with SMT vs. without SMT
with SMT vs. without SMT
with SMT vs. without SMT
ScienceMark Membench | |||
| 厂商 | Dawning | Dawning | |
| 产品型号 | I620r-G Nehalem-EP Intel Gainestown Xeon E5540 2.53GHz | I620r-G Nehalem-EP Intel Gainestown Xeon E5540 2.53GHz 无超线程 | |
| 内存技术参数 | 2GB R-ECC DDR3-1066 SDRAM x12 | 2GB R-ECC DDR3-1066 SDRAM x12 | |
| L1带宽(MB/s) | 47705.89 | 47572.00 | |
| L2带宽(MB/s) | 19499.26 | 19353.08 | |
| 内存带宽(MB/s) | 8620.40 | 8623.25 | |
| L1 Cache Latency(ns) | |||
| 32 Bytes Stride | 2 cycles 0.79 ns | 2 cycles 0.79 ns | |
| L1 Algorithm Bandwidth(MB/s) | |||
| Compiler | 41485.23 | 42272.23 | |
| REP MOVSD | 43346.81 | 43279.73 | |
| ALU Reg Copy | 11990.39 | 12024.42 | |
| MMX Reg Copy | 47705.89 | 23374.45 | |
| SSE PAlign | 46105.34 | 47567.95 | |
| SSE2 PAlign | 48167.88 | 47572.00 | |
| L2 Cache Latency(ns) | |||
| 4 Bytes Stride | 3 cycles 1.19 ns | 3 cycles 1.19 ns | |
| 16 Bytes Stride | 3 cycles 1.19 ns | 3 cycles 1.19 ns | |
| 64 Bytes Stride | 9 cycles 3.56 ns | 9 cycles 3.56 ns | |
| 256 Bytes Stride | 8 cycles 3.17 ns | 8 cycles 3.17 ns | |
| 512 Bytes Stride | 8 cycles 3.17 ns | 7 cycles 2.77 ns | |
| L2 Algorithm Bandwidth(MB/s) | |||
| Compiler | 18042.06 | 12053.96 | |
| REP MOVSD | 19499.26 | 19353.08 | |
| ALU Reg Copy | 8766.52 | 8691.59 | |
| MMX Reg Copy | 13988.43 | 13947.73 | |
| SSE PAlign | 18664.20 | 18458.88 | |
| SSE2 PAlign | 18664.20 | 18287.87 | |
| Memory Latency(ns) | |||
| 4 Bytes Stride | 3 cycles 1.19 | 3 cycles 1.19 | |
| 16 Bytes Stride | 5 cycles 1.98 | 5 cycles 1.98 | |
| 64 Bytes Stride | 23 cycles 9.10 | 23 cycles 9.10 | |
| 256 Bytes Stride | 105 cycles 41.55 | 105 cycles 41.55 | |
| 512 Bytes Stride | 118 cycles 46.70 | 118 cycles 46.70 | |
| Memory Algorithm Bandwidth(MB/s) | |||
| Compiler | 8013.28 | 7990.45 | |
| REP MOVSD | 8620.40 | 8623.25 | |
| ALU Reg Copy | 7066.53 | 7014.69 | |
| MMX Reg Copy | 8098.63 | 8082.58 | |
| MMX Reg 3dNow | - | - | |
| MMX Reg SSE | 7288.34 | 7373.09 | |
| SSE PAlign | 7121.20 | 7133.16 | |
| SSE PAlign SSE | 8001.72 | 7999.09 | |
| SSE2 PAlign | 7123.08 | 7140.24 | |
| SSE2 PAlign SSE | 7985.25 | 8000.34 | |
| MMX Block 4kb | 6499.16 | 6505.21 | |
| MMX Block 16kb | 6873.16 | 6907.44 | |
| SSE Block 4kb | 6582.42 | 6573.33 | |
| SSE Block 16kb | 4681.34 | 6917.23 | |
关闭超线程之后，L1和内存性能着微弱的提升，但是L2性能下降比较明显，这表明Nehalem的L2 Cache可以充分满足处理器的需要，你不需要关闭超线程以获得极微弱的L1/内存性能提升。
