为了体现出超线程对系统性能的影响，我们特地在另一台Nehalem-EP平台上作了打开/关闭超线程的测试。
| SiSoftware Sandra Pro Business 2009 | ||
|---|---|---|
|
测试对象
|
Intel Nehalem-EP
双路Intel Gainestown Xeon X5570 2.93GHz |
Intel Nehalem-EP
双路Intel Gainestown Xeon X5570 2.93GHz 无超线程 |
|
Processor Arithmetic Benchmark
处理器架构测试 |
||
|
Dhrystone ALU
|
142977MIPS
|
147034MIPS
|
|
Dhrystone ALU vs SPEED
|
48.75MIPS/MHz
|
50.13MIPS/MHz
|
|
Whetstone iSSE3
|
124035MFLOPS
|
80990MFLOPS
|
|
Dhrystone iSSE3 vs SPEED
|
42.29MFLOPS/MHz
|
27.61MFLOPS/MHz
|
|
Processor Multi-Media Benchmark
处理器多媒体测试 |
||
|
Multi-Media Int x16 iSSE4.1
|
296.85MPixel/s
|
260.47MPixel/s
|
|
Multi-Media Int x16 iSSE4.1 vs SPEED
|
101.21MPixel/s/MHz
|
88.81MPixel/s/MHz
|
|
Multi-Media Float x8 iSSE2
|
228.24MPixel/s
|
196.13MPixel/s
|
|
Multi-Media Float x8 iSSE2 vs SPEED
|
77.82kPixels/s/MHz
|
66.87kPixels/s/MHz
|
|
Multi-Media Double x4 iSSE2
|
125.88MPixel/s
|
102.33MPixel/s
|
|
Multi-Media Double x4 iSSE2 vs SPEED
|
42.92kPixels/s/MHz
|
34.89kPixels/s/MHz
|
|
Multi-Core Efficiency Benchmark
|
||
|
Inter-Core Bandwidth
|
75.61GB/s
|
32.66GB/s
|
|
Inter-Core Bandwidth vs SPEED
|
26.40MB/s/MHz
|
11.40MB/s/MHz
|
|
Inter-Core Latency
(越小越好) |
16ns
|
48ns
|
|
Inter-Core Latency vs SPEED
(越小越好) |
0.01ns/MHz
|
0.02ns/MHz
|
|
Memory Bandwidth Benchmark
内存带宽测试 |
||
|
Int Buff'd iSSE2 Memory Bandwidth
|
16.93GB/s
|
38.71GB/s
|
|
Float Buff'd iSSE2 Memory Bandwidth
|
16.90GB/s
|
38.52GB/s
|
|
Memory Latency Benchmark
内存延迟测试 |
||
|
Memory(Random Access) Latency
(越小越好) |
81ns
|
78ns
|
|
Speed Factor
(越小越好) |
61.40
|
61.60
|
|
Internal Data Cache
|
4clocks
|
4clocks
|
|
L2 On-board Cache
|
10clocks
|
9clocks
|
|
L3 On-board Cache
|
48clocks
|
46clocks
|
|
Cache and Memory Benchmark
缓存及内存测试 |
||
|
Cache/Memory Bandwidth
|
143.24GB/s
|
141.40GB/s
|
|
Cache/Memory Bandwidth vs SPEED
|
50.01MB/s/MHz
|
49.37MB/s/MHz
|
|
Speed Factor
(越小越好) |
20.90
|
21.90
|
|
Internal Data Cache
|
448.46GB/s
|
450.77GB/s
|
|
L2 On-board Cache
|
421.42GB/s
|
425.31GB/s
|
|
.NET Arithmetic Benchmark
.NET架构测试 |
||
|
Dhrystone .NET
|
32904MIPS
|
31208MIPS
|
|
Dhrystone .NET vs SPEED
|
11.22MIPS/MHz
|
10.64MIPS/MHz
|
|
Whetstone .NET
|
78286MFLOPS
|
55638MFLOPS
|
|
Whetstone .NET vs SPEED
|
26.69MFLOPS/MHz
|
18.97MFLOPS/MHz
|
|
.NET Multi-Media Benchmark
.NET多媒体测试 |
||
|
Multi-Media Int x1 .NET
|
62.28MPixel/s
|
55.60MPixel/s
|
|
Multi-Media Int x1 .NET vs SPEED
|
21.23kPixels/s/MHz
|
18.96kPixels/s/MHz
|
|
Multi-Media Float x1 .NET
|
26.19MPixel/s
|
15.95MPixel/s
|
|
Multi-Media Float x1 .NET vs SPEED
|
8.93kPixels/s/MHz
|
5.44kPixels/s/MHz
|
|
Multi-Media Double x1 .NET
|
51.45MPixel/s
|
29.85MPixel/s
|
|
Multi-Media Double x1 .NET vs SPEED
|
17.54kPixels/s/MHz
|
10.18kPixels/s/MHz
|
SiSoftware Sandra对比，用蓝色标出了性能特出的项目
关闭超线程之后，计算能力普遍下降，对L1/L2缓存的压力降低（因而L1/L2缓存带宽也就略为上升了），然而内存吞吐量也下降了，这表明每CPU三通道DDR3-1333对8个逻辑处理器是足够的。
只有极少数的项目中，关闭超线程获得了更好的测试成绩。Nehalem-EP的超线程比起Pentium 4时代有了不少的改进，你不应该将其关闭。
