Hi there,

I am considering running both SETI and GPUGRID on this Pc. It has 2 graphic cards, both of them are Nvidia 9500 GT's. Does anyone else use both projects? I am wondering what the impact will be as GPUGRID has a 4 day turnaround and I think SETI is set up differently. I don't want to find that I will be putting results in after the allotted time for the GPUGRID only to find I may not recieve credit.
I ask because there are times when GPUGRID does not send any work after finishing a task and the gpu is idle. Would it be better not to combine them both? Any views would be helpfull.
Thanks.

Eric.

Hi there,

I am considering running both SETI and GPUGRID on this Pc. It has 2 graphic cards, both of them are Nvidia 9500 GT's. Does anyone else use both projects? I am wondering what the impact will be as GPUGRID has a 4 day turnaround and I think SETI is set up differently. I don't want to find that I will be putting results in after the allotted time for the GPUGRID only to find I may not recieve credit.
I ask because there are times when GPUGRID does not send any work after finishing a task and the gpu is idle. Would it be better not to combine them both? Any views would be helpfull.
Thanks.

Eric.

I run ClimatePrediction on the Quad, and split my 9800GTX+ 50% S&H, and 50% GPUGrid. I have had no genric issues running SETI & GPUGrid, and have not heard of many reports of a clash on the GPU, or reports attributed to crash clash SETI/GPUGrid. The GPU can only handle one type of App at a time at the moment, software still "young" and not as flexible as CPU based tasking, but nonetheless the mix works fine providing its run separately.

I use Raistner's Team Opt App for SETI (V7), which cuts processing time down vastly. Its worth using one of his Opt Apps, there are several variants covering different uses and needs (Version numbewrs indicate use not quality - ie V7 & V10 for different uses, not so much one is faster than the other). They are designed to run SETI on both the CPU & GPU, therefore some manual intervention is needed at times to do what I am doing, but its not onerous, works well. You should use a opt app at SETI no matter which WU you crunch - its much faster.

SETI have two tpes of WU:
Astropulse: Broadly speaking its an equivalent length to the GPUGrid WU, but runs only on the CPU. Its "new" to SETI, a few weeks since full Production, so has had its share of introduction niggles as you would expect, but its settling nicely. Most of the comment/controversy on the SETI Boards is caused by a "Culture" change. Traditionaly SETI WUs are short (these are the MB WUS - Multi Beam), takes 5-60 mins to crunch depending on a lot of factors. The Astropulse length of crunching was a shock to the traditional SETI cruncher of 10 years standing, and took some getting used to. Astropulse currently cannot run on a GPU.

MB - Multi Beam WU, the traditional SETI WU. These will run on both the CPU and the GPU, and are EXACTLY the same WU - there is no different one for CPU & GPU. The only difference is the application will put the "CUDA" label against those initially earmarked to run on the GPU. A fast i7 / 295 will knock these off every 3 mins or so on the GPU with minimal CPU loading, very rapid with an opt app. A slow Pentium D on the CPU will run them every few hours.

The only issues I ave seen on the boards from those mixing the Projects were in the end due to their own errors - usually too enthusiastic on the overclocking, or incorrect use of the Opt Apps. Its worked well for me, just be aware you will not get automatic switching between apps (GPUGrid/SETI), unless you run the cache of one dry, then BOINC will pick up the GPU capable remainder and feed that. So there is an element of baby sitting, but as I use the PC a lot, one manual switch a day is no hassle.

SETI Opt Apps are very good, the guys there know their stuff, and constantly improve the Opt Apps.

Regards
Zy

I run both (seti & CPUGRID) projects on all 5 of my CUDA PC's. Although seti CUDA work units have gotten stalled from time to time, with BOINC 6.6.17 (which I have been testing since it came out yesterday), it seams to be working great. Stalled work units now seem to auto abort and become computation errors, and BOINC moves on to the next work unit. The work unit caching for both CUDA projects have worked real well ever since BOINC 6.6.15.

Rick
____________

lol 5 to 60 minutes for a AP unit you must have a very fast machine, they use up to 750.000 seconds on my machines to complete.

lol 5 to 60 minutes for a AP unit you must have a very fast machine, they use up to 750.000 seconds on my machines to complete.