Amiga Performance Scale

[trekiej]

Is there an Amiga Performance Scale and where do you think Natami and FPGA-Arcade will go on this scale?

[thedocbwarren]

Natomi sounds kind of neat, but wouldn't a compatible OS on a say a G4 or some-such be better?  Anyway, 68060 is faster than any classic Amiga.

Measuring performance is always tricky since you have the graphics co-processors to take into account, plus blitter, etc.

A base 68000 at 8Mhz is 1 Mip compared to 110 Mips for 68060 75Mhz.

But the original Amiga 1000 had graphics co-processors that could push pixels faster than the 68K could.  So what is performance?

[J-Golden]

Isn't that what SysInfo and those style of programs were for?

[trekiej]

Well, I knew I should have been clearer.
I need to find a sysinfo comparison chart or something similar.
I will hunt down Natami and FPGA-Arcade sysinfo shots.

[Damion]

Well, I knew I should have been clearer.
I need to find a sysinfo comparison chart or something similar.
I will hunt down Natami and FPGA-Arcade sysinfo shots.

Maybe not what you're looking for, but still interesting:

http://www.amigaspeed.de.vu/

[Kesa]

Natomi sounds kind of neat, but wouldn't a compatible OS on a say a G4 or some-such be better?  Anyway, 68060 is faster than any classic Amiga.

Measuring performance is always tricky since you have the graphics co-processors to take into account, plus blitter, etc.

A base 68000 at 8Mhz is 1 Mip compared to 110 Mips for 68060 75Mhz.

But the original Amiga 1000 had graphics co-processors that could push pixels faster than the 68K could.  So what is performance?

I agree that measuring performance is tricky as there are many different variables to consider so i think the way to approach this is to envision what it is going to be used for and start from there otherwise you may end up just confusing yourself

My 2 cents is not focusing on the technical aspect but more on what we will be using on the NatAmi and FPGA. Is there any classic software that can take advantage of all this power? I mean for an Amiga a 75Mhz 68060 is ultra fast but the 150 Mhz(?) NatAmi will make the 68060 seem slow.

Maybe another point i am trying to make is maybe the current performance measuring tools will not be relevant for the ng of classic hardware?

[trekiej]

I can go with that.
I just want to see how the other stack up in some kind of chart or graph.
Natami and Replay should be fairly fast as most everything is on chip, with some exception to other external chips.

[J-Golden]

Maybe another point i am trying to make is maybe the current performance measuring tools will not be relevant for the ng of classic hardware?

+1

I remember SysInfo's scale not going far enough when measuring certain processors.  And that was way back then!

[Franko]

Bit off topic but related in a way...

SysSpeed unlike SysInfo will let you check all the Amiga processors including some basic PPC checks, can't say that any of these utils are very accurate but they do allow you to get an idea of a particular Amiga's performance...

Don't quite think it's what you're looking for but in this THREAD (on Page 5) a while back I posted various SysSpeed test results for an A1200 with 060 board, might help if you find similar results elsewhere on the site here for other set ups to compare them to...

I'd love to see one of the betatesters of the NatAmi post some SysSpeed results in order to compare them to my 060 results...

Back on topic...

If the NatAmi when released lives up to it's claims then this would really put it at the top of any Amiga performance chart I'd reckon (even better than my Blizzard 060) not sure about FPGA-Arcade as I've never really followed that thread...

[Karlos]

A base 68000 at 8Mhz is 1 Mip compared to 110 Mips for 68060 75Mhz

Even this is a somewhat awkward comparison. The usual comment about "Meaningless Indicator of Performance Scale" aside, code has to be well-tuned for superscalar execution (bearing in mind that the two execution units are non symmetrical, the secondary execution unit can't do everything the primary unit can) to get anything like that level of performance on 68060.

[Kesa]

Franko that depends. I think what this thread is trying to say is that it wants graphs and not numbers to look at. You being a die hard Amigan should know this. On Amiga.org if it ain't a picture we ain't looking at it    :p

[Franko]

Franko that depends. I think what this thread is trying to say is that it wants graphs and not numbers to look at. You being a die hard Amigan should know this. On Amiga.org if it ain't a picture we ain't looking at it    :p

I would have drawn a graph but me big red crayon broke and that only left me the blue one, so all the results would have got mixed up, that and and I'm crap at drawing...

PS: I ain't an Amigan today, today I'm a SNESIAN cos I'm playing some old SNES RPG games... SNES RULEZ... AMIGA SUX... :p

[Forcie]

This is indeed an interesting topic.
Benchmarks are quite complicated things. What does one actually measure, and does the results make sense to compare with ones run on another architecture?

SysInfos "speed" test, for example, is quite a "stupid" benchmark. It tests very few instructions, and in a very repetitive manner. It makes more sense for testing 000-020 than the more advanced architectures. If we wanted, we could optimise the Natami 050 core to look super good in SysInfo. But this would not necessarily make much sense for achieving good real life performance.

AIBB is a lot more advanced. Maybe we should make an AIBB profile for Natami. I have not used AIBB much myself, though.

However, I think it makes more sense writing "new" benchmark suites for the new classics like FPGAArcade and Natami. There are so many aspects which get lost when running benchmarks made for an A500. Not saying the old ones should not be run too, though

Here is a small benchmark suite written by Rune of the Natami Team. It runs code optimised for 030, 060 and N050 in sequence. Source included.

http://www.4shared.com/file/me92duH4/sp_benchmark_11april_.html

[Matt_H]

I think it's AIBB that is/was considered the gold standard in benchmarking. A Natami module for it would be great to have.

SysInfo is good for some quick'n'dirty numbers, but you shouldn't read too much into it. Its tests are not very comprehensive, and like most software from the 3.1-and-earlier era it doesn't know about anything newer than an 040!

[Ral-Clan]

I think it's AIBB that is/was considered the gold standard in benchmarking. A Natami module for it would be great to have.

I agree.  I would love it if a Natami beta board tester could release an AIBB module for the Natami.  That way I could run tests again my own Amiga hardware (end even emulator setups) and see just how it compares.