Multiple GPUs Do I have to buy an identical GPU?

I have a real time application running on a GTX 8800. I am using the same GPU for visualization. I realized that CUDA is limited if the same card is used for graphics therefore I’m thinking about getting a new GPU. Do I have use SLI if I want to add another GPU? . i.e should I buy an identical card or I can use an older card that I have (6800 GT on another PCI exp)? PS my application is real time and I’m using the GlLmapbuffer right now. So I’m not sure how much the performance is going to increase if I use another GPU since I can’t use GlLmapbuffer anymore.

BTW when I tested the bandwidth of CUDA I found out that my bandwidth is too low. Is it because I’m running CUDA on the same card that I’m doing graphics?

Host to Device Bandwidth for Pageable memory

Transfer Size (Bytes)   Bandwidth(MB/s)

 33554432               576.9

Device to Host Bandwidth for Pageable memory

Transfer Size (Bytes)   Bandwidth(MB/s)

 33554432               807.1

Device to Device Bandwidth

Transfer Size (Bytes)   Bandwidth(MB/s)

 33554432               71109.9

CUDA doesn’t use SLI.

As for your bandwidth numbers, those are most often limited by your motherboard (and any bugs/problems in its BIOS).

Thanks.

So I moved my GTX 8800 to another computer. Now I have a 6600 GT for visualization and my motherboard is an ASUS p5nd2-sli but now the results of my bandwidth test are worst!

Is there any manual for using another GPU with cuda? Now that I have another GPU, Do I need to change anything in the codes?

Here is my new bandwidth test results.

>bandwidthtest

Quick Mode

Host to Device Bandwidth for Pageable memory

Transfer Size (Bytes)   Bandwidth(MB/s)

 33554432               184.8

Quick Mode

Device to Host Bandwidth for Pageable memory

Transfer Size (Bytes)   Bandwidth(MB/s)

 33554432               201.8

Quick Mode

Device to Device Bandwidth

Transfer Size (Bytes)   Bandwidth(MB/s)

 33554432               45942.6

>bandwidthTest --memory=pinned

Quick Mode

Host to Device Bandwidth for Pinned memory

Transfer Size (Bytes)   Bandwidth(MB/s)

 33554432               186.3

Quick Mode

Device to Host Bandwidth for Pinned memory

Transfer Size (Bytes)   Bandwidth(MB/s)

 33554432               205.0

Quick Mode

Device to Device Bandwidth

Transfer Size (Bytes)   Bandwidth(MB/s)

 33554432               69996.7

It’s very weird that your Device-to-Device performance went up with the pinned test, since pinned memory should have no impact on Device-to-Device copies.

Lacking any other good ideas: Are you sure both PCI-Express power connectors on your 8800 GTX are plugged in fully? I know the card will downclock itself if it isn’t getting enough power. What kind of power supply is in this system?

I have checked my card on two different computers and got these results. I’m pretty sure the power connections are fine. I also checked the voltage of the power; it is 11.98. Is there any software for testing the card?

Which driver version are you using?
What kind of power supply are you using?

in post #1 :

Cuda compilation tools, release 1.1, V0.2.1221

Power:Antec SL350, 350W

in post #3 :

Cuda compilation tools, release 1.1, V0.2.1221

Power: Antec TPII-550, 550W

I have also tried using only 8800 GTX in the second computer (post #3) and I got similar bandwidth results. :(

I have installed CUDA 2, why it prints release 1.1?

You installed the CUDA_2.0-beta.

Do you think I should ask my question in the hardware section?