A Simple Memory Benchmark

Introduction

I wrote a very simple memory benchmark. It performs a sequential read or write on a block of memory until 256 MB was read or written, for different block sizes. This should show the difference in memory performance for cache hits and cache misses.

Below are the results for some machines I have access to. callisto is my PowerPC.

Memory Read Performance (MB/s)

block size	mercator	stevin	sp05	callistoE	callistoS	piefje	sp17	bartm
64	350.685	287.640	104.065	609.524	609.524	433.898	218.803	640.000
128	345.946	263.918	110.823	609.524	609.524	501.961	222.609	731.429
256	400.000	290.909	110.823	640.000	640.000	522.449	237.037	775.758
512	419.672	290.909	117.972	673.684	673.684	522.449	246.154	825.806
1K	433.898	308.434	116.895	691.892	711.111	544.681	234.862	800.000
2K	441.379	308.434	117.431	711.111	711.111	544.681	253.465	825.806
4K	449.123	316.049	118.519	711.111	711.111	556.522	243.810	825.806
8K	419.672	320.000	119.626	711.111	711.111	556.522	237.037	825.806
16K	161.006	324.051	117.972	731.429	711.111	533.333	220.690	825.806
32K	162.025	174.150	118.519	200.000	711.111	201.575	224.561	800.000
64K	146.286	180.282	117.431	130.612	139.891	192.481	269.474	253.465
128K	142.222	172.973	89.510	134.737	134.031	188.235	237.037	253.465
256K	117.972	179.021	89.199	113.274	123.671	175.342	211.570	250.980
512K	109.871	168.421	89.510	100.787	113.778	163.057	201.575	250.980
1M	94.465	110.823	89.199	90.780	106.224	146.286	211.570	142.222
2M	89.199	109.871	94.465	87.372	101.587	143.017	204.800	141.436
4M	88.276	109.402	93.773	87.372	101.186	140.659	200.000	142.222
8M	88.889	108.936	91.103	87.075	101.587	139.891	206.452	142.222

Memory Write Performance (MB/s)

block size	mercator	stevin	sp05	callistoE	callistoS	piefje	sp17	bartm
64	82.848	426.667	253.465	522.449	533.333	83.388	465.455	581.818
128	82.315	433.898	297.674	568.889	568.889	83.934	595.349	691.892
256	82.848	465.455	332.468	609.524	595.349	85.050	640.000	731.429
512	84.211	512.000	350.685	624.390	640.000	84.211	691.892	775.758
1K	83.934	512.000	355.556	640.000	640.000	84.211	731.429	752.941
2K	82.581	522.449	360.563	656.410	656.410	84.211	731.429	775.758
4K	83.934	522.449	365.714	656.410	640.000	85.050	691.892	775.758
8K	83.660	544.681	365.714	656.410	656.410	83.388	656.410	800.000
16K	83.117	556.522	365.714	656.410	673.684	84.211	711.111	800.000
32K	83.117	533.333	365.714	148.837	656.410	85.050	711.111	731.429
64K	81.529	544.681	360.563	93.431	100.787	84.211	711.111	131.282
128K	82.848	533.333	119.626	97.338	94.815	83.388	691.892	131.282
256K	79.012	501.961	119.626	78.769	85.619	84.211	400.000	131.282
512K	82.848	406.349	120.188	69.377	78.287	84.211	371.014	129.949
1M	82.581	148.837	120.188	61.836	72.113	84.488	400.000	69.755
2M	82.848	142.222	119.626	59.535	68.633	84.488	400.000	69.565
4M	82.848	146.286	116.895	59.259	68.817	83.388	406.349	69.565
8M	82.581	145.455	114.286	59.953	69.003	84.211	328.205	70.330

Machines

Hostname	CPU	Clock Speed	Chipset	Bus Speed	L1 cache	L2 cache	Main memory
mercator	Pentium	133 MHz	i430FX	66 MHz	2x8K	512K	64M EDO
stevin	UltraSPARC	167 MHz	Ultra-2	83 MHz	2x16K	1M	256M
sp05	POWER2	66 MHz	SP2 thin	66 MHz	64K	-	256M
callistoE	604e	200 MHz	VLSI GG2	66 MHz	2x32K	512K	32M EDO
callistoS	604e	200 MHz	VLSI GG2	66 MHz	2x32K	512K	32M SD
piefje	Pentium MMX	166 MHz	i430TX	66 MHz	2x16K	512K	64M EDO
sp17	POWER2	135 MHz	SP2 wide	??	128K	-	512M
bartm	K6	233 MHz	i430TX	66 MHz	2x32K	512K	64M SD

Program Source

The C source is to be compiled with the command

gcc -O3 -fomit-frame-pointer MemSpeed.c -o MemSpeed

This page is maintained by Geert Uytterhoeven.
$Date: 2006-01-25 21:24:49 $