Pegasos & AmigaOne benchmarks.

[Alkemyst]

http://personal.inet.fi/cool/pekosbil/a1benchmarks.htm

[LaBodilsen]

Lets just start by saying the same as Ole E. have already said several times on ANN.lu

Don't just say that these benchmarks are useless, but do come up with some better app's to use for benchmarking.

Do bear in mind though, that they must not be Hardware specific, as the test setups don't have similar Harddrives and gfx adaptors.  (and CPU in the AMD case)

be helpfull, not hatefull  :-)

[strobe]

More useful benchmarks like lmbench and x11perf would be useful.

As the discussion on ann.lu points out this is very preliminary and some more volunteers could help shed some light on the differences.

I still don't know why, for example, the faster G4 is slower at FPU fourier than the slower G3.

[L8-X]

interesing results.

[JoannaK]

strobe wrote:
More useful benchmarks like lmbench and x11perf would be useful.

As the discussion on ann.lu points out this is very preliminary and some more volunteers could help shed some light on the differences.

I still don't know why, for example, the faster G4 is slower at FPU fourier than the slower G3.

That G4 /G3 difference may be beacuse people may have used different set of link libraries while building up those binaries.

It's not stated on page but I think that those Pegasos numbers are from April-1 machine.. April-2 was supposed to make some diference, but I can't be sure.
 

[downix]

Worthless benchmark, as it gives no information on the setups used.  Maybe the Pegasos had a different form of RAM than the A1 and Athlon.  Maybe they had different kernel revisions or configurations.  What cards are in each machine?  What about the hard drives, that can throw off results too.

Without this data, those benchmarks are worth less than the bandwidth used to download them.

[tonyw]

Just about the only things you can read into those figures are:

(1) raising the clock rate of a system usually increases its computing speed;
(2) a Pegasus G3 @ 600 MHz performs about the same as an A1 G3 @ 600 MHz.

Not very surprising, really.

tony

[downix]

The problem is, no data about the actual hardware used.  What video card, HD, RAM, etc.  Any and all of these can schew benchmarks off by up to 30% in any direction, depending on the benchmark.

Heck, I remember a benchmark where the hardware was chosen specifically to cripple a particular OS for an OS 2 OS comparison, causing an artificial "boost" in the sponsoring company.  That's why public disclosure of the hardware is so critical.

JoannaK wrote:

strobe wrote:
More useful benchmarks like lmbench and x11perf would be useful.

As the discussion on ann.lu points out this is very preliminary and some more volunteers could help shed some light on the differences.

I still don't know why, for example, the faster G4 is slower at FPU fourier than the slower G3.

That G4 /G3 difference may be beacuse people may have used different set of link libraries while building up those binaries.

It's not stated on page but I think that those Pegasos numbers are from April-1 machine.. April-2 was supposed to make some diference, but I can't be sure.
 

Those test (all AmigaOnes and Pegaos) were done with exactly the same executable. Yes, the Pegasos used on those test is a April-1 one...

[Warface]

That a G3 600 machine performs similar to a G3 600 is nothing surprising methinks. However, something strikes me: AmigaONE-s ran Debian, only the Pegasos had Suse, while Genesi delivers Pegasoses with Debian and Eyetech delivers AONEs with Suse...

Huhh?

[Hammer]

@Alkemyst
My Athlon XP 1800+ with ASUS nForce 2 (MCP-T) board delivers about +11,800,000  OGR: (Nodes/sec).

http://www.amiga.org/forums/showthread.php?t=1287
==========================================
Athlon XP 1800+, ASUS nForce 2, WinXP-SP1

[Feb 18 22:25:23 UTC] Automatic processor type detection found
an AMD K7-6 (Athlon XP/MP/-4) processor.
...
[Feb 18 22:25:44 UTC] OGR: using core #1 (GARSP 5.13-B).
[Feb 18 22:26:03 UTC] OGR: Benchmark for core #1 (GARSP 5.13-B)
0.00:00:16.31 [11,800,528 nodes/sec]
==========================================

http://www.amiga.org/forums/showthread.php?t=1234
Athlon XP 1800+, ASUS nForce 2, WinXP-SP1
==========================================
[Feb 19 01:15:35 UTC] Automatic processor type detection found
an AMD K7-6 (Athlon XP/MP/-4) processor.
[Feb 19 01:15:35 UTC] OGR: using core #0 (GARSP 5.13-A).
[Feb 19 01:15:55 UTC] OGR: Benchmark for core #0 (GARSP 5.13-A)
0.00:00:16.71 [12,000,449 nodes/sec]
...
==========================================
Running 24 processes in the background (WinXP).
Why just those benchmarks programs?

What about “OpenSSL”?

[Alkemyst]

@by Hammer

LOL Dont ask me i just posted the link.

[strobe]

@tonyw

Don't confuse the CPU with the overall system

[JoannaK]

Pecosbil wrote:
Those test (all AmigaOnes and Pegaos) were done with exactly the same executable. Yes, the Pegasos used on those test is a April-1 one...

Ok.  Didn't see that mentioned on Saku foorum, it's good to know.

IIRC many of those old bytemarks are so smal tests they fit into L1 cache of C3 CPU... only couple of them were affected at all of absense of L2 cache as shown on those first numbers from Gunne.

One that would show performance diffies might be real life size tests. Like doing batch of 50 2megapixel 1600*1200 sized digital images and make some basic effects (smooth?) to them and  store back to disk (in png?). Main idea is that even if software itself would it to CPU caches the data definitely would not. Jpeg would be about 500k but it expanded to 32-bit image for manipulation takes bit over 7 megabytes or Ram each.

That imagesize  is nothing biggie these days (actually 5 megapixel is more top of line and batch conversions (scaling etc) are everyday items with anyone taking pictures and using them.

I think this kind of test could be made quite simply with standard Unix tools (PPM etc..) and some shellscripting?

[bbrv]

Posted to ANN where this discussion started:

Amigaone v Peg  Benchmark : Comment 142 of 142

Posted by bbrv (212.198.0.93) on 09-Mar-2003 06:27:27

In Reply to Comment 141:
   
Good Morning...just had the chance to catch up on this thread.  The issue was NEVER G3 vs. G4.  The issue was application oriented performance -- what you could do and how well you could do it.  Processing speeds as an end in themselves can only find amusement with a crowd of integrated circuit engineers.  Detrackers and malcontents have twisted the original post to mean something different than what was stated.  Strobe's post just above is in the proper vein for a "real" discussion that begins with system performance while really "doing" something.  :-)  Gunne too has posted well here.
 
Please read our original comment again and if you are so inclined, please feel free to continue this discussion by email with us if you so desire.
 
 Thanks and best regards,
 
 Raquel Velasco and Bill Buck
 Genesi


  ;-)

Note: Stobe's post here is correct too!