Sorry, dnetc is sort of spammy, but for clarificiation:

A1XEG4-933 OS4 update 3
RC5-72: [41,788 keys/sec]
OGR-P2 [223,574 nodes/sec]

From Chain-Q, Peg2/MOS 1GHZ G4:
RC5-72: [2,410,575 keys/sec]
OGR-P2: [2,938,373 nodes/sec]

So, MOS 68k emulation here is apparently also emulating an '060, while OS4 emulates an '020. This should in theory give better compatibility, but no clue as to any optimizations that may be available via an emulated '060 versus an emulated '020, which _may_ well have some effect in these numbers, but as it is, we can pretty much say that at least with respect to computations, the OS4 68k emu appears to be..quite slow.

Hmm. Doesn't OS4 NOT emulate an FPU? If so, that's possible why dnet scores are lousy. Anyone with more info feel free to share... (any old school 68k asm devs?

_________________
Are we not done with the same silly arguments and flames yet??!

@wegster
Quote:

Doesn't OS4 NOT emulate an FPU? If so, that's possible why dnet scores are lousy.

It does emulate an FPU:
5.RAM Disk:> cpu
System: PPC750(G3) (emulated MC68020/FPU) (INST: Cache) (External Cache)

Anyway since dnetc benchmarks (esp. RC5-72) more require heavy integer calculations, than float, i doubt lack of FPU would kill the performance. It's simply the difference between static and dynamic emulation. I'll turn of Trance on my MOS later, and do a test with MOS's static emulation, when i'll be at my Peg. I'm curious now. :P

_________________
MorphOS, classic Amiga, demoscene, and stuff
"When a bridge is not enough, build a Viaduct!"
"Strip the Amiga community of speculation and we can fit every forum on a 720k floppy" (by resle)

@Chain-Q

I'm not following the thread but here i post my dnet bench on A1 XE G4 1Ghz running native+altive

Quote:

rc5-72 10.680.748 k/s ogr 24.094.077 k/s

_________________
ikir

@wegster
For the record, i made dnetc benchmarks on my good old A2000/060 also. It's a phase5 Blizzard 2060 CPU card, running at 50Mhz. Results:

RC5-72: 138.461 keys/sec
OGR-P2: 336.595 keys/sec

dnetc is heavy number crunching, so it's not well suited for interpretive emulation, so the numbers mean really nothing in bigger perspective, however i'm still a bit surprised that a plain old 060 is so much faster than interpretive emulation on a 800Mhz CPU for this task.

Last edited by Chain-Q on 13-Jan-2006 at 01:34 AM.

_________________
MorphOS, classic Amiga, demoscene, and stuff
"When a bridge is not enough, build a Viaduct!"
"Strip the Amiga community of speculation and we can fit every forum on a 720k floppy" (by resle)

@Chain-Q
Quote:

t does emulate an FPU: 5.RAM Disk:> cpu System: PPC750(G3) (emulated MC68020/FPU) (INST: Cache) (External Cache)

Hmm. I could have sworn either no FPU emulation or there were know issues, petty sure the current/standard recommendation is to install software on OS4 with '020, no FPU, will look into.

Quote:

Anyway since dnetc benchmarks (esp. RC5-72) more require heavy integer calculations, than float, i doubt lack of FPU would kill the performance. It's simply the difference between static and dynamic emulation.

Thanks, you're right of course (FPU vs int ops). Yeah, please post back the other numbers. If nothing else it will give some sort of ballpark when JIT is available, or a comparison.

_________________
Are we not done with the same silly arguments and flames yet??!

@ikir

Quote:

I'm not following the thread but here i post my dnet bench on A1 XE G4 1Ghz running native+altive

Yep, those are a bit higher 1.X mill keys/second for rc5-72 per second using the PPC native client than my G4-933, which looks right in line.

The other numbers were from running the 68k dnet client, which thus is run under emulation...

_________________
Are we not done with the same silly arguments and flames yet??!

@wegster

dnetc v2.9010-495-CTR-05051421 for AmigaOS (OS 4.0pre, 68K).

[Jan 13 03:03:39 UTC] Automatic processor type detection found
a Motorola 68020 processor.
[Jan 13 03:03:39 UTC] RC5-72: using core #0 (MH 1-pipe 68020/030).
[Jan 13 03:03:59 UTC] RC5-72: Benchmark for core #0 (MH 1-pipe 68020/030)
0.00:00:16.69 [41,970 keys/sec]
[Jan 13 03:03:59 UTC] OGR-P2: using core #1 (GARSP 6.0 68020).
[Jan 13 03:04:19 UTC] OGR-P2: Benchmark for core #1 (GARSP 6.0 68020)
0.00:00:17.83 [186,412 nodes/sec]

This with 7450/51 at 800 mhz.

_________________
Site admins are people too..pooff!

@tomazkid

Quote:

tomazkid wrote: (Benchmarks from Dnetc 68k for example)

That's not a benchmark, that's maybe a way to see if there's something wrong with your system if the results deviate too much from what's expected on the respective CPU! :)

I suppose nobody's ported e.g. lmbench to MOS/AOS/et al. yet?

Edit: Crud, my too quick and too tired visual cortex just noticed you were talking about 68k dnetc and emulation. OK, dnetc would be a benchmark then. How good or useful a benchmark I wouldn't know. Nightynight.

Last edited by Seehund on 13-Jan-2006 at 04:19 AM.

_________________
Oh, bother.

@wegster

Here are my results for dnetc_68k without Trance:

Ram Disk:dnetc_68k> dnetc_68k -cpuinfo

distributed.net client for AmigaOS Copyright 1997-2005, distributed.net
RC5-72 68K assembly by Malcolm Howell and John Girvin
Please visit http://www.distributed.net/ for up-to-date contest information.


dnetc v2.9010-495-CTR-05051421 for AmigaOS (OS 4.0pre, 68K).
Please provide the *entire* version descriptor when submitting bug reports.
The distributed.net bug report pages are at http://www.distributed.net/bugs/

Automatic processor identification tag: 68060
name: Motorola 68060
Estimated processor clock speed (0 if unknown): 0 MHz
Number of processors detected by this client: 1
Number of processors supported by this client: 128
Ram Disk:dnetc_68k> dnetc_68k -benchmark

distributed.net client for AmigaOS Copyright 1997-2005, distributed.net
RC5-72 68K assembly by Malcolm Howell and John Girvin
Please visit http://www.distributed.net/ for up-to-date contest information.


dnetc v2.9010-495-CTR-05051421 for AmigaOS (OS 4.0pre, 68K).
Please provide the *entire* version descriptor when submitting bug reports.
The distributed.net bug report pages are at http://www.distributed.net/bugs/

[Jan 13 05:16:15 UTC] Automatic processor type detection found
a Motorola 68060 processor.
[Jan 13 05:16:15 UTC] RC5-72: using core #2 (MH 2-pipe 68060).
[Jan 13 05:16:37 UTC] RC5-72: Benchmark for core #2 (MH 2-pipe 68060)
0.00:00:18.73 [66,268 keys/sec]
[Jan 13 05:16:37 UTC] OGR-P2: using core #4 (GARSP 6.0 68060).
[Jan 13 05:16:59 UTC] OGR-P2: Benchmark for core #4 (GARSP 6.0 68060)
0.00:00:18.81 [277,273 nodes/sec]
Ram Disk:dnetc_68k>


This was Pegasos II with G4 at 1000MHz running MorphOS.

However benchmarks are not valid because autodetection choosed 060 optimized cores.

-----------------------------------------------------------------------------
Same machine, no JIT, but overriding autodetection from performance settings:

Ram Disk:dnetc_68k> dnetc_68k -benchmark

distributed.net client for AmigaOS Copyright 1997-2005, distributed.net
RC5-72 68K assembly by Malcolm Howell and John Girvin
Please visit http://www.distributed.net/ for up-to-date contest information.


dnetc v2.9010-495-CTR-05051421 for AmigaOS (OS 4.0pre, 68K).
Please provide the *entire* version descriptor when submitting bug reports.
The distributed.net bug report pages are at http://www.distributed.net/bugs/

[Jan 13 05:34:07 UTC] RC5-72: using core #0 (MH 1-pipe 68020/030).
[Jan 13 05:34:28 UTC] RC5-72: Benchmark for core #0 (MH 1-pipe 68020/030)
0.00:00:18.91 [62,156 keys/sec]
[Jan 13 05:34:28 UTC] OGR-P2: using core #1 (GARSP 6.0 68020).
[Jan 13 05:34:50 UTC] OGR-P2: Benchmark for core #1 (GARSP 6.0 68020)
0.00:00:19.00 [273,950 nodes/sec]
Ram Disk:dnetc_68k>

_________________
Amiga Developer
Amiga 500, Efika, Mac Mini and PowerBook

@itix
Thanks itix!

Just as a summary: ([b[all test runs below run using dnetc 68k client, thus under emulation[/b])

From Wegster, A1XEG4-933 OS4 update 3
RC5-72: [41,788 keys/sec]
OGR-P2 [223,574 nodes/sec]

Itix, 1GHz Peg2/MOS (release or dev/any difference?) using 68020 emulation.
RC5-72: [62,156 keys/sec]
OGR-P2: [273,950 nodes/sec]

Itix, same system, '060 emulation:
Jan 13 05:16:15 UTC] Automatic processor type detection found
a Motorola 68060 processor.
RC5-72: [66,268 keys/sec]
OGR-P2: [277,273 nodes/sec]

Using JIT/MOS:
From Chain-Q, Peg2/MOS 1GHZ G4: (released MOS or dev build?)
RC5-72: [2,410,575 keys/sec]
OGR-P2: [2,938,373 nodes/sec]

So, it looks like the '060 emulation buys _some_ percentage of speed increase on MOS, although percentage-wise it's pretty minimal over '020, and might be within the tolerance of multiple runs, but the 68k emulation in MOS, at least in this case is faster, by just under 1.4x for RC5 (adjusted for speed difference of 1GHz vs 933Mhz), and ~1.14x faster (again, adjusted for speed difference) in OGR..

Not terrible by any means IMO, but look at the speed diff with MOS' JIT enabled

Anyways, this is not a real comparison of 'oh, X is so much faster than Y' really, as dnetc is a number cruncher only, so while important, it's certainly not the 'whole picture.' I don't know enough about 68k or asm to even begin to make a guess as to how relevant this is likely to be (as in similar operations as used for dnetc vs other apps), but there are certainly other parts in each OS. Note- I'm not trying to diminish the difference here, just point out that random benchmarks, even moreso when done without analysis and understanding the test, can be misleading to someone casually reading.

Having said all that as a 'fair disclaimer,' of course it'll be re-run under OS4 JIT when we get it (actually, probably without JIT as well, as things could have changed as well since ud3)

RE: lmbench - not that I'm aware of, looked for it whwnever it was did the ramspeed tests initially and came up blank. Might be an easy port or not, but would be nice to have.

_________________
Are we not done with the same silly arguments and flames yet??!

@wegster
Quote:

From Chain-Q, Peg2/MOS 1GHZ G4: (released MOS or dev build?)

Dev build, but i don't think it these numbers changed by more than fractions of percentage since 1.4.5, if at all.

Maybe it would be a nice idea to setup a simple webpage with all these benchmarks.

Last edited by Chain-Q on 13-Jan-2006 at 10:25 AM.

_________________
MorphOS, classic Amiga, demoscene, and stuff
"When a bridge is not enough, build a Viaduct!"
"Strip the Amiga community of speculation and we can fit every forum on a 720k floppy" (by resle)

@wegster

Quote:

Itix, 1GHz Peg2/MOS (release or dev/any difference?) using 68020 emulation

It is a development version but static emulation havent changed over years.

_________________
Amiga Developer
Amiga 500, Efika, Mac Mini and PowerBook

@Chain-Q/itix RE: dev build differences
Ok, good enough for me then, just to get full info..

Quote:

Maybe it would be a nice idea to setup a simple webpage with all these benchmarks.

Umm. I think I mostly refrain from comment

I've dealt with SPEC before, and am not a _huge_ fan of 'unqualified benchmarks.' Meaning benchmarks that people take to mean something they don't, without having that info put in front of them (means to understand specifics of what was tests specifically, and it's relevance).

If someone felt like it, they could certainly duplicate this off of AW, but IMO it's the discussion that's half of the value. Plus, I get to learn stuff I haven't thought about since college and find out how much I thought I knew, and don't, along the way

_________________
Are we not done with the same silly arguments and flames yet??!

@Seehund

to aw.net


Quote:

Crud, my too quick and too tired visual cortex just noticed you were talking about 68k dnetc and emulation.

Yes, the thread is about 68k benchmarks under the MorphOS 68k emulation and OS4 68k emulation.
OS4-PR update #4 is coming, with Petunia, so I thought it would be fun (and maybe interesting too ) , to see some numbers, and compare them.
And of course native 68k benchmarks from Amiga Classics are most interesting too.

The Dnetc benchmarks are, as any other benchmarks, not to be taken too seriously, but gives some idea of how well the emulation performs.


*edited typo (again)

Last edited by tomazkid on 13-Jan-2006 at 01:20 PM.

_________________
Site admins are people too..pooff!

@tomazkid

I did some benches comparing Peg1, BlizzardPPC, WinUAE and 68040. Site is in German, but numbers and configs are clear anyway.

http://via.i-networx.de/bench_en.html (English)
http://via.i-networx.de/bench.html (German)

List is vey short, but gives a brief estimate between the 68k, MOS on the Blizzard and Peg and UAE.

edit: added english link

Last edited by Zylesea on 16-Jan-2006 at 10:29 PM.

_________________
My programs: via.bckrs.de
MorphOS user since V0.4 (2001)

@Zylesea

Re: Your link: Straight babelfish translation
Drop me a PM if you have corrections to the translation and I will fix.
--------------------------------------------------------------------------------
Bench mark values f?Pegasos, WinUAE, Amiga

Because there are again and again questions about the performance of the Pegasos with MorphOS compared with WinUAE with AOS or 68k-Amigas, some reference values are here listed.
During the 68k JIT emulation under MOS 1.4.5 on the 603e processor is to be noted that this does not f?JIT emulation is well suitable, since the L1 Cache is small and L2 Cache does not exist
The list extended...

Benoit:
Standard almond bread in 1024x768 pixels Aufl?ng

Pegasos1 - G3/600: 0.47s (PowerUp) | 2.50 - 2.62s (68k)

WinUAE PIII/650: 4.20s

WinUAE Celeron 2.4: 4.09 (strong diagram errors in the expenditure)

A1200: 68k(AOS): 48.70 | 68k Emu (MOS 0,4): 109.22s | MOS 0,4 PowerUp: 1.87s

A1200MOS: 1.87s (PowerUp) | 8.37s (68k Emu)

Maxon Cinema
Scene glass letter 640x480, Rendering without announcement.

Pegasos1 - G3/600: 54 - 49s (JIT) 218s (NO JIT)

WinUAE PIII/650: 344s

WinUAE Celeron 2.4: 138s

A1200: 68k (AOS): approx. 1620s | 68k Emu (MOS 0,4) approx. 2220s

A1200MOS: 677s (JIT)

Non removable disk performance (tested with master ISO von Asimware)

Pegasos1 - G3/600: VIA IDE DEVICE - 10328 KB/s with max gate dia. moon max of 80GB 7200 RPM 2MB Cache, approx. 18% CPU load

A1200MOS: SCSI Blizzard DEVICE - 2156 KB/s with quantity Atlas ii 4,5 GB, 7200 RPM??? Cache, 15% CPU load
X-Surf IDE DEVICE - 1490 KB/s with IBM DNJA 13,5 GB, 7200 RPM, 512 KB Cache (?), approx.. 30% CPU load
CPU load in the enterprise (determines with task manager)

Pegasos1-G3/600: Idle ~0.9% || Songplayer mp3-Replay (128 kbps) ~ 6.7%

A1200MOS: Idle ~4% || Songplayer mp3-Replay (128 kbps) ~40%
Configurations:

Pegasos1 = Pegasos1 April1, G3/600, 256 MT RAM, Morphos 1.4.5

WinUAE PIII/650: Sony Vaio PCG QR10 PentiumIIImobile (Celeron) copilot by mine 650 MHz, 320 MT RAM, WinMe.
WinUAE 0.9.91, AOS3.9 BB2, 64 MT Z-3 RAM, 4 MT chip RAM, 8 MT gfx RAM, ECS, Picasso96, NO sound, 68040 with FPU, max Cache.

WinUAE Celeron 2.4: Celeron 2,4 Ghz, 512 MT RAM, W2k, WinUAE 8,17, other Settings like above

A1200: Amiga1200 with BlizzardPPC 68040/25 and 603e/200, 2 MT chip RAM, 64 MT 32 bits nearly RAM, BVision/8MB, AOS3.9 BB2 and MOS 0.4

A1200MOS: Identical hardware Config such as A1200 however with MorphOS V 1.4.5

07-10-2005
--------------------------------------------------------------------------------

#6

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

@Chain-Q

I think it's exactly the opposite of what you write (that 7447 is somewhat better overall). Mostly because when we talk RC5 Altivec-enabled clients always go way, way faster. However, in all-round use the 750GX really shines (the FX too actually). I think if you run some low-res Quake demos (for instance) - and I choose lowres so the graphics card doesn't matter much - I think you'll see better performance from the FX/GX than the 7447 in a timedemo. Of course if we're to test that we'll need similar compiler setups, same compilation flags and so on for the Quake executable - even better would be the same machine with an exchange of cpus.

@number6

Some babelfish translations are always funny e.g. making 'fast ram' to 'nearly ram', because in German 'fast' is a word meaning 'almost' (or nearly) in English. The 'almond bread' might be a tasty thing (I know breads with walnuts which are quite tasty') but are of course Mandelbrot fractal sets.
I guess it is understandable enough for a rough view but if I find the time tonight I'll do a translation by hand.

_________________
My programs: via.bckrs.de
MorphOS user since V0.4 (2001)

@Zylesea

Yep. It also seems that babelfish saw the word "for" as "f?". Odd that.

#6

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

@Zylesea

Put a translation online, check it out at:

http://via.i-networx.de/bench_en.html

Currently my A1200 is not connected (I moved recently), when I
reconnect it I'll do more tests.

_________________
My programs: via.bckrs.de
MorphOS user since V0.4 (2001)