jfpoole

I've been doing a bit of work lately on cross-platform benchmarks, and I thought it might be interesting to share some of the results I've collected so far. I've run the benchmarks on three different machines (one PC, two Macs) and here are the results. Higher numbers, as always, are better.

Dell Dimension 4500 (P4 2.26GHz)

<pre>cpu (float) mandelbrot (sqrt) 800.53 megaflops<br />cpu (float) mandelbrot (nosqrt) 720.65 megaflops<br /><br />cpu (integer) blowfish (cache) 37.32 megabytes/sec<br />cpu (integer) blowfish (memory) 27.98 megabytes/sec<br />cpu (integer) emulate 6502 275.00 megahertz<br /><br />memory (stdlib) fill 575.69 megabytes/sec<br />memory (stdlib) sequential access 542.38 megabytes/sec<br />memory (stdlib) random access 658.40 megabytes/sec<br />memory (stdlib) copy 448.52 megabytes/sec<br /><br />memory (stream) copy 957.93 megabytes/sec<br />memory (stream) scale 893.63 megabytes/sec<br />memory (stream) add 1010.30 megabytes/sec<br />memory (stream) triad 1011.56 megabytes/sec<br /></pre>

Power Mac G5 (G5 1.6GHz)

<pre>cpu (float) mandelbrot (sqrt) 441.54 megaflops<br />cpu (float) mandelbrot (nosqrt) 697.13 megaflops<br /><br />cpu (integer) blowfish (cache) 32.52 megabytes/sec<br />cpu (integer) blowfish (memory) 24.07 megabytes/sec<br />cpu (integer) emulate 6502 193.07 megahertz<br /><br />memory (stdlib) fill 2.13 gigabytes/sec<br />memory (stdlib) sequential access 2.14 gigabytes/sec<br />memory (stdlib) random access 2.21 gigabytes/sec<br />memory (stdlib) copy: 1.02 gigabytes/sec<br /><br />memory (stream) copy 1.43 gigabytes/sec<br />memory (stream) scale 1.39 gigabytes/sec<br />memory (stream) add 1.51 gigabytes/sec<br />memory (stream) triad 1.50 gigabytes/sec</pre>

Power Mac G3 (G3 450MHz)

<pre>cpu (float) mandelbrot (sqrt) 49.51 megaflops<br />cpu (float) mandelbrot (nosqrt) 198.20 megaflops<br /><br />cpu (integer) blowfish (cache) 11.50 megabytes/sec<br />cpu (integer) blowfish (memory) 9.74 megabytes/sec<br />cpu (integer) emulate 6502 52.85 megahertz<br /><br />memory (stdlib) fill 284.41 megabytes/sec<br />memory (stdlib) sequential access 283.53 megabytes/sec<br />memory (stdlib) random access 288.61 megabytes/sec<br />memory (stdlib) copy 81.90 megabytes/sec<br /><br />memory (stream) copy 165.33 megabytes/sec<br />memory (stream) scale 165.04 megabytes/sec<br />memory (stream) add 168.51 megabytes/sec<br />memory (stream) triad 168.48 megabytes/sec</pre>

DBerG

As expected the G5 exceed by far in memory operations.

Chealion

jfpoole - What's the bus speed on the P4?

I was under the impression that the G5/G4s were better at integer operations while the P4 was better at floating point operations. (Or was it vice versa?)

__________________
I still remember Macnutt.

mguertin

As always though it comes down to the tools used and how well they are (or are not) optimized for the arch in question. Things like MMX/SSE/Altivec make HUGE differences! Also hardware differences (such as FSB speed, hyperthreading, chipsets used, etc) make huge differences in things like ram access. It's like comparing a Camaro to a Volkswagon

Kosh

Ooooh, lots of big beautiful numbers!

Yes, I believe you are right Chealion, I think the 970 is best at integer operations, while the P4 and I believe G4 is best at floating point operations. With the proviso that you don't take Altivec into consideration. If you put floating point ops through Altivec, then of course the 970 wins overall, I believe.

__________________
-- Mac Pro Dual 3.0 Dual Core, 4GB RAM, 150GB Raptor HD, 500GB HD, Superdrive, Radeon X1900XT, Bluetooth, Airport Exteme, Logitech Z-5500 5.1 Speakers, 30" ACD
-- PowerMac G5 Dual 2.5, 2.5GB RAM, 160GB HD, Superdrive, Radeon X800XT, Bluetooth, Airport Extreme
-- Macbook Pro 2.53 15", 4GB, 320GB HD, SuperDrive, Bluetooth, Airport Extreme

comprehab

Glad to see all those powermacs are going to good use! Hehehe

gnatsum

so long story short.

P4 is a speed demon

G5 is a beast of burden

G3 is a royal waste of time



eh?

how can i get about benchmarking my sawtooth? lol...just for fun? or is it not a matter of fun?

i wouldn't mind seeing what difference in floating point operations the "AltiVec" engine makes...

When would a person ever use a Floating Point Operation?

mguertin

I wouldn't say the G3 is a waste of time ... but it you compare it to the PC world you would be comparing it to P2 processors and the G3 will likely come up much nicer in that comparison

someone

I don't think the float point tests are fair. I believe the P4 ones are optimized for SSE. while the G5 ones are *not* optimized for Altivec.

jfpoole

The Pentium 4 has a 533MHz FSB (it's an older machine).

The floating point tests aren't optimized for any particular platform; they don't use SSE or Altivec instructions. I might tweak the benchmarks to use SIMD instructions (if available), but I'm more keen on getting the benchmark suite out the door first.