CUDA Tesla

I am planning to go for TESLA C1060. How well does CUDA fit into the TESLA products. Will I get a higher performance of my existing CUDA applications running in 8800 GTX, with TESLA.

BTW, what is the CPU-GPU transfer rate in TESLA. This is something I need to keenly watch.Very poor performance for smaller data size. Fed up of this. TESLA should overcome this teething problem.

The Tesla C1060 is basically a GTX 280 with somewhat less memory bandwidth, 4 GB of memory, and better quality assurance testing. (Well, and no DVI ports, obviously.)

The GPU in the C1060/GTX 280 has 240 stream processors clocked at 1.3 GHz compared to the 128 stream processors at 1.35 GHz in the 8800 GTX, so you should see a substantial speed improvement.

CPU/GPU transfer rate is set by the speed of the PCI-Express bus (and the motherboard quality). The C1060 (like all NVIDIA cards after the 8800 GT) is a PCI-Express 2.0 device, which in practice should give you 5 to 6 GB/sec transfer rate for pinned memory, depending on your motherboard and RAM performance. The 8800 GTX was a PCI-Express 1.0 device, and therefore maxed out around 3 GB/sec.