Another Tabor benchmark:
E-UAE JIT SysInfo:

45.79 Mips (2.40x A4000/40), 3.42 MFlops


For comparison
PowerBook G4 1666 MHz with E-UAE JIT Beta 4 (compiled by Fab):
81.83 Mips (4.29x A4000/40), 14.30 MFlops

Note: Results vary across E-UAE versions (eg. Beta 5 gives 34 Mips in MorphOS), however Tabor should be in raw emulation (non FPU) speed comparable to Pegasos 2 G4.

Source: amiga.ikirsector.it, ppa.pl

Last edited by pavlor on 07-Nov-2015 at 12:02 PM.

@pavlor: on Linux/Debian PowerPC (or PowerPCSPE)?

@cdimauro

Not that it matters in this case (except FPU emulation), but it is SPE.

@pavlor

But most 680x0 code does not contain a lot of FPU instructions, naturally if you emulating integer operation you use integer operating's to do so, I expect mostly just bench-marking integer operations.

And the SPE SIMD instructions won't be used unless program is optimized to use this instructions.

I other words I don't think it proves anything.

"Tabor" can possibly run well, under optimal conditions. But under conditions where it has to emulate FPU and not using SPE SIMD is not optimal at all.

Last edited by NutsAboutAmiga on 07-Nov-2015 at 01:07 PM.
Last edited by NutsAboutAmiga on 07-Nov-2015 at 12:56 PM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@pavlor: I've found nothing on both sources. Can you provide some link? Thanks

@pavlor

Quote:

...PowerBook G4 1666 MHz with E-UAE JIT Beta 4 (compiled by Fab): 81.83 Mips (4.29x A4000/40), 14.30 MFlops... ...Beta 5 gives 34 Mips in MorphOS...

You seem to like to present MorphOS results negatively, when the second benchmark (the one that mops up the Tabor's results with dated hardware) would appear to be run under it.
Unless Fab has suddenly felt the need to use OS4 or Linux instead of the OS he usually works with.

And, do you really think throwing out a Linux spe bench is all that valid if the OS4 solution will be based on emulation?

Further, was that bench produced using one core or two?

Again, not to lose sight of this, I am not anti OS4 I just don't care for the idea of running it on hardware that can be bested by other hardware that is almost ten years old.

Last edited by iggy on 07-Nov-2015 at 01:32 PM.

@iggy

Quote:

You seem to like to present MorphOS results negatively,

No, not at all! I tried to point out that these results aren´t directly comparable.

Quote:

And, do you really think throwing out a Linux spe bench is all that valid if the OS4 solution will be based on emulation?

Mips results reflect pure integer code, so yes it is valid.

Quote:

that can be bested by other hardware that is almost ten years old.

Of course. Success or failure of Tabor depends on its end user price.

@iggy

Quote:

if the OS4 solution will be based on emulation?

Well for program that has not been compiled for the CPU it, it has too, for program compiled for CPU target, you might not need to.

What I think we end up with is multiple versions of programs, and more time has to be wasted on optimizing for different CPU targets. And programs compiled for SPE won't run on normal CPU's like G4/G3 and PA6T, unless this CPU's emulates SPE.

This is not like if your missing one CPU/FPU instruction, this is a radical different CPU.

For example the CPU instruction on this CPU, does not use float registers, it use the GPRS to store and hold float's.

Last edited by NutsAboutAmiga on 07-Nov-2015 at 02:09 PM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@pavlor

Sorry, all your replies are completely valid.
In particular, the price issue is very important.

@NutsAboutAmiga

Yes, we have repeatedly re-hashed this.
If compiled for this cpu, there is far less of of a performance difference.
I imagine some programmers will write with this in mind.
Having an application test for what fpu is available shouldn't be that hard (although there ought to be a simple flag incorporated into the OA).
Then you would just have to have two sets of alternative fpu instruction paths.
It not like fpu code makes up a large percentage of most programs.

Last edited by iggy on 07-Nov-2015 at 02:15 PM.

@NutsAboutAmiga

Quote:

NutsAboutAmiga wrote: And programs compiled for SPE won't run on normal CPU's like G4/G3 and PA6T, unless this CPU's emulates SPE.

That's impossible for G4 and PA6T, since SPE instructions use space from the Altivec opcode space (and some from the "string" instructions; here even G3s might be affected).


@pavlor: there are not sources for such "new". So, doesn't make sense to debate with possibly invented numbers.

@cdimauro

Quote:

there are not sources for such "new".

Try harder.

https://v8.googlecode.com/svn/data/benchmarks/v7/run.html

V8 Benchmark Suite - version 7
===========================================
Sam460ex Ubuntu Mate 16.04 -

midori 0.5.11

Score: 41.3
Richards: 26.8
DeltaBlue: 19.2
Crypto: 101
RayTrace: 63.9
EarleyBoyer: 3.06
RegExp: 25.9
Splay: 227
NavierStokes: 140
========================================
Sam460ex Ubuntu Mate 16.04 -

Qupzilla 1.8.6

Score: 91.3
Richards: 61.8
DeltaBlue: 34.8
Crypto: 108
RayTrace: 92.9
EarleyBoyer: 192
RegExp: 31.3
Splay: 246
NavierStokes: 151
========================================

Sam460ex AmigaOS 4.1 FE

Odyssey

Score: 29.3
Richards: 10.6
DeltaBlue: 7.18
Crypto: 51.9
RayTrace: 32.1
EarleyBoyer: 75.1
RegExp: 11.3
Splay: 60.8
NavierStokes: 84.3
========================================
V8 Benchmarks
==============================================
Tabor Ubuntu 12.04 non-smp kernel

Midori Midori 0.4.3

Score: 35.4
Richards: 104
DeltaBlue: 64.8
Crypto: 2.53
RayTrace: 29.6
EarleyBoyer: 192
RegExp: 50.8
Splay: 107
NavierStokes: 4.71
===============================================
Tabor Debian 8 PowerPCSPE chroot non-smp kernel

Qupzilla 1.8.6

Score: 139
Richards: 97.0
DeltaBlue: 75.0
Crypto: 133
RayTrace: 163
EarleyBoyer: 339
RegExp: 41.4
Splay: 374
NavierStokes: 165
===============================================
Tabor Debian 8 PowerPCSPE chroot smp kernel

Qupzilla 1.8.6

Score: 158
Richards: 105
DeltaBlue: 90.7
Crypto: 143
RayTrace: 179
EarleyBoyer: 391
RegExp: 48.6
Splay: 422
NavierStokes: 198

Last edited by Spectre660 on 07-Nov-2015 at 03:58 PM.

_________________
Sam460ex : Radeon Rx550 Single slot Video Card : SIL3112 SATA card

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: there are not sources for such "new". Try harder.

I don't find them. Is it that hard for you to provide such links, since you seem to know them?

@cdimauro

Quote:

I don't find them.

It is not that hard...

Quote:

Is it that hard for you to provide such links, since you seem to know them?

Not at all...


Some clue: On both sites, numbers I mentioned were posted as screengrab. That is why Google probably found nothing for you.

Enjoy your search!

@pavlor: I've NOT used Google, but directly accessed to the sites.

OK, I take note that you don't want to provide sources for such numbers.

Last edited by cdimauro on 07-Nov-2015 at 04:40 PM.

@cdimauro

Quote:

That's impossible for G4 and PA6T, since SPE instructions use space from the Altivec opcode space (and some from the "string" instructions; here even G3s might be affected).

Can be done with JIT, as that sort interpreter with a compiler stage. However, I don't won't give people fall hope, it’s a lot of work.

Last edited by NutsAboutAmiga on 07-Nov-2015 at 06:05 PM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@Spectre660

Interesting scores.
Thanks for including a Webkit based browser (outside of Odyssey).
It make the SAM460 comparisons easier to grasp.


OK..after checking out Midori I have a question.
Is there a version that works outside of the Linux variant?

Because versions 0.5.10, 0.5.11, and 0.5.100.5.11 all seem to have issues under Windows.

In the case of the first two it appears to be a missing library (and no, that is not likely to be my problem).

And the last 'portable' version, like so many Linux ports, appears to be a massive pile of dlls without a decent instillation batch file (at least none that I could locate).

Last edited by iggy on 07-Nov-2015 at 06:27 PM.

@Spectre660

well..on integer that board seems REALLY faster than Sam460ex...nice!
Really curious why A-Eon wont choosed the standard fpu variant...

I really wait for the price and the becnhmarks on OS4.1 whrn its available...

_________________
Simone"Tuxedo"Monsignori, Perugia, ITALY.

@cdimauro

http://amiga.ikirsector.it/forum/viewtopic.php?f=39&t=18792&start=30

latest page post

_________________
Simone"Tuxedo"Monsignori, Perugia, ITALY.

@Tuxedo

Quote:

Really curious why A-Eon wont choosed the standard fpu variant...

Price, e500mc based CPUs cost about twice what the e500v2 based CPUs do.

In fact, in quantity, they are more expensive than the low end e5500 cored CPUs (which are priced similarly to the e500v2 processors).

BUT, we have been over this, the choice was made awhile ago.
It is not changing.

Doesn't surprise me that a Frescale processor would outperform an AMCC CPU at integer calculations though.

And, since Freescale was spun out of Motorola (and inherited the 68K), it feels more comfortable using their products (and they have had the best support in the industry since we were using 8 bit CPUs).