3 ms·
also the code tends to be highly SIMD, code with a lot of data dependent branches benefits very little from higher memory bandwidth when the fetch latency from
by throwaway2048 12y ago
also the code tends to be highly SIMD, code with a lot of data dependent branches benefits very little from higher memory bandwidth when the fetch latency from ram is on the order of hundreds or even thousands of cycles.
- m_mueller 12y agoYes that's a good point. So low arithmetic intensity + SIMD benefits POWER mostly.