Semi-Interesting Memory Benchmarks

[jjsimas]

I have an old system, but I ran some interesting benchmarks.

Performance Increases For Different Memory Configurations:
Memory Configuration: Performance Increase:
Crucial PC100 CL2 1x128MB ECC +00.0% (base case)
Kingston PC100 CL2 1x128MB Non +03.6%
Crucial PC100 CL2 2x128MB ECC +06.4%
Kingston PC100 CL2 2x128MB Non +13.6% (this one is speculative, see note at bottom)

Benchmark:
OfficeBench 1.2

My System:
WinME
AMD K6-400
Tyan S1598
FastTrak66 RAID 1
2 x IBM UDMA66 13GB
Matrox G200

Jason

(The last configuration is just speculative and was not actually tested. It seems that I get +6.4% by adding 1 stick of 128MB (tested). It seems that I get +3.6% per stick of 128MB if I drop ECC (tested). Therefore, I might get 13.6% increase, if move to 2 sticks of 128MB non-ECC (not tested).)

 

[StanTheMan]

not interesting at all....

 

[jjsimas]

Well that was a useful comment..

 

[pm]

Well, this changes a few of my preconceptions.

I had thought that the performance difference between ECC and non-ECC was relatively insignificant in the real world. But 3.6% is definitely something. I also thought that the performance increase of interleaving memory was less than 5% in the real world.

I don't think you can make that speculative prediction though. First, if going from ECC -> non-ECC got you 3.6% and going from non-interleaved ECC to interleaved ECC got you 6.4%, then going from ECC non-interleaved to non-ECC interleaved could be 3.6+6.4 = 10%, not 13.6%. I see where you are getting the 13.6% (because there are two of them), but this seems like a stretch to me. But this is aside from the fact that generally these things generally aren't additive.

They are interesting, but I think you are going out on a limb with that speculative benchmark result.