SE10P:
1. 单精度峰值性能: 32 SP FLOPs/clock/core * 61 cores * 1.1GHz =2147.2 GFLOP/s
2. 双精度峰值性能: 16 DPFLOPs/clock/core * 61 cores * 1.1GHz = 1073.6 GFLOP/s
3. 内存带宽: 4 Bytes/channel * 16 mem. channels * 5.5GT/s= 352GB/s
5110P:
1. 单精度峰值性能: 32 SP FLOPs/clock/core * 60 cores * 1.053GHz =2021.76 GFLOP/s
2. 双精度峰值性能: 16 DPFLOPs/clock/core * 60 cores * 1.053GHz = 1010.88 GFLOP/s
3. 内存带宽: 4 Bytes/channel * 16 mem. channels * 5.0GT/s= 352GB/s
注:32SP FLOPs/clock/core =512/32*2,是指512bits向量化和FMA指令,双精度类似。