这三个测试是Intel推荐的项目之一,因此笔者也使用其进行了测试:
MMM - Matrix-Matrix Multiplicaion Benchmark | ||||
处理器 | 双路Intel Nehalem-EP Xeon X5570 | 双路Intel Westmere-EP Xeon X5670 | 双路Intel Westmere-EP Xeon X5680 | DELL PowerEdge M910 四路Intel Nehalem-EX Xeon E7540 |
单位 | GFLOPS | GFLOPS | GFLOPS | GFLOPS |
Threads 1 | ||||
5000 step | 7.821975 | 7.842319 | 8.877563 | 5.867645 |
10000 step | 7.890761 | 7.840417 | 8.883291 | 5.865347 |
15000 step | 7.888751 | 7.845479 | 8.881528 | 5.826604 |
Threads 2 | ||||
5000 step | 15.59136 | 15.62796 | 17.5891 | 11.570261 |
10000 step | 15.7544 | 15.66469 | 17.73566 | 11.689317 |
15000 step | 15.7445 | 15.64657 | 17.67208 | 11.602247 |
Threads 4 | ||||
5000 step | 30.69218 | 29.99696 | 34.85343 | 21.788607 |
10000 step | 31.02227 | 29.75883 | 34.90105 | 22.21115 |
15000 step | 31.04954 | 30.55926 | 34.92557 | 22.073702 |
Threads 8 | ||||
5000 step | 36.2252 | 49.03697 | 45.99856 | 41.228878 |
10000 step | 38.21083 | 50.30305 | 45.99856 | 43.472432 |
15000 step | 40.71236 | 56.00031 | 47.74417 | 43.37777 |
Threads 16 | ||||
5000 step | 59.38371 | 64.04222 | 66.10022 | 73.379889 |
10000 step | 61.44583 | 62.42291 | 72.38159 | 78.596851 |
15000 step | 61.83442 | 64.3761 | 73.2495 | 79.099092 |
Threads 24 | ||||
5000 step | 54.82514 | 84.13599 | 66.10022 | 94.000418 |
10000 step | 54.82514 | 88.58685 | 72.38159 | 124.028823 |
15000 step | 59.18915 | 90.12297 | 73.2495 | 124.574801 |
Threads 48 | ||||
5000 step | 97.335138 | |||
10000 step | 119.780984 | |||
15000 step | 121.637469 |
MMM是一个类似矩阵乘法基准测试软件,得到的结果单位是GFLOPS,也就是说它是一个浮点测试。可以看到,M910平台/Nehalem-EX平台具有着最高的计算能力:121.6GFLOPS,表现比Sandra的结果更抢眼。MMM和Linpack一样,都能充分地利用CPU运算核心的能力,因此它实际上建议关闭超线程来测试。
SunGard Adaptiv Analytics Benchmark v4.0 | ||||
处理器 | 双路Intel Nehalem-EP Xeon X5570 | 双路Intel Westmere-EP Xeon X5670 | 双路Intel Westmere-EP Xeon X5680 | DELL PowerEdge M910 四路Intel Nehalem-EX Xeon E7540 |
Threads | 16 | 24 | 24 | 48 |
Time (lower is better) | 138.076s | 110.331s | 94.911s | 139.512s |
这个测试程序是SunGard风险分析管理套件的一个部分,多线程的优势并不大,反而是高频率的好处多一些,因此M910平台是垫底的。
black_scholes | ||||
处理器 | 双路Intel Nehalem-EP Xeon X5570 | 双路Intel Westmere-EP Xeon X5670 | 双路Intel Westmere-EP Xeon X5680 | DELL PowerEdge M910 四路Intel Nehalem-EX Xeon E7540 |
Threads | 16 | 24 | 24 | 48 |
Time (lower is better) | 9.17s | 6.16s | 5.51s | 4.40s |
black_scholes是对布莱克-肖尔斯期权定价模型进行计算的一个程序,布莱克-肖尔斯期权定价模型是由1997诺贝尔经济学奖的两个获得者创立和发展的模型。可以知道,这个测试对多线程的支持比较好,M910取得了最好的性能。