ScienceMark v2.0 Membench
ScienceMark v2.0是一款用于测试系统特别是处理器在科学计算应用中的性能的软件，MemBenchmark是其中针对处理器缓存、系统内存而设计的功能模块，它可以测试系统内存带宽、L1 Cache延迟、L2 Cache延迟和系统内存延迟，另外还可以测试不同指令集的性能差异。
| ScienceMark Membench | |||
|---|---|---|---|
| 厂商 | Intel | Intel | Intel |
| 产品型号 | Intel 单路至强W3570 3.2GHz | Intel 单路至强E5504 2GHz | PowerEdge 2900 III Intel Harpertown Xeon E5430 2.66GHz |
| 内存技术参数 | 2GB R-ECC DDR3-1333 SDRAM x4 | 2GB R-ECC DDR3-1333 SDRAM x4 | 4GB R-ECC DDR3-1333 SDRAM x6 |
| L1带宽(MB/s) | 103754.51 | 47877.41 | 55376.16 |
| L2带宽(MB/s) | 42247.04 | 19596.05 | 16757.55 |
| 内存带宽(MB/s) | 13884.17 | 8833.57 | 4485.09 |
| L1 Cache Latency(ns) | |||
| 32 Bytes Stride | 0.62 ns | 1.50 ns | 1.13 ns |
| L1 Algorithm Bandwidth(MB/s) | |||
| Compiler | 71112.48 | 42198.88 | 25201.96 |
| REP MOVSD | 94211.67 | 43498.52 | 25467.15 |
| ALU Reg Copy | 25659.21 | 12067.10 | 13093.65 |
| MMX Reg Copy | 52379.62 | 24173.05 | 25242.19 |
| SSE PAlign | 103651.52 | 47830.32 | 52826.21 |
| SSE2 PAlign | 103754.51 | 47877.41 | 55376.16 |
| L2 Cache Latency(ns) | |||
| 4 Bytes Stride | 0.94 | 2.00 ns | 1.13 ns |
| 16 Bytes Stride | 0.94 | 2.00 ns | 1.50 ns |
| 64 Bytes Stride | 2.81 | 5.00 ns | 4.51 ns |
| 256 Bytes Stride | 2.50 | 4.50 ns | 4.51 ns |
| 512 Bytes Stride | 2.50 | 4.00 ns | 4.89 ns |
| L2 Algorithm Bandwidth(MB/s) | |||
| Compiler | 37758.50 | 17957.58 | 11880.48 |
| REP MOVSD | 42247.04 | 19596.05 | 12536.88 |
| ALU Reg Copy | 19039.82 | 8778.56 | 8577.86 |
| MMX Reg Copy | 30510.41 | 14063.17 | 13408.31 |
| SSE PAlign | 40513.22 | 18656.42 | 16719.97 |
| SSE2 PAlign | 40513.22 | 18677.19 | 16757.55 |
| Memory Latency(ns) | |||
| 4 Bytes Stride | 0.94 | 2.00 ns | 1.13 ns |
| 16 Bytes Stride | 1.87 | 2.00 ns | 4.89 ns |
| 64 Bytes Stride | 8.44 | 8.50 ns | 19.17 ns |
| 256 Bytes Stride | 31.25 | 46.00 ns | 59.77 ns |
| 512 Bytes Stride | 35.94 | 52.00 ns | 68.04 ns |
| Memory Algorithm Bandwidth(MB/s) | |||
| Compiler | 8901.34 | 7918.04 | 3178.45 |
| REP MOVSD | 12489.75 | 8833.57 | 3220.23 |
| ALU Reg Copy | 7988.86 | 5631.16 | 2789.34 |
| MMX Reg Copy | 9030.04 | 5880.52 | 2972.91 |
| MMX Reg 3dNow | - | - | - |
| MMX Reg SSE | 13389.96 | 8398.25 | 3978.53 |
| SSE PAlign | 13210.83 | 8750.74 | 4128.59 |
| SSE PAlign SSE | 13876.73 | 8715.17 | 4390.48 |
| SSE2 PAlign | 13181.13 | 8749.69 | 4326.42 |
| SSE2 PAlign SSE | 13884.17 | 8724.84 | 4441.71 |
| MMX Block 4kb | 10887.56 | 7648.23 | 4063.30 |
| MMX Block 16kb | 11795.95 | 8515.20 | 4479.88 |
| SSE Block 4kb | 10974.57 | 7731.28 | 4074.79 |
| SSE Block 16kb | 11850.05 | 8620.84 | 4485.09 |
从测试中我们看到，至强W3570凭借着高主频在L1和L2的测试中遥遥领先，而在内存带宽和其他指令（集）项目的测试中，也领先对比平台至强E5504和至强E5430许多，三通道的性能优势发挥明显。
