question on v1.1's Atomic ops and Async kernels

[yk_cadcg]

according to Guide0.9, 8600/8500 is of compute capability 1.1. While 8800 and Quadros are not. Sadly mine is 8800. But still I’m curious about potential big changes that 1.1 brings forward:
1, atomic ops, which solves write conflict to global memory among threads. This will enable massive old hindered works.
2, async kernels, which enables co-processing. This will boost old performance numbers.
Am I correct? Thanks for suggestions!

[pluk]

according to Guide0.9, 8600/8500 is of compute capability 1.1. While 8800 and Quadros are not. Sadly mine is 8800. But still I’m curious about potential big changes that 1.1 brings forward:

1, atomic ops, which solves write conflict to global memory among threads. This will enable massive old hindered works.

2, async kernels, which enables co-processing. This will boost old performance numbers.

Am I correct? Thanks for suggestions!

[snapback]212723[/snapback]

1. yes, atomic operations will only be available in devices of compute capability 1.1 (8600/8500).

2. no, asynchronous kernels are available in the 0.9 release of CUDA for all G8x cards, so not only for the devices of compute capability 1.1.

Ps. compute capability 1.1 has nothing to do with the software (CUDA or driver release) but only with the hardware platform (improvements in architecture).