the speed of cuda work units

[whocrazy]

Hello.
I used to have a geforce 9600gt, and now I have a geforce gt240.
I noticed that certain cuda work units would run at different speeds on my 9600gt.

Some would finish quicker than others.
Now that I've upgraded, I dont see any real difference in speed or the time it takes to complete the work unit. Did I make the right choice in upgrading?
Thanks.

[Ravage7779]

I think you are just seeing a slew of the same work units right now. That's what I see right now.

[codysluder]

All projects are benchmarked by the Pande Group on designated hardware. If your GPU matches their benchmark machine, you can expect WUs on your machine to run very, very nearly at a constant rate. If your machine is radically different than the benchmark machine, there will be significant variations in time.

[P5-133XL]

Work units vary in the amount of time they take to complete. Some are big, some are small. the big ones, may take longer but they give more points. So don't worry too much about how long a WU takes to complete.

[whocrazy]

I long for the day where I can pick and choose what work units I want to do, be it cuda, cpu or linux or play station. I dont like the idea of limiting by platform IE, you can only do this protein if you use linux, or you can't do this protein unless you use a ps3.
one day I hope that every work unit can be done on any platform no matter what.
I use the console version of folding at home cuda. Other projects like gpugrid.net and setiathome have cuda, but do not let you see the graphics, for obvious reasons. I dont think that trying to designed a graphical viewer for cuda work units is such a good idea.

[codysluder]

I long for the day where I can pick and choose what work units I want to do, be it cuda, cpu or linux or play station. I dont like the idea of limiting by platform IE, you can only do this protein if you use linux, or you can't do this protein unless you use a ps3.
one day I hope that every work unit can be done on any platform no matter what.

I'm sure that F@h dreams of universal clients that can run on any platform, but ati and nvidia don't seem to want to cooperate with each other. Maybe the dream can be fulfilled if OpenCL becomes a reality, but that means abandoning the work that has been done with Brook/ati and with CUDA/nvidia. Unfortunately they can't do that because that would mean shutting down a bunch of active projects. Apparently developing for Brook/ati was suspended some time ago, awaiting OpenCL, but the current projects continue to run. A beta CUDA client was developed first for Windows and I'm sure they'd expand that to OS-X and Linux if it were a universally supported API, but now it appears that further CUDA development may also be dead-end, too, if the dream of OpenCL pans out. There's still a lot of work to be done to make OpenCL a reality, and if, say 50% of the work being done on CUDA can be reused in OpenCL, then it's still worth developing for CUDA.

I use the console version of folding at home cuda. Other projects like gpugrid.net and setiathome have cuda, but do not let you see the graphics, for obvious reasons. I dont think that trying to designed a graphical viewer for cuda work units is such a good idea.

Doesn't the CUDA viewer work pretty good now? From what I understand, it's the DirectX viewer that has lots of problems but since I never use either one, I really don't know.

[Baowoulf]

I agree whocrazy. I think it would be interesting for fun to try and have at least one of each client. CPU, GPU, SMP, and PS3 to be able to work on any type of WU. Since there is no universal client yet atm.

On the viewer I know the CPU one is wonky, but I don't know about the GPU one. Even if the CUDA one is working good what type does the ATI client use, a DirectX one as well?