@Spirantho

From MorphOS user point of view PA6T is interesting chip in comparison. It is PowerPC chip and in theory MorphOS could run on that hardware, too. If you dont like you can always ignore MorphOS threads.

AROS users can provide their own benchmarks against MorphOS hardware. I dont mind that at all.

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

@itix

The problem isn't that, though. The problem is that he's not benchmarking the CPU, he's benchmarking the entire system - the OS, the CPU, the RAM , the hard disk, the graphics card.. even the installation of the OS. That's just far too many variables to make the benchmark mean anything at all.

If I give an X1000 2 GB of RAM and run everything from the RAM disk with all graphics disabled and nothing else running at all, and then give a G5 MOS machine 128MB of RAM, run 50 apps in the background, and tell it to encode from an ATA33 CD-ROM drive onto a Zip disk, then I could make it seem much slower too.

Of course that's an extreme example - but the point is these graphs don't tell you what you need to know. You can't compare two different platforms running on two different systems, using two differently optimised versions of two different ages of one program, and then say "look how much faster our one is than yours". It's ludicrous.

The whole thing is misleading. The graph is engineered to make it look like the SAM460 is dog slow, and worse than a 500MHz G4. Of course what it doesn't tell you is that if you run the non-Altivec version, you get very different results. Sure, in some circumstances, it's good to have Altivec, but the graphs are deliberately designed to hide the fact that for most things the SAM460 is much faster than the G4/500, and made up to make it look like MOS machines are a million times faster than AOS 4 machines.

Right, rant over. I still look forward to the day when this petty sniping stops though.

Won't say anything else in this thread as it's getting OT.

@itix

Quote:

Yup. Luckily there is G5 @ 2.5GHz.

Good for winter months.


I think using multiple CPU cores would be best approach. There are even multi-threaded forks of lame (and of course other similar applications). I can imagine scenarios, where PA6T 1.8 GHz (dual-core) outperforms 970FX 2.7 GHz (single core).

Benchmarks are useful. However, Piru did rather unfortunate mix of hardware/OS/benchmark versions (sadly, it was not for the first time). Now, I´m even more interested in Blender benchmark - not that much AltiVec centric as lame, MPlayer or OGR.

@pavlor

Quote:

I can imagine scenarios, where PA6T 1.8 GHz (dual-core) outperforms 970FX 2.7 GHz (single core).

And I think using multiple CPUs would be best approach: Apple Power Macintosh G5 2.7 DP

Quote:

Benchmarks are useful. However, Piru did rather unfortunate mix of hardware/OS/benchmark versions (sadly, it was not for the first time). Now, I´m even more interested in Blender benchmark - not that much AltiVec centric as lame, MPlayer or OGR.

AFAIK (since I am not 100% sure, busy with other things) Piru found out AltiVec code in lame was trashing random memory and built new version.

But what is sad in mixing hardware or OS versions? MorphOS developers are obviously running the 3.2 and Piru is testing it on his G5.

Edit: we could probably provide our own benchmarks for SAM460, too. But if our SAM460 benchmarks were slower you wouldnt be happy. If our SAM460 benchmarks were faster you wouldnt be happy with it either. If they are same then what is the point? =P

MPlayer is certainly one of the most interesting benchmarks. Everyone is using it and it is one of most important applications.

Last edited by itix on 03-Jan-2013 at 08:12 PM.

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

@Spirantho
Quote:

Sorry for the rant but I get really fed up with engineered "benchmarks" that are basically works of fiction.

No problem, I see your point and understand you. I am not a fan of synthetic benchmarks, too. If you do such benchmarks at all (even more, if you compare different OSes), they should be done with the same software and under the same conditions. I see such comparisons not as competition, more as an orientation (depending on the tasks you intend to do, not everyone is using Blender for example).

And I agree: The stupid red-versus-blue-thing shouldn´t count anymore. We are in a tiny niche-"market" with no commercial relevance anyway and have the same roots ("Classic" Amigas with AmigaOS), so no need to fight anymore.

And: As 90% of us use Amiga as a hobby these days, it doesn´t really matter, how fast you render a scene in blender. At the end of the day we should have fun with our hobby. I mean I owned an SAMflex in the past and enjoyed it, and there are people, who enjoy their Efika with MorphOS and so on. I see MorphOS as an addition. For me it will be the key to an "Amiga"-Notebook, which I wanted for years (I know there is AROS on AspireOne, but AROS isn´t on the same level as AmigaOS4 or MorphOS yet).

_________________
AmigaOne X5000/40, 2.2 Ghz, 4 GB RAM, Radeon R9 280X, M-Audio Revolution 5.1, 240 GB SSD

@itix

Quote:

But what is sad in mixing hardware or OS versions? MorphOS developers are obviously running the 3.2 and Piru is testing it on his G5.

G4 1.42 GHz lame 3.99.4 - 18s
G4 1.42 GHz lame 3.99.5 - 14s

G5 2.0 GHz lame 3.99.4 - 13s
G5 2.0 GHz lame 3.99.5? - 11s

I think differences are obvious.

Quote:

And I think using multiple CPUs would be best approach: Apple Power Macintosh G5 2.7 DP

Hopefuly MorphOS will support such configuration soon.

However, 2x970MP 2.5 GHz in newest PoweMac G5 will be still faster in both single core and (of course) multi core operations.

Quote:

MPlayer is certainly one of the most interesting benchmarks. Everyone is using it and it is one of most important applications.

Without doubt.

@pavlor

Quote:

G4 1.42 GHz lame 3.99.4 - 18s G4 1.42 GHz lame 3.99.5 - 14s G5 2.0 GHz lame 3.99.4 - 13s G5 2.0 GHz lame 3.99.5? - 11s

Hold your horses. Post #18:

Quote:

Test new Lame 3.99.5 on Mac mini G4 1.42 GHz For MorphOS there is only old version LAME 3.98.4 (slower)

And your post #20:

Quote:

I hope older version was used for G5 benchmarks. However, same speed as G5 2 GHz looks impressive.

Last edited by itix on 03-Jan-2013 at 08:22 PM.

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

@itix

If Piru can, why not me?

@Spirantho
Quote:

Spirantho wrote: Sorry for the rant but I get really fed up with engineered "benchmarks" that are basically works of fiction.

Actually, I think you just don't want to see those benchmarks. They are synthetic benchmarks and as such they have a lot of limits. But they do give nice ballpark figures that I, for one, happen to find both interesting and useful. People that owns other hardware than Piru's might update the numbers for their rig, so that the final result should be more "fair", even if I am not really convinced it is unfair right now. Denying any value to the benchmarks is indeed unfair, in my humble opinion. It just makes you look like a child covering his ears and shouting "la-la-la" because he doesn't want to hear something...

Kind regards,
Andrea

@pavlor

Because you are making OS4 users angry. You confuse them, you spread misinformation and mix different operating systems.

Lame 3.98.4 was used in old MorphOS benchmarks. Lame 3.99.5 was used in OS4 benchmarks i.e. in a post #15 it scores 47 seconds for sam460 using lame 3.99.5 and the graph was fixed accordingly.

Lame 3.99.5 benchmark results on Mac mini G4 1.42GHz running OSX scored 14 seconds. You are mixing it here with MorphOS results. There is no such result in graphs. There never was and there still isnt.

However G5 results with lame benchmark were better using generic AltiVec version than with G5 sepcific optimizations. But it is still lame 3.99.5 version (i.e. same was used on sam460/os4 and presumably on X1000).

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

@itix

Quote:

Because you are making OS4 users angry. You confuse them, you spread misinformation and mix different operating systems.

Expert opinion, I think.

Quote:

Lame 3.98.4 was used in old MorphOS benchmarks

Sure.

Quote:

Lame 3.99.5 was used in OS4 benchmarks

Not for X1000 benchmarks (Lame 3.99.5 was released in September 2012 on OS4depot).

Quote:

Lame 3.99.5 benchmark results on Mac mini G4 1.42GHz running OSX scored 14 seconds. You are mixing it here with MorphOS results.

I didn´t mentioned OS version in my results table. However, in my original post (55), I noted both MorphOS and MacOS X versions.

It was Piru, who compared old lame and newer/optimised lame in single graph without neccessary description. I hope you got my point.

Benchmarks are nothing to get upset about as long as we're sticking to single core PowerPC. I'm sure we've all got PCs and even mobile devices with more processing power available.

Besides, my BlizzardPPC is better than all your kit put together

_________________
Doing stupid things for fun...

@Karlos

Would be nice with some aros results here.

And why is not the powermac g4 500mhz tested with mplayer????

_________________
I'm an antique. Don't light my fuse

@Antique

I'm sure any SSE3 optimized x86/x64 port on a modern processor would roast 'em. It wouldn't matter which OS was used, since a task like that will be compute bound while decoding and IO bound the rest of the time.

_________________
Doing stupid things for fun...

@pavlor

Quote:

It was Piru, who compared old lame and newer/optimised lame in single graph without neccessary description. I hope you got my point.

It was you who was posting benchmark results here. The thread is at MorphZone and information was also posted to #morphos channel. Perhaps you should take a visit (irc.freenode.net) and take a look what is going on there. And while you are there maybe take a peek at #aros, too.

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

@pavlor

"G5 2 GHz is only less than 30 % faster than G4 1.42 GHz in lame benchmark."

G4 is awesome! The best Altivec so far. It will be interesting to see next Freescale Altivec product!

I think MPC6810 (or was it MPC8610, anyway) is so far the best per Mhz (in blender, IIRC).

((PA6T might be best per watt, if both cores are used, untill then...))



Other than that, people take benchmarks too seriously. They can be fun and interesting, but real life usability matters most.


Btw. has anyone tried how different optimizations affect PA6T performance. Piru pointed out that G4 version of Lame was faster than G5 version. I wonder how G5 vs G4 version perform on PA6T... (grim?)

Last edited by KimmoK on 04-Jan-2013 at 02:14 PM.
Last edited by KimmoK on 04-Jan-2013 at 08:21 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

Another update from Piru :post 33

New lame benchmark results:

2.5GHz G5: 6 seconds
2.0GHz G5: 8 seconds


Good! That is result one expect from G5.

@itix

I assume all results in new graph are from latest lame (3.99.5), except X1000. Right?

@pavlor

Quote:

I assume all results in new graph are from latest lame (3.99.5), except X1000. Right?

I think so.

http://www.amiga.org/forums/showpost.php?p=678876&postcount=55

Using lame 3.99.5 could win few seconds for X1000, perhaps?

Last edited by itix on 04-Jan-2013 at 03:39 PM.
Last edited by itix on 04-Jan-2013 at 03:39 PM.
Last edited by itix on 04-Jan-2013 at 03:36 PM.

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

considering the clockrates and coresponding benchmark results and remembering that bothsystems are using one core on double core processors imho its safe to say that pasemi/os4 and g5/mos are roughly equivalent. that means pasemi _is_ actually an g5 class cpu even if it doesnt mean that its much different from g4, other than by clockrate. that said g5 is available in higher clockrates and withmore cores than pasemi and therefore is a superior solution even if not as power efficient.

is that just? i think so.. both sides have their advantages and even x1k has some reason to be, even considering there are faster ppc systems, still its one of the fastest. right?

Last edited by wawa on 04-Jan-2013 at 03:59 PM.

@wawa

According to benchmarks I saw (and even the latest MorphOS ones) G4 1420 MHz is roughly half as fast as G5 2300. PA6T 1800 MHz (single core) is as fast as G4 1420 MHz. So one PA6T CPU (2 cores) is as fast as one G5 CPU (970FX) in multicore operations (eg Blender).

Note: There are some cases, where PA6T is faster than G4 (memory throughput) or even G5 (FFT).

Quote:

and even x1k has some reason to be

2 PCIe x16 slots. Nice if you want develop and test Radeon HD drivers.


Edit:
For comparison (and some perspective):
Assuming G4 1.5 GHz would score 4 SpecInt2006 (my rather optimistic estimate), then...
PA6T 1.8 GHz would score 4 SpecInt2006
970FX 2.3 GHz would score 8 SpecInt2006
Core 2 Q6600 2.4 GHz scores 18 SpecInt2006 (in more than 5 years old computer of my father)
Core i5-2500K 3.3 GHz scores above 40 SpecInt2006

All single core integer performance.

Last edited by pavlor on 04-Jan-2013 at 04:55 PM.
Last edited by pavlor on 04-Jan-2013 at 04:54 PM.