SiSoftware Sandra Pro Business 2009
SiSoftware Sandra是一款可运行在32bit和64bit Windows操作系统上的分析软件，这款软件可以对于系统进行方便、快捷的基准测试，还可以用于查看系统的软件、硬件等信息。从Sandra 2007开始支持SSE4指令集。SiSoftware Sandra所有的基准测试都针对SMP和SMT进行了优化，最高可支持32/64路平台，这也是我们选择这款软件的原因之一。
SiSoftware Sandra Pro Business 2009 | |||
测试对象 | Dawning A650 双路AMD Shanghai Opteron 2378 2.4GHz | Dawning A650 双路AMD Barcelona Opteron 2350 2.0GHz | DELL PE2900 III 双路Intel Harptown Xeon E5430 2.66GHz |
Processor Arithmetic Benchmark 处理器架构测试 | |||
| Dhrystone ALU | 63082MIPS | 51480MIPS | 91006MIPS |
| Dhrystone ALU vs SPEED | 26.28MIPS/MHz | 25.74MIPS/MHz | 34.21MIPS/MHz |
Whetstone iSSE3 | 62993MFLOPS | 51400MFLOPS | 78385MFLOPS |
| Dhrystone iSSE3 vs SPEED | 26.25MFLOPS/MHz | 25.70MFLOPS/MHz | 29.47MFLOPS/MHz |
Processor Multi-Media Benchmark 处理器多媒体测试 | |||
Multi-Media Int x8 aSSE2 | 187.70MPixel/s | 155.64MPixel/s | |
| Multi-Media Int x8 iSSE4.1 | 199.33MPixel/s | ||
Multi-Media Int x8 aSSE2 vs SPEED | 78.21kPixels/s/MHz | 77.82kPixels/s/MHz | |
| Multi-Media Int x8 iSSE4.1 vs SPEED | 74.94kPixels/s/MHz | ||
Multi-Media Float x4 iSSE2 | 81.53MPixel/s | 67.86MPixel/s | 108.69MPixel/s |
Multi-Media Float x4 iSSE2 vs SPEED | 33.97kPixels/s/MHz | 33.93kPixels/s/MHz | 40.86kPixels/s/MHz |
Multi-Media Double x2 iSSE2 | 44.51MPixel/s | 37.15MPixel/s | 55.75MPixel/s |
Multi-Media Double x2 iSSE2 vs SPEED | 18.55kPixels/s/MHz | 18.58kPixels/s/MHz | 20.96kPixels/s/MHz |
Multi-Core Efficiency Benchmark | |||
Inter-Core Bandwidth | 6.54GB/s | 2.91GB/s | 20.54GB/s |
Inter-Core Bandwidth vs SPEED | 2.79MB/s/MHz | 1.49MB/s/MHz | 7.91MB/s/MHz |
Inter-Core Latency (越小越好) | 128ns | 185ns | 90ns |
Inter-Core Latency? vs SPEED (越小越好) | 0.05ns/MHz | 0.09ns/MHz | 0.03ns/MHz |
Memory Bandwidth Benchmark 内存带宽测试 | |||
Int Buff'd iSSE2 Memory Bandwidth | 16.59GB/s | 7.12GB/s | 6.13GB/s |
Int Buff'd iSSE2 Memory Bandwidth vs SPEED | 25.52MB/s/MHz | 10.94MB/s/MHz | 9.43MB/s/MHz |
Float Buff'd iSSE2 Memory Bandwidth | 16.58GB/s | 7.13GB/s | 6.13GB/s |
Float Buff'd iSSE2 Memory Bandwidth vs SPEED | 25.50MB/s/MHz | 10.96MB/s/MHz | 9.43MB/s/MHz |
Memory Latency Benchmark 内存延迟测试 | |||
Memory(Random Access) Latency (越小越好) | 106ns | 157ns | 108ns |
Memory(Random Access) Latency vs SPEED (越小越好) | 0.16ns/MHz | 0.24ns/MHz | 0.16ns/MHz |
Speed Factor (越小越好) | 83.80 | 103.40 | 95.20 |
Internal Data Cache | 3clocks | 3clocks | 3clocks |
L2 On-board Cache | 16clocks | 16clocks | 18clocks |
L3 On-board Cache | 58clocks | 47clocks | |
Cache and Memory Benchmark 缓存及内存测试 | |||
Cache/Memory Bandwidth | 77.08GB/s | 51.17GB/s | 68.88GB/s |
Cache/Memory Bandwidth vs SPEED | 32.89MB/s/MHz | 26.20MB/s/MHz | 26.52MB/s/MHz |
Speed Factor | 36.00 | 45.50 | 111.90 |
| Internal Data Cache | 299GB/s | 244.31GB/s | 421.23GB/s |
| L2 On-board Cache | 162.91GB/s | 135.04GB/s | 122.68GB/s |
.NET Arithmetic Benchmark .NET架构测试 | |||
Dhrystone .NET | 12736MIPS | 9551MIPS | 10562MIPS |
Dhrystone .NET vs SPEED | 5.31MIPS/MHz | 4.78MIPS/MHz | 3.97MIPS/MHz |
Whetstone .NET | 38737MFLOPS | 31231MFLOPS | 45399MFLOPS |
Whetstone .NET vs SPEED | 16.14MFLOPS/MHz | 15.62MFLOPS/MHz | 17.07MFLOPS/MHz |
.NET Multi-Media Benchmark .NET多媒体测试 | |||
Multi-Media Int x1 .NET | 24.48MPixel/s | 20.11MPixel/s | 31.28MPixel/s |
Multi-Media Int x1 .NET vs SPEED | 10.20kPixels/s/MHz | 10.06kPixels/s/MHz | 11.76kPixels/s/MHz |
Multi-Media Float x1 .NET | 5.29MPixel/s | 4.34MPixel/s | 8.68MPixel/s |
Multi-Media Float x1 .NET vs SPEED | 2.20kPixels/s/MHz | 2.17kPixels/s/MHz | 3.26kPixels/s/MHz |
Multi-Media Double x1 .NET | 21.31MPixel/s | 17.49MPixel/s | 24.75MPixel/s |
Multi-Media Double x1 .NET vs SPEED | 8.88kPixels/s/MHz | 8.74kPixels/s/MHz | 9.30kPixels/s/MHz |
SiSoftware Sandra对比，用蓝色标出了性能特出的项目
和老的巴塞罗那相比，Multi-Core Efficiency Benchmark、Memory Bandwidth Benchmark、Memory Latency Benchmark、Cache and Memory Benchmark的成绩都得到了明显的提升，而且是在除以处理器主频的情况下，这表明处理器核心之间的Crossbar设计以及内存控制器的效率都得到了很大的改进，性能得到了成倍以上的提升。
不过，我们注意到Shanghai L3缓存的延迟似乎要比Barcelona的要高一些，此外，在纯粹的整数/浮点运算性能还具有5%左右的提升。另外，SiSoftware Sandra通常看起来会偏向Intel处理器一些，因为它可以很好地支持Intel的SSE4指令集，而AMD Shanghai的SSE4A等指令集就没这么好运了，这也能表现出Intel在编译器方面做出的努力。
详细的处理器运算效能，请期待我们的SPEC CPU 2006测试。这个测试要运行数天之久，调试也比较麻烦，因此本文未能包及。
