SSP Memory Bandwidth: Vector

Previous slide

Next slide

Back to first slide

View graphic version

• This is because the cache is not being utilized and we are not getting the memory bandwidth needed. The high-bandwidth memory system of the T90 provides almost four times the throughput as that obtained on the SV1.

• Also note the almost one to one correlation between MFLOP performance and memory bandwidth. This will be a reoccuring observation that floating point performance scales with memory bandwidth