Desktop Manager caused poor utilisation, therefore I disabled it. Cant say i saw any difference but who am i to argue.
I did notice that with 2 WUs on each gpu that my system responsiveness became quite slow at times, as this is my main pc i have decided to fall back to just 1 WU per gpu for now.
Thanks forum.

Ignore my last post it seems on further reading its WDDM not WDM that causes utilisation issues.

Okay, thanks. That's what I thought. Just making sure there isn't some tweak I'm missing out on.

Hello: I experienced again with two tasks per GPU using new ultra short tasks and in this case the result is a performance increase of 17% which is quite interesting.

Anyway I said is still valid, the choice of two tasks per GPU can be used only in very specific cases.

Hello: I experienced again with two tasks per GPU using new ultra short tasks and in this case the result is a performance increase of 17% which is quite interesting.

Anyway I said is still valid, the choice of two tasks per GPU can be used only in very specific cases.

I run 2 long WUs on 1 GPU for a few months - stable and improves perf by about 10%. and has low risk of going over 24hrs (longest I've seen is 16 hrs for some santi jobs)

I'm not sure what they show under GPU use - if it's per CU, then it's really hard to operate at 100% utilization. it's it's "something running on at least 1 CU", then I wonder why it's so far from 100%.

Hi :)

After reading tons of pages about multiple WUs on a single gpu, one can`t stop thinking to himself.

"Which would give the best result, error wise, and which solution gives the best performance boost:

1. Multiple WUs on one gpu.
2. Multiple gpus pr. WU.

I want to build my own "Folding-Cluster" and I don`t like abusing my gpus, so I hope nr. 2 wins.

Best regards :)

Number 2 doesn't even make sense to me. Each GPU can run 1-or-more tasks. You cannot have multiple GPUs tackle a single task, to my knowledge.

Hmm, ok.

Thanks for the answer though :)

The real answer is:
It's probably best to run 1 task per GPU.

If you have 2GB or more of memory on the GPU, you can attempt multiple tasks, and if you're not doing other things with the GPU (like suspending/resuming, gaming every so often, etc.), then it'll probably work fine. But if you plan on using the system for anything besides BOINC, then I'd recommend 1 task per GPU.