multi-thread suite

[deanjo]

Hey Michael, how about putting the multi-threaded tests into a suite of their own to show the speed difference with x number of cores?

[Michael]

All of the compilation tests are able to handle multiple jobs concurrently and is done automatically... In regards to making an SMP suite, that may be a good idea. There is also the multi-thread friendly ET:QW and Quake 4 and a few others.

[deanjo]

All of the compilation tests are able to handle multiple jobs concurrently and is done automatically... In regards to making an SMP suite, that may be a good idea. There is also the multi-thread friendly ET:QW and Quake 4 and a few others.

Sunflow is multithreaded as well and if you include p7zip into your tests you would have another multithread as well (which would be really easy since it has a built in benchmark "7za b" will invoke it).

Sample output:


dean@linux:~/Desktop/phoronix-test-suite> 7za b

7-Zip (A) 4.57 Copyright (c) 1999-2007 Igor Pavlov 2007-12-06
p7zip Version 4.57 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,4 CPUs)

RAM size: 3965 MB, # CPU hardware threads: 4
RAM usage: 850 MB, # Benchmark threads: 4

Dict Compressing | Decompressing
Speed Usage R/U Rating | Speed Usage R/U Rating
KB/s % MIPS MIPS | KB/s % MIPS MIPS

22: 6213 298 2028 6044 | 93329 383 2510 9606
23: 6356 312 2074 6476 | 88663 368 2513 9252
24: 5887 306 2070 6329 | 87567 367 2526 9259
25: 6169 334 2108 7043 | 84764 370 2455 9080
----------------------------------------------------------------
Avr: 312 2070 6473 372 2501 9299
Tot: 342 2285 7886


PS there is also the multithreaded verison of bzip2 as well @ http://compression.ca/pbzip2/

[Michael]

Both pbzip2 and p7zip profiles are now in the git tree :)

[Michael]

./phoronix-test-suite benchmark multicore

is in there now.

[apaige]

I'd be interested to see if "true" quad-cores such as the Phenom have an edge over multi-chip processors such as the Core 2 Duos/Quads as far as scalability goes (i.e., how much performance is gained from using one core to using all cores). It'd be nice if the "multicore" tests ran with only one core, then two, etc... Performance gain ratios could then be computed (i.e. 1.97x with two cores, 3.72x with four cores, etc).

Obviously, scalability relies primarily on the app itself, but at least we could compare ratios between CPUs.