juin

Distinguished
May 19, 2001
3,323
0
20,780
http://www.cs.virginia.edu/stream/standard/Bandwidth.html

Gb\S per cpu

Alpha EV7
HP_AlphaServer_GS1280 1 5816.0 5816.0 6488.0 6345.0
HP_AlphaServer_GS1280 64 347712.3 341890.6 373126.5 377727.
Almost 6GB per CPU not bad for a dead end platform thank to a very powerful RDRAM systemes with also about 40 CPU cycle latency that also the best latency i know.Pc 2100 or also know as Pepsi can 2100 cannot rival with this.


Hp super Dome

HP_SuperDome_16cell 64 82276.0 81269.0 83037.0 84049.0

Not bad but far from Alpha Ev7 still better that the rest (sun IBM) can be equipe with less cpu for better CPU\GBs.


IBM P690 power series 1.7 GHZ
IBM_eServer_p690+ 32 33737.0 33718.0 40198.0 41064.0

Barely 1GB per cpu i wonder if there is not better score for power.It scaling seen to be bad when moving over 16 cpu (32 core).


SGI altix (best itanium systems)

SGI_Altix_3000 64 106457.0 105359.0 123968.0 124601.0

Better score that HP and perfect scaling.A bit more 1.5GB per cpu.

SUN slowpoke

Sun_F25K-1050 72 68184.1 67462.7 75402.7 76097.8

Not a surprise coming from sun less that 1GB\S.

Opteron

AMD_Opteron_246 1 2398.9 2233.5 2414.1 2209.9 data

Like expected the intergrade memory controleur make the cut.Over 2.3 GB per cpu with PC 2700 faster score can be archive with PC 3200.

When you compare opteron 1 way to 2 way you a very large decrease in bandwith.This can be a result of linux NUMA and not the opteron itself.
PC3200 bench
AMD_Opteron_248 2 5488.7 5435.7 6025.0 6002.4 data
AMD_Opteron_248 1 2819.8 2815.4 3146.0 3144.0 data

PC2700
AMD_Opteron_244 2 4040.8 4009.2 4436.9 4555.4 data
AMD_Opteron_244 1 2342.6 2272.8 2497.3 2362.7 data



Apple_PwrMac_G5 2 1901.4 1772.9 2088.8 2059.2 data
Apple_PwrMac_G5 1 1578.3 1548.1 1746.7 1764.1 data
Asus_P4T533-P4-2800 1 1832.6 1736.2 2249.2 2249.2
Intel_875_P4_2.80 1 2391.0 2376.0 2742.0 2715.0 data
Reference score

i need to change useur name.