I'm curious to see some benchmarks from MorphOS.
Especially 68k emulation, and the difference between Amiga Classics with CSPPC/BPPC and Pegasos 1/2.

(Benchmarks from Dnetc 68k for example)

Last edited by tomazkid on 11-Jan-2006 at 05:11 AM.
Last edited by tomazkid on 11-Jan-2006 at 05:11 AM.

_________________
Site admins are people too..pooff!

@tomazkid

Last benchmark battle what we've made was on fYa#12, on the big screen... It was Cinema4D rendering, of some default pic (earth with stars or stg).

Petunia on uA1/750GX@800 completted the task in 16 secs.
Trance on Pegasos 2/G4@1Ghz (MOS 1.4.5, not internal beta) completted the task in 8 secs.

Rachy later explained the results as "Cinema4D uses mathlibs, which are not yet native on OS4". I've some doubt though if Cinema4D, esp. the FPU version uses mathlibs while rendering. But i haven't investigated it further.

I'll do some benchmarks after my uA1 got Upd4, you can be sure about that.

_________________
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:

Rachy later explained the results as "Cinema4D uses mathlibs, which are not yet native on OS4". I've some doubt though if Cinema4D, esp. the FPU version uses mathlibs while rendering.

Would have been interesting to have that checked using SnoopDOS or something alike.

I am also eagerly waiting to see how Petunia from Update 4 performs. Then we can do some "real" benchmarking, since there won't be much changes to the final version, I guess.

_________________
This signature is in the middle of a much needed facelift!

@tomazkid:

Once Petunia is released, I intend to do a couple of comparisons using HD-Rec.

The main reason is that, currently, my 600mHz, 128meg Peg1/MOS1.4.5 absolutely zooms compared to my 800mHz, 256meg A1/upd3.

For instance, I can have several effects running on the Peg at one time but anymore than two n the A1 and I get "not enough CPU".

I expect the A1 to outdo it once Petunia is released but we shall see...

_________________
Robert
--
A1XE G4, OS4.1. Peg1 G3, MOS 1.4.
Abel Soul - Check out our tunes on Spotify

@Chain-Q

Basically it's useless to do these tests unless you're running them on the same
hardware.

Wait for OS4 final and do some tests on a classic setup. That's a more fair test.
Because testing a Pegasos vs a µA1-C (with its lightning fast™ north bridge)
wouldn't really be fair at all.

If I remember correctly the Pegasos II had more than twice the memory speed
of a µA1-C, someone posted ages ago.

_________________
/Harry

[SOLD] µA1-C - 750GX 800MHz - 512MB - Antec Aria case

Avatar by HNL_DK!

@polka.
Quote:

Would have been interesting to have that checked using SnoopDOS or something alike.

That would mean nothing. One app could still open the mathlibs to use some helper stuff (for example to get some float values more precisely) then render 100% using the FPU only. Some '060 demos known to do that for example. They generate tables and some pre-calculated stuff using mathlibs on init, then don't touch the libs any more.

_________________
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)

@Robert

It should really. Petunia is 4-10 times faster than the current emulation.
It will truly be nice to play with when released

_________________
/Harry

[SOLD] µA1-C - 750GX 800MHz - 512MB - Antec Aria case

Avatar by HNL_DK!

@Samwel
Quote:

Basically it's useless to do these tests unless you're running them on the same hardware.

Well, i agree. However, there was much hype - here in Hungary at least - about 750GX's FPU being faster than G4's, and about Petunia's speed "because it's in assembly", while the only benchmark i know shows quite the opposite.

Quote:

If I remember correctly the Pegasos II had more than twice the memory speed of a µA1-C, someone posted ages ago.

If my memory serves me well, the Pegasos II with G4/1Ghz has uncached memory write speed of ~420-440MB/sec, while has read speed of ~210-220MB/sec, according to MorphOS's own memtest utility. That's more like the advantage of G4 though, because with G3, the memory speeds are much slower, also on a Pegasos II. What utility is there to test the memory speed of an uA1 on OS4?

Anyway, a twice-as-slow-as-trance Petunia would be still a HUGE speedup compared to the current interpretive emulation. Still, you can't blame anything on the hardware, because DVPlayer runs quite fast on uA1, and sometimes outperforms an MPlayer running on a Peg2/G4.

_________________
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
Original thread was started by me,
here

Summary of it basically:
uA1-C, G3-750GX 800MHz/1MB cache CPU:
INT READ: 188 MB/sec>
INT WRITE: 121 MB/sec
FP READ: 193 MB/sec
FP WRITE: 122 MB/sec

XE-G4-933Mhz 7557/256K cache CPU: (AOS4 pre-2, 'fixed' XE board): (Wegster)
INT READ: 205MB/second
INT WRITE: 190MB/second
FP READ: 209MB/second
FP WRITE: 415MB/second

XE-G4 7557/256K cache CPU: (Debian 2.4.26 Linux): (Wegster)
INT READ: 206 MB/sec
INT WRITE: 200 MB/sec
FP READ: 205 MB/sec
FP WRITE: 378 MB/sec

1XE-G4-800Mhz/7451/256K L2 cache, 'unfixed' XE board, AOS4 pre-2:,
ramspeed binary from os4depot:
INT READ: 202MB/sec
INT WRiTE: 210MB/sec
INTMEM: 154MB/sec
FP READ: 202MB/sec
FP WRITE: 366MB/sec
FPMEM: 156MB/sec

Pegasos 2 G4/7447-1000MHz RAM: 768MB Using OS4/os4depot memspeed binary

INT READ: 224MB/sec
INT WRITE: 237MB/sec
INTMEM: 207MB/sec
FP READ: 228MB/sec
FP WRITE: 322MBs/sec
FPMEM: 213MB/sec

Peg2, 7447 1GHz, native MOS/PPC compile:
INT READ: 226MB/sec
INT WRITE: 466MB/sec
INTMEM: 216MB/sec
FP READ: 227MB/sec
FP WRITE: 466MB/sec
FPMEM: 213MB/sec

Apple PowerMac Quicksilver, Dual G4-1GHz/7557 CPU/256L2/2MB L3, PC133 RAM, but non SMP benchmark:
INT READ: 394 MB/sec
INT WRITE: 352 MB/sec
FP READ: 397 MB/sec
FP WRITE: 460 MB/sec

PowerBook G4 (7447A 1.33GHz): (512k L2 cache?)
INT READ: 355MB/sec
INT WRITE: 375MB/sec
INTMEM: 383MB/ec
FP READ: 345MB/sec
FP WRITE: 470MB/sec
FPMEM: 440MB/sec
Cache (max)
Reading 9 GB / second
Writing 3.975 GB / Second


FreeBSD 6, 2x667Mhz PIII CPU/256K L2 cache, PC133 Registered ECC RAM, xSeries 45
00R:
INT READ: 465 MB/sec
INT WRITE: 236 MB/sec
FP READ: 465 MB/sec
FP WRITE: 236 MB/sec

3.2GHZ CPU, 1MB L2 cache, DDR2-400 RAM, Linux 2.6.10/gentoo:

INT READ: 3340 MB/sec
INT WRITE: 1610 MB/sec
FP READ: 3700 MB/sec
FP WRITE: 1612 MB/sec

Some question remains as far as reproducability with ramspeed (source available), and what was happening in the uA1-C's numbers for writes. The numbers above were taken by basically throwing out the 'in-cache' hits, which are pretty boviousl when you run the test, to use as a mostly apples to apples comparison.

The Peg2 and A1s overall (well, unsure about INT writes on the uA1-C given) are 'pretty comparable,' while similar Apple hardware does better across the boards, perhaps not too surprisingly. I added in the 1GHz PowerMac, PIII server and P4 numbers ofr relative comparison, from compiled binaries using the same version of gcc where possible, as native compilations on each system.

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

@wegster

Did you run any tests with registered versus unregistered RAM in the A1's ?

_________________
Site admins are people too..pooff!

@tomazkid
Good question. I've got a _huge_ data file with data from all my test, just to handle questions like that (and evidently because of the fact that I suck and forget things ). I'll check, but may not have written down which module was in it (If I had given a ram size in the tests, I'd know, it's either a 512MB Registered ECC stick or the 256MB stick supplied by EyeTech, which I'm unsure right now if it's reg or not).

The RAM I _currently_ run is Registered, I'll try to kick off a few runs later tonight after work and see if there are any changes.

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

@tomazkid
Oh, at least with respect to the original tests, the tests done under Linux were the same hardware, with little to no difference, so at least we're showing the test is valid for relative (hardware) comparisons, although cache sizes vary so still need to 'adjust' to a standard measurement of sorts. DrBombCrater and others more knowledgable than I piped into the original thread and clarified some aspects, so it's worth a read of the original linked thread even with the many re-posts of tests etc.

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

@wegster

I asked due to the "fact" that unregistered is usually considered to be faster afaik.
I've got an unregistered 512Mb dimm in my A1, might run some tests later.

@thread

Isn't there any Peg-owner that is willing to give some Dnetc 68k benchmarks from MorphOS ?

_________________
Site admins are people too..pooff!

@tomazkid
Quote:

I asked due to the "fact" that unregistered is usually considered to be faster afaik.

Yes, it is generally a 'fact' that's been confirmed many times, but yeah, definitely run a few but refer to the original thread as to how we were arriving at the final numbers.

As far as I'm concerned, it's not a huge deal here, but is interesting. We know the A1s aren't speed demons with respect to most things (of course, AltiVec rocks where it can be used though!), and while benchmarks can be interesting, they can easily be taken as showing 'the wrong thing,' if a test isn't doing exactly what you think, or you're using the wrong criteria for testing, etc etc..

(Yeah, I'd still like to see your test though! )

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

@tomazkid

Quote:

Isn't there any Peg-owner that is willing to give some Dnetc 68k benchmarks from MorphOS ?

Here (running on Peg2/G4):

8.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

8.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 12 15:24:23 UTC] Automatic processor type detection found
a Motorola 68060 processor.
[Jan 12 15:24:23 UTC] RC5-72: using core #2 (MH 2-pipe 68060).
[Jan 12 15:24:42 UTC] RC5-72: Benchmark for core #2 (MH 2-pipe 68060)
0.00:00:16.97 [2,410,575 keys/sec]
[Jan 12 15:24:42 UTC] OGR-P2: using core #4 (GARSP 6.0 68060).
[Jan 12 15:25:02 UTC] OGR-P2: Benchmark for core #4 (GARSP 6.0 68060)
0.00:00:16.83 [2,938,373 nodes/sec]

For the record, the same machine does approx. 3,5-3,7Mkeys/sec, in RC5-72 with native code, while it gives well above 10Mkeys/sec when using Altivec.

_________________
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
A1XEG4-933, OS4 update 3, no JIT..
68k mode, current version from dnet:

9.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 12 17:34:33 UTC] Automatic processor type detection found
a Motorola 68020 processor.
[Jan 12 17:34:33 UTC] RC5-72: using core #0 (MH 1-pipe 68020/030).
[Jan 12 17:34:52 UTC] RC5-72: Benchmark for core #0 (MH 1-pipe 68020/030)
0.00:00:16.28 [41,788 keys/sec]
[Jan 12 17:34:52 UTC] OGR-P2: using core #1 (GARSP 6.0 68020).
[Jan 12 17:35:13 UTC] OGR-P2: Benchmark for core #1 (GARSP 6.0 68020)
0.00:00:17.72 [223,574 nodes/sec]


------------------------------------------------------------------------------
Same system, current AOS-PPC version of dnetc:

9.RAM Disk:dnetc> dnetc -benchmark

distributed.net client for AmigaOS Copyright 1997-2005, distributed.net
RC5-72 PowerPC assembly by Malcolm Howell and Didier Levet
Enhancements for 604e CPUs by Roberto Ragusa
RC7-72 Altivec and OGR assembly by Didier Levet
Please visit http://www.distributed.net/ for up-to-date contest information.


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

[Jan 12 15:42:09 UTC] Automatic processor type detection found
a PowerPC 7445/7455 (G4) processor.
[Jan 12 15:42:09 UTC] RC5-72: using core #4 (KKS 7450).
[Jan 12 15:42:28 UTC] RC5-72: Benchmark for core #4 (KKS 7450)
0.00:00:16.10 [9,977,856 keys/sec]
[Jan 12 15:42:28 UTC] OGR-P2: using core #1 (KOGE 2.0 Hybrid).
[Jan 12 15:42:47 UTC] OGR-P2: Benchmark for core #1 (KOGE 2.0 Hybrid)
0.00:00:16.37 [22,493,553 nodes/sec]

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

@wegster

Wopsie daisy.. Is that twice the MOS speed I'm reading or am I misreading
the numbers?

This really shows how slow the 68k emulation is when interpreted. JIT will
make it atleast 10 times faster I think.

_________________
/Harry

[SOLD] µA1-C - 750GX 800MHz - 512MB - Antec Aria case

Avatar by HNL_DK!

@Chain-Q

Quote:

If my memory serves me well, the Pegasos II with G4/1Ghz has uncached memory write speed of ~420-440MB/sec, while has read speed of ~210-220MB/sec, according to MorphOS's own memtest utility. That's more like the advantage of G4 though, because with G3, the memory speeds are much slower, also on a Pegasos II. What utility is there to test the memory speed of an uA1 on OS4?

Well That depends on the G3 CPU. The GX is quite much faster than most of the
G4's we're talking about (up to 7447). But the bad memory controller and slow bus
(133MHz vs GX own 200MHz) slow everythig down on the µA1-C.
Excluding altivec code of course.


EDIT: about DVPlayer vs MPlayer.. Yes but DVPlayer is heavily optimized for AmigaOS
and that's not the case for MPlayer which on OS4 is quite slow in comparison.

Last edited by Samwel on 12-Jan-2006 at 11:18 PM.

_________________
/Harry

[SOLD] µA1-C - 750GX 800MHz - 512MB - Antec Aria case

Avatar by HNL_DK!

@Samwel
Quote:

Wopsie daisy.. Is that twice the MOS speed I'm reading or am I misreading the numbers?

You are misreading the numbers. 22Mkeys/sec is OGR speed, not RC5-72. I can give you exact dnetc benchmark results from Peg2/G4 running beta MOS kernel (with AltiVec support) tomorrow if you wish.

_________________
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)

@Samwel
Quote:

The GX is quite much faster than most of the G4's we're talking about (up to 7447).

No it's not. My GX/800 gives 2.7Mkeys/sec in dnetc. That's multiplied by 1.25 (to have a result from a theoretical 1Ghz version) gives 3,3Mkeys/sec, which is good, but still a bit slower than my G4/1Ghz is able to archieve, without AltiVec. (And don't forget, the GX has double the cache.) And it doesn't count IMO that the GX could go with faster buses, since the 7447 also has a 166Mhz bus mode, which is unused on the current machines.

I think it all depends on optimization. You can write code, and there is code which runs faster on the GX, but for generic cases, i still think the 7447 is somewhat better, and has more spare performance in it, thanks to AltiVec.

_________________
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)