Fedora 7 64bit
Asus 8800GS
NVIDIA-Linux-x86_64-169.09-pkg2.run

Thanks

I tested two drivers, 173.14 and 177.13, no difference

I will test 169.09 ... maybe

I tested two drivers, 173.14 and 177.13, no difference

I initially had 173.14 installed (from here) but wasn't sure if that was a cuda driver or not
So followed the http://www.nvidia.com/object/cuda_get.html link and it suggested the 169.09 package for my setup.

So what do you guys think? It appears my older Nvidia cards don't support CUDA. :(
Looking for a card to put in my dedicated cruncher for this project.
I think I have settled on this:

ZOTAC ZT-98XES2P-FSP GeForce 9800 GTX

---

The 9800 GTX shouldn't be a bad cruncher, you can see the runtimes per WU of my card in the other thread you started...

But if you would like to, or are able to spend a little bit more money on the graphics card, one of the new GTX 260, or GTX 280 cards would be much faster!

But for a GTX 280, you would probably also need a new PSU, because it needs a 8pin plus a 6pin power connector - the GTX 260 only two 6 pin connectors...

---

pixelicious.at - my little photoblog

The 9800 GTX shouldn't be a bad cruncher, you can see the runtimes per WU of my card in the other thread you started...

But if you would like to, or are able to spend a little bit more money on the graphics card, one of the new GTX 260, or GTX 280 cards would be much faster!

But for a GTX 280, you would probably also need a new PSU, because it needs a 8pin plus a 6pin power connector - the GTX 260 only two 6 pin connectors...

A way to know which card is better for the money do this: Compute the peak Gflops of the card.

Peak Gflops = (shader clock)x(number of stream processors)x(3 flop)

The highest the better.


GDF

Peak Gflops = (shader clock)x(number of stream processors)x(3 flop)

The highest the better.
GDF

It's x2 flop I think

Thanks guys!
What I would like to know, is how CPU intensive is this? I am imagining the GPU is doing most the work. A core is tied up with it but not running 100% like we do with "traditional" projects.
So my conjecture/question is if the same RAC, with whatever card, can be achieved with a slightly slower (read cheaper) cpu?
I am very keen to see if it will be successful to run multiple tasks on multiple cards simultaneously.
If all my assumptions above are correct. Then rather than spend more to get a faster card, that ties up one of my good crunching cores.
Wouldn't it be cool to build a budget dedicated twin or even quad GPU cruncher?
I invision two of the $200 dollar cards. (although if one had the funds the GTX 280 is an amazing machine) A basic $50 atx board and a $50 processor, probably a 4200+. Using Stefan's numbers that would be like 8-10k RAC a day for the price of a PS3. Really exciting development you guys have here. I will do whatever I can to encourage this technology. And to think I thought the Cell BE was cool. ;)

---

Thanks guys!
What I would like to know, is how CPU intensive is this? I am imagining the GPU is doing most the work. A core is tied up with it but not running 100% like we do with "traditional" projects.

Actually the application uses also one core of the CPU to 100% plus the GPU as coprocessor...
I really can't say how important the CPU speed is for the GPU application, but earlier test have shown that if the app uses only 50% of a CPU core, the WUs were a good bit slower.

Would be interesting to see some comparisons with the same graphics cards but other CPUs in the thread you started...

---

pixelicious.at - my little photoblog

Thanks guys!
What I would like to know, is how CPU intensive is this? I am imagining the GPU is doing most the work. A core is tied up with it but not running 100% like we do with "traditional" projects.

Actually the application uses also one core of the CPU to 100% plus the GPU as coprocessor...
I really can't say how important the CPU speed is for the GPU application, but earlier test have shown that if the app uses only 50% of a CPU core, the WUs were a good bit slower.

Would be interesting to see some comparisons with the same graphics cards but other CPUs in the thread you started...

The CPU is not important at all. It appears to be using 100% of resources just because it is polling waiting for the accelerated kernel to finish.
So, any CPU should do.

We have built a machine with 6 GPU cores by putting together 3 Geforce 9800 GX2, a power supply of 1200W, an Nvidia 780i motherboard with 3 PCI-E 16x slots and a quad core CPU.

We are looking into building another one with Geforce 280.

GDF

The CPU is not important at all. It appears to be using 100% of resources just because it is polling waiting for the accelerated kernel to finish.
So, any CPU should do.

We have built a machine with 6 GPU cores by putting together 3 Geforce 9800 GX2, a power supply of 1200W, an Nvidia 780i motherboard with 3 PCI-E 16x slots and a quad core CPU.

We are looking into building another one with Geforce 280.

GDF

Wow! This sounds like a nice machine! :D
Want to give it away to me as a present? ;-)

---

pixelicious.at - my little photoblog

Actually the application uses also one core of the CPU to 100% plus the GPU as coprocessor...

Please excuse my ignorance, but does that mean the remaining cores are left free to do 'traditional' BOINC projects?

Here's to hoping this can get ported to windows soon. I'd love to donate my new 8800 GT to the cause. :)

The "100%" resource use of one CPU core doesn't interfere with other traditional BOINC projects, does it?

Actually the application uses also one core of the CPU to 100% plus the GPU as coprocessor...

Please excuse my ignorance, but does that mean the remaining cores are left free to do 'traditional' BOINC projects?

Yes, they are.

gdf

Here's to hoping this can get ported to windows soon. I'd love to donate my new 8800 GT to the cause. :)

The "100%" resource use of one CPU core doesn't interfere with other traditional BOINC projects, does it?

No, it should not.

gdf

We have built a machine with 6 GPU cores by putting together 3 Geforce 9800 GX2, a power supply of 1200W, an Nvidia 780i motherboard with 3 PCI-E 16x slots and a quad core CPU.

GDF

Could your clarify this.

Does a GPU task use only 1 CPU core and all avaialable GPU cores, in this case 6, or does each gpu task use 1 cpu and 1 graphics card (2 gpu cores) or just 1 gpu core ?

So your rig above, how many gpu tasks could run at the same time to use all cores, and how many cpus would that use and how many cpus would be left for other boinc projects ?

We have built a machine with 6 GPU cores by putting together 3 Geforce 9800 GX2, a power supply of 1200W, an Nvidia 780i motherboard with 3 PCI-E 16x slots and a quad core CPU.

GDF

Could your clarify this.

Does a GPU task use only 1 CPU core and all avaialable GPU cores, in this case 6, or does each gpu task use 1 cpu and 1 graphics card (2 gpu cores) or just 1 gpu core ?

So your rig above, how many gpu tasks could run at the same time to use all cores, and how many cpus would that use and how many cpus would be left for other boinc projects ?

This is tunable. At the moment we are using 1 CPU core for each GPU core. Regarding the machine, we are using it mainly outside BOINC.

gdf

news ?

For Linux64:
1) Install the latest Nvidia driver.....
GDF

yes sir :-))

thx for your efforts ;-)