Intel Lynnfield Memory Performance Analysis

November 2, 2009 | 10:22

Tags: #12800 #1333 #1600 #2000 #2133 #analysis #article #benchmarks #c7 #c9 #channel #cl7 #cl8 #cl9 #core #ddr3 #dimm #dual #i5 #i7 #kit #lynnfield #perform #performance #review #tests

Companies: #intel

Lavalys Everest 5.3 Memory Performance

Website: Lavalys

Everest Memory Read Test

  • 1,600MHz CL9
  • 1,600MHz CL8
  • 1,600MHz CL7
  • 1,600MHz CL6
  • 2,000MHz CL9
  • 2,000MHz CL8
  • 2,000MHz CL7
    • 16783
    • 17354
    • 18395
    • 18117
    • 19546
    • 20039
    • 20267
0
5000
10000
15000
20000
MB/s
  • MB/s

Everest Memory Read Test

  • 1,800MHz CL9
  • 1,800MHz CL8
  • 1,800MHz CL7
  • 2,160MHz CL9
  • 2,160MHz CL8
    • 17669
    • 18005
    • 17729
    • 19187
    • 18607
0
5000
10000
15000
20000
MB/s
  • MB/s

Read performance is, as we'd expect, progressively better as we lower the latency and jump to a higher clock speed. Note that it actually drops for the lowest latency settings - this indicates that we are on the verge of stability and that some bandwidth is lost in errors rather than getting stuff sent quicker.

Everest Memory Write Test

  • 1,600MHz CL9
  • 1,600MHz CL8
  • 1,600MHz CL7
  • 1,600MHz CL6
  • 2,000MHz CL9
  • 2,000MHz CL8
  • 2,000MHz CL7
    • 15568
    • 15553
    • 15540
    • 15558
    • 15546
    • 15546
    • 15506
0
2500
5000
7500
10000
12500
15000
17500
MB/s
  • MB/s

Everest Memory Write Test

  • 1,800MHz CL9
  • 1,800MHz CL8
  • 1,800MHz CL7
  • 2,160MHz CL9
  • 2,160MHz CL8
    • 14202
    • 14206
    • 14097
    • 14214
    • 14135
0
2500
5000
7500
10000
12500
15000
MB/s
  • MB/s

The write performance is clearly dominated by CPU northbridge frequency, rather than any memory timing or frequency as the write results are virtually entirely uniform. This is to be expected, since the CPU is writing to memory, and clearly there's no limitation even from the slowest 1,600MHz CL9 setting.

Everest Memory Copy Test

  • 1,600MHz CL9
  • 1,600MHz CL8
  • 1,600MHz CL7
  • 1,600MHz CL6
  • 2,000MHz CL9
  • 2,000MHz CL8
  • 2,000MHz CL7
    • 22878
    • 20682
    • 20148
    • 21968
    • 22066
    • 21667
    • 22466
0
5000
10000
15000
20000
25000
MB/s
  • MB/s

Everest Memory Copy Test

  • 1,800MHz CL9
  • 1,800MHz CL8
  • 1,800MHz CL7
  • 2,133MHz CL9
  • 2,133MHz CL8
    • 20354
    • 20248
    • 20172
    • 21034
    • 19814
0
5000
10000
15000
20000
MB/s
  • MB/s

Copy performance is a little bit everywhere. It averages around the 20,000MB/s mark across all the tests, but there's no real trend line.

Everest Memory Latency Test

  • 1,600MHz CL9
  • 1,600MHz CL8
  • 1,600MHz CL7
  • 1,600MHz CL6
  • 2,000MHz CL9
  • 2,000MHz CL8
  • 2,000MHz CL7
    • 42.0
    • 39.9
    • 38.5
    • 36.9
    • 38.2
    • 36.6
    • 35.2
0
10
20
30
40
nanoseconds
  • nanoseconds

Everest Memory Latency Test

  • 1,800MHz CL9
  • 1,800MHz CL8
  • 1,800MHz CL7
  • 2,133MHz CL9
  • 2,133MHz CL8
    • 40.9
    • 39.6
    • 38.4
    • 38.1
    • 36.8
0
10
20
30
40
nanoseconds
  • nanoseconds

The overall latency is obviously affected by reducing the memory latency, with a clear reducing trend-line as we drop the timings or increase the frequency equating to a 13 per cent reduction across the 1,600MHz range alone and 17 per cent from 1,600MHz CL9 to 2,000MHz CL7. Varying CPU memory controller frequency doesn't have much affect, with a similar 11 per cent reduction from 1,800MHz CL9 to 2,133MHz CL8.
Discuss this in the forums
YouTube logo
MSI MPG Velox 100R Chassis Review

October 14 2021 | 15:04