Phenom II and sandy bridge integer
performance is fairly similar (clock for clock). The big advantage of sandy bridge (and what is shown by most benchmark software) is float
calculations. IF Llanos is intercepting the CPU float instructions, converting them, and sending them to the GPU, then the big advantage of sandy bridge disappears. A small GPU has many times the flops of a fast AVX-using sandy bridge.
The questions here are:
Is AMD rerouting float operations to the GPU in real time?
If so, can they offload all float operations or just some and what penalties are incurred?
Originally Posted by xd_1771
Nope it was a head start all along:
AMD takes 2 minutes to complete, SB takes 1m 36s or so
Even with this being true, sandy bridge is a generation newer (architecture wise) and is running at a 52%-89% faster clockspeed (top/top and bottom/bottom calculation though there is a potential difference of 111%) while only completing the task >25% faster and while (visibly) lagging behind in the other applications.
I think that the idea of a "fast enough" processor with a much faster GPU is the future (thus why Intel licensed IP from Nvidia).Edited by hajile - 4/26/11 at 8:33am