This slide is out of date now :P
GK110 is finally available to the everyday consumer
This comment was marked helpful 0 times.
ypk
Warp execution makes GPU possible to hide latency.
This comment was marked helpful 0 times.
kayvonf
@ypk: Could you clarify what you meant? Executing the same instruction for all CUDA threads in a warp on SIMD ALUs enables parallel execution, but does not hide latency. The processor however interleaves execution of many warps to hide large latency operations performed by any one warp.
This slide is out of date now :P GK110 is finally available to the everyday consumer
This comment was marked helpful 0 times.
Warp execution makes GPU possible to hide latency.
This comment was marked helpful 0 times.
@ypk: Could you clarify what you meant? Executing the same instruction for all CUDA threads in a warp on SIMD ALUs enables parallel execution, but does not hide latency. The processor however interleaves execution of many warps to hide large latency operations performed by any one warp.
This comment was marked helpful 0 times.