last_updated: 2026-06-25

Memory Benchmark#

I was interested how my RAM performs. Now I know more.

sudo apt install mbw
mbw 512

example result#

locutus $ mbw 512
Long uses 8 bytes. Allocating 2*67108864 elements = 1073741824 bytes of memory.
Using 262144 bytes as blocks for memcpy block copy test.
Getting down to business... Doing 10 runs per test.
0       Method: MEMCPY  Elapsed: 0.06258        MiB: 512.00000  Copy: 8181.135 MiB/s
1       Method: MEMCPY  Elapsed: 0.06202        MiB: 512.00000  Copy: 8255.934 MiB/s
2       Method: MEMCPY  Elapsed: 0.06202        MiB: 512.00000  Copy: 8255.401 MiB/s
3       Method: MEMCPY  Elapsed: 0.06224        MiB: 512.00000  Copy: 8226.089 MiB/s
4       Method: MEMCPY  Elapsed: 0.06216        MiB: 512.00000  Copy: 8237.338 MiB/s
5       Method: MEMCPY  Elapsed: 0.06259        MiB: 512.00000  Copy: 8180.743 MiB/s
6       Method: MEMCPY  Elapsed: 0.06220        MiB: 512.00000  Copy: 8232.041 MiB/s
7       Method: MEMCPY  Elapsed: 0.06564        MiB: 512.00000  Copy: 7799.647 MiB/s
8       Method: MEMCPY  Elapsed: 0.06342        MiB: 512.00000  Copy: 8072.654 MiB/s
9       Method: MEMCPY  Elapsed: 0.06319        MiB: 512.00000  Copy: 8102.163 MiB/s
AVG     Method: MEMCPY  Elapsed: 0.06281        MiB: 512.00000  Copy: 8152.100 MiB/s
0       Method: DUMB    Elapsed: 0.03948        MiB: 512.00000  Copy: 12966.949 MiB/s
1       Method: DUMB    Elapsed: 0.03845        MiB: 512.00000  Copy: 13314.956 MiB/s
2       Method: DUMB    Elapsed: 0.03828        MiB: 512.00000  Copy: 13374.432 MiB/s
3       Method: DUMB    Elapsed: 0.03829        MiB: 512.00000  Copy: 13370.939 MiB/s
4       Method: DUMB    Elapsed: 0.03875        MiB: 512.00000  Copy: 13212.562 MiB/s
5       Method: DUMB    Elapsed: 0.03825        MiB: 512.00000  Copy: 13384.921 MiB/s
6       Method: DUMB    Elapsed: 0.03842        MiB: 512.00000  Copy: 13326.393 MiB/s
7       Method: DUMB    Elapsed: 0.03824        MiB: 512.00000  Copy: 13389.471 MiB/s
8       Method: DUMB    Elapsed: 0.03798        MiB: 512.00000  Copy: 13482.554 MiB/s
9       Method: DUMB    Elapsed: 0.03800        MiB: 512.00000  Copy: 13472.266 MiB/s
AVG     Method: DUMB    Elapsed: 0.03842        MiB: 512.00000  Copy: 13328.023 MiB/s
0       Method: MCBLOCK Elapsed: 0.04010        MiB: 512.00000  Copy: 12767.761 MiB/s
1       Method: MCBLOCK Elapsed: 0.04006        MiB: 512.00000  Copy: 12782.105 MiB/s
2       Method: MCBLOCK Elapsed: 0.04010        MiB: 512.00000  Copy: 12769.035 MiB/s
3       Method: MCBLOCK Elapsed: 0.04013        MiB: 512.00000  Copy: 12759.807 MiB/s
4       Method: MCBLOCK Elapsed: 0.03992        MiB: 512.00000  Copy: 12824.366 MiB/s
5       Method: MCBLOCK Elapsed: 0.04024        MiB: 512.00000  Copy: 12724.923 MiB/s
6       Method: MCBLOCK Elapsed: 0.04009        MiB: 512.00000  Copy: 12770.946 MiB/s
7       Method: MCBLOCK Elapsed: 0.03997        MiB: 512.00000  Copy: 12810.569 MiB/s
8       Method: MCBLOCK Elapsed: 0.04027        MiB: 512.00000  Copy: 12715.758 MiB/s
9       Method: MCBLOCK Elapsed: 0.03997        MiB: 512.00000  Copy: 12808.646 MiB/s
AVG     Method: MCBLOCK Elapsed: 0.04008        MiB: 512.00000  Copy: 12773.304 MiB/s

interpret#

According to man mbw:

-t <number>
      Select tests to be run. If no -t parameters are given the default is to run all tests. -t0: memcpy() test, -t1: dumb (b[i]=a[i] style) test, -t2: memcpy() with arbitrary block size

So

  • MEMCPY means memcpy()

  • DUMB means b[i]=a[i] style

  • MCBLOCK means memcpy() with arbitrary block size

explain#

  1. MEMCPY is slowest. I believe because of blocksize 8 (bytes). I get similar results with mbw 512 -t2 -b8.

  2. It’s interesting that DUMB is faster than MCBLOCK with default block size (whatever that is). I guess it’s because of compiler optimizations.

try#

Increasing block size improves performance - up to a point:

$ for i in 1 2 4 8 16 32 64 128 1024; do echo "$i: $(mbw 512 -q -n1 -t2 -b $i | rg ^AVG)"; done
1: AVG  Method: MCBLOCK Elapsed: 0.53046        MiB: 512.00000  Copy: 965.196 MiB/s
2: AVG  Method: MCBLOCK Elapsed: 0.26510        MiB: 512.00000  Copy: 1931.354 MiB/s
4: AVG  Method: MCBLOCK Elapsed: 0.12078        MiB: 512.00000  Copy: 4239.077 MiB/s
8: AVG  Method: MCBLOCK Elapsed: 0.05672        MiB: 512.00000  Copy: 9026.003 MiB/s
16: AVG Method: MCBLOCK Elapsed: 0.04311        MiB: 512.00000  Copy: 11876.870 MiB/s
32: AVG Method: MCBLOCK Elapsed: 0.04184        MiB: 512.00000  Copy: 12237.386 MiB/s
64: AVG Method: MCBLOCK Elapsed: 0.04000        MiB: 512.00000  Copy: 12799.040 MiB/s
128: AVG        Method: MCBLOCK Elapsed: 0.04045        MiB: 512.00000  Copy: 12658.541 MiB/s
1024: AVG       Method: MCBLOCK Elapsed: 0.04078        MiB: 512.00000  Copy: 12556.406 MiB/s
re: The Coming Loop | Armin Ronacher’s Thoughts and Writings