latency latency on tesla c1060

on paragraph 5.1.1.3 of progarmming guides I read about 400 to 600 clock cycle latency in memory when I log in memory … I think are very large latencies!

actually an algorithm performed by me on Tesla C1060 (cholesky) I have some great times when accessing the global … someone could motivate this latency so high?

This is by design - in general, GPUs cover memory latency by having thousands of threads in flight, so when one processor stalls on a memory access, there’s always plenty of other work to do. There’s more detail in the documentation.

please, what documentation??

Here: Get CUDA
Select your operating system. When you want to install CUDA, read Getting Started Guide. If you want to know how to program CUDA, read Programming Guide.