Connecting 2 or more Tesla S U1 series is it possible?


I was wandering if it is possible to connect two or more Tesla S1070 (or S870) together? I have read that every such unit is connected to a PC via special PCIex card. So for example if I had 4 x Tesla S1070 I would have to connect them separately to 4 PCs and do my distributed computation via Ethernet.

My question is: is it possible to connect a few Tesla S1070 (or S870) to one computer?

If you’re asking whether its possible to attach more than one S1070 unit to a single system, the answer is yes. However, there could be SBIOS problems that cause problems.

How such connection works? Are a few Tesla S1070 connected directly to one computer via PCIex card? The number of devices would be limited by amount of PCIex slots in a PC. Or maybe a number of Tesla units are connected together via a special link and only one Tesla is connected directly to a computer?

I don’t know what SBIOS is. Is it a non typical solution to connect many Tesla in one system? I haven’t seen any information about this on nvidia’s pages.

Every GPU requires a CPU core for optimal performance… if you don’t have an octacore/2-quadcores give up.
Insterad, it’s very common to have many Teslas attached to many computers, and then have a normal MPI cluster (with local processing on GPUs)

Thank you for reply. I thought it works in such a way (every Tesla to a PC and then MPI), but I wasn’t sure.