Am Montag, 13. August 2012, 20:55:23 schrieb Frank Steinmetzger: > Hey there > > As I mentioned in an earlier thread, I switched from 32 to 64 bit after some > convinction work done by the ML and a friend. In order to justify the > switch for myself, I made some performance comparisons. > > So, in case anyone is interested, here are my results. > > The only thing I don't really like is of course the increased RAM usage. > While the old installation took 400 MB of RAM after Login to KDE (Akonadi is > a hog), it now takes 500. The memory meter now stands always at least at > 50% (3 GB available). I will have to tune down multitasking a bit. > > > > The following items first display the command excuted (denoted by $), and > then the output of time for the command; first for 32 bit and then 64 bit. > > All tests were done on my Core 2 Duo laptop (T7200, max. 2GHz) fixed at 1 > GHz and with 3 GB of RAM. This is not a theoretical benchmark, but rather > about stuff I usually to do in my every-day computing. I excluded > compiling, because it involves more than just crunching. > > Resulting observation: there seems to be an inherent increase of about 10% > in memory throughput. I was most surprised by the performance of lilypond > and blender, two computing-intensive applications I tend to use regularly. > I wanted to do a framerate comparison of the Java-based CPU hog Minecraft, > but didn't get around to it. > > > All the following tasks were done in ramdisk to rule out HDD hindrance. > > > > $ 7z b (7zip's own benchmark function, abridged output) > > 32 bit | 64 bit > =========================================================================== > RAM size: 3037 MB | RAM size: 3013 MB > RAM usage: 425 MB | RAM usage: 425 MB > > Dict Compressing | Decomp. | Dict Compressing | Decomp. > Speed Rating | Speed Rating | Speed Rating | Speed Rating > KB/s MIPS | KB/s MIPS | KB/s MIPS | KB/s MIPS > > 22: 1487 1446 | 19039 1719 | 22: 1612 1568 | 20974 1893 > 23: 1443 1470 | 19049 1744 | 23: 1612 1642 | 20758 1900 > 24: 1499 1612 | 18854 1749 | 24: 1591 1711 | 20292 1883 > 25: 1489 1700 | 18611 1750 | 25: 1584 1809 | 20030 1884 > -------------------------------------------------------------------------- > Avr: 1557 1740 | 1682 1890 > Tot: 1649 | 1786 > > > > Various compressions of the High Voltage SID collection version 56 > (41356 files, 1416 folders, total dir size 307.676k according to du -s). > > Extract: > $ unrar x hvsc.rar > real 0m38.582s 0m38.763s > user 0m36.031s 0m36.190s > sys 0m2.523s 0m2.496s > ---> neglibible > > Repack witz p7zip, resulting archive size 54.8 MB: > $ 7za a -t7z -m0=lzma -mx=9 -mfb=64 -md=32m -ms=on hvsc.7z C64Music/ > > /dev/null real 3m0.530s 2m41.780s > user 5m22.359s 4m55.810s > sys 2m2.973s 0m3.144s > ---> 1/9 faster > > Extract from 7z: > $ 7z x hvsc.7z > real 0m24.541s 0m21.437s > user 0m19.302s 0m16.929s > sys 0m4.403s 0m4.472s > ---> 1/10 faster > > Simple taring of the directory: > $ tar cf hvsc.tar C64Music/ > real 0m1.334s 0m1.226s > user 0m0.297s 0m0.304s > sys 0m1.020s 0m0.872s > ---> ~1/10 faster > > XZing the tar, resulting archive size 54.2 MB: > $ xz -k -z hvsc.tar > real 6m26.383s 4m31.747s > user 6m23.375s 4m30.969s > sys 0m2.733s 0m0.728s > ---> ~1/3 faster > > XZing with --extreme option (about 4% smaller archive): > $ xz -e -k -z hvsc.tar > real 15m37.732s 10m39.348s > user 15m36.592s 10m38.900s > sys 0m0.977s 0m0.456s > ---> ~1/3 faster > > Packing in squashfs: > $ mksquashfs C64Music/ hvsc.sqfs > real 0m57.380s 0m44.697s > user 1m45.136s 1m20.377s > sys 0m9.059s 0m6.116s > ---> ~1/4 faster > > > > Some memory shuffling: > $ dd if=/dev/urandom of=random bs=1M count=500 > real 2m0.306s 1m49.348s > user 0m0.003s 0m0.000s > sys 2m0.292s 1m49.315s > ---> 1/12 faster > > $ cp random r2 > real 0m1.069s 0m0.917s > user 0m0.000s 0m0.004s > sys 0m1.067s 0m0.908s > ---> 1/10 faster > > > > Compile Bach's Christmas Oratorio, Tenor part, 16 pages A5: > $ lilypond wo.ly > real 0m31.430s 0m23.737s > user 0m30.711s 0m23.129s > sys 0m0.717s 0m0.592s > ---> 1/3 faster > > Compile Oratorio de Noël by Saint-Saëns, 4 voices, 16 pages A4: > $ lilypond noel.lyk > real 0m41.575s 0m26.494s > user 0m41.177s 0m25.870s > sys 0m0.390s 0m0.604s > ---> >1/3 faster > > > > Optimising a PNG (photo of Orion nebula, 1400x1050 pixel): > $ optipng -o9 Orion.png > real 0m23.491s 0m21.337s > user 0m23.465s 0m21.281s > sys 0m0.027s 0m0.008s > ---> 1/10 faster > > > Encoding a video file to x264 (1280x960, 1600 frames, no sound): > First pass: > $ mencoder bike.flv -ovc x264 -x264encopts bitrate=2000:pass=1 -nosound -o > /dev/null real 1m57.379s 1m44.500s > user 3m48.048s 3m19.728s > sys 0m0.837s 0m0.796s > ---> 1/8 faster > > Second pass: > $ mencoder bike.flv -ovc x264 -x264encopts bitrate=2000:pass=2 -nosound -o > bike.avi real 4m53.233s 4m15.533s > user 9m41.169s 8m26.660s > sys 0m0.990s 0m1.144s > ---> 1/8 faster > > > Rendering a small test scene in Blender 2.63 > (http://www.eofw.org/bench/test.blend): (file's scene defaults: 800x600, > one thread) > win32 2:59 > Gentoo32 2:57 > Gentoo64 2:09 > ---> > 1/3 faster
so all in all you got performance improvements you had to spend several hundred of dollars for just through recompiling. Should give you food for thought. Oh and the ram? Ram is cheap. Get yourseld 8gb. Costs as much as a good lunch. Or a couple of beers on friday night. -- #163933