@Tomppeli

I'm completely sure, yes. What was used is what was pasted in the post. There's even some leitmotiv.flv file inside.
Now if it doesn't have altivec enabled, well, that's too bad. :)
If an older version with altivec exists, it can be tested as well, though. But altivec generally makes something like 15% difference at best (and sometimes it doesn't help at all).

@Crumb

Quote:

Crumb wrote: @COBRA Quote: It would be interesting to run glQuake OS4 version on MorphOS via OS4emu, that would rule out the differences in the port itself. That would force MorphOS to use the slower MiniGL through its Warp3D wrapper. Why don't recompile blitzquake for OS4?

It depends on what it is you're trying to show, or not, as the case may be:
1. OS4 performance on Pegs vs A1
2. OS4 relative optimization on A1 vs Peg
3. OS4 on Peg vs MOS on Peg
etc etc.
4. Some combination of the above
or, of course, the one that will get threads cleanup
5. Being rude and 'my os is better than yours' type crap. (note - this isn't saying to disregard results, or debate on what's valid or not, or what is actually being shown by a given benchmark, but leave the trolling and ''told you so's" elsewhere, please)

So far, it's pretty obvious at least, OS4 graphics drivers are less optimized that those on MOS.

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

@Jakodemus

Quote:

Jakodemus wrote: @Mufa First you could tell us which version of Quake II,III you are using. Second is that you are using much weaker graphic card on pegasos. Radeon 9250 uses 64bit memory bus compared to 128bit in Radeon 9000 Pro. So, atleast use the same graphic card for both systems. Thread was about comparing MorphOS and AOS4.1 on the same hardware.

Agreed. Benchmarks can easily enough lead to the wrong conclusions, without apples to oranges comparisons.

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

@wegster

Quote:

Agreed. Benchmarks can easily enough lead to the wrong conclusions, without apples to oranges comparisons.

Comparing different hardware can be misleading.. but it can also be misleading to compare an OS and software thats not optimised 100% for its hardware, so I think both sets of benchmarks are required here.

I'd like to see someone do more benchmarks with the mentioned software, but with an A1 that has the same Radeon as Fab's Peg2.

@jahc

There are no benchmarks where Amiga 1 could outperform Pegasos 2 G4.

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

@jahc

Quote:

jahc wrote: @wegster Quote: Agreed. Benchmarks can easily enough lead to the wrong conclusions, without apples to oranges comparisons. Comparing different hardware can be misleading.. but it can also be misleading to compare an OS and software thats not optimised 100% for its hardware, so I think both sets of benchmarks are required here. I'd like to see someone do more benchmarks with the mentioned software, but with an A1 that has the same Radeon as Fab's Peg2.

I'd be somewhat surprised personally to see an A1 with the same card outperform a Peg2, including both running OS4. It's certainly possible, just unlikely.

Whether or not OS4 has been 'optimized' for A1s, is something someone like Thomas, HJF, or some of the OS4 driver devs can really only answer..and after writing software for a pretty long time, I can say there's a *lot* of code that never does get optimized. There were also some 'surprises' on the A1, for example, issues with sound and network drivers, etc...and some of the attempts to fix these issues (on the A1s) may or may not also be executed on the Pegs as well, for all we know. Only someone intimate with the OS4 source for the drivers and HAL could even begin to tell you with much certainty 'levels of optimization' done there, for either the A1 or Peg. And it's also possible the goal was 'feature complete' over optimization. So, it's all guesses regarding 'optimization' other than to say 'some or more can always be done.'

Prior tests that mostly have to do with the hardware (RAM tests for example, and number crunching) have shown the Peg2 to be generally faster than the A1s.
Regardless, you can test anything you'd like, as can others..just hopefully it will be more apples to apples in attempt to produce *valid* benchmarks, or as close to valid as possible, without improper conclusions being drawn, and without this becoming a flamefest. So far, all I've really seen is surrounding the same graphics hardware, on Peg2, for OS4 and MOS, and MOS graphics tests are faster. It's for a limited set of tests, and sure, it can like in the software being used (q2, q3), in the underlying drivers themselves, or both. Someone was working on some improved graphics drivers for OS4, perhaps we'll see some differences once those are released.

As you said, it's more likely A1 drivers are 'optimized' vs OS4 for Peg2, so if anything, I'd expect those to simply show the Peg2 to be a superior machine for the most part. Personally, I don't expect to see any cases where the A1 outperforms the Peg2, given the same OS on each, or tests that are more hardware involved vs significant overhead on the app and/or system calls. And really, at this point, it's just not that interesting.

Likewise, for the Peg2, I'd expect MOS drivers to be more optimized for the Peg2, than OS4 on the Peg2. Either way, *good* tests can give a starting point, and as each OS evolves assuming they both remain available on the same hardware), we can see what changes there, such as when new network drivers (Peg Gigabit ethernet or RTL add-in card) or graphics drivers become available.

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

@itix

I never said I expected A1's to be faster than Peg2.......... ok, I'll try and explain it again.

Edit: To put it simply.. Peg2 OS4 vs Peg2 MOS tests might less lopsided when you compare to the A1 which may be more optimised WHILE AT THE SAME TIME taking into account the hardware specs.

Last edited by jahc on 18-Feb-2009 at 07:02 AM.

@jahc
From what i've learned so far, "OS4 is optimized for A1" is not the case at all. On the contrary, *some* OS4 applications (and maybe system components) seem to miss some *generic* optimizations, which made close to no difference on A1s, but causes significant perormance loss on other hardware. Which means: the code tops out at A1 speeds, or little above, while other hardware could do a lot more.

I'm mainly referring the Articia here, which always has been, and always will be a bottleneck on these systems, while "other" hardware includes the Pegasos II, but also any recentish PowerPC Macs as well. You can call this "optimized for the A1", but i'd simply call it unoptmized, or not properly optimized. This is not the developers fault, of course, since they've got no other hardware to test with.

_________________
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

Okay. My opinion wasn't set in stone, it was just a theory. :)

But yeah, posts like this from Cobra got me thinking:
Quote:

I discovered a few interesting things while testing DvPlayer on OS4-Pegasos2. Playback actually turned out to be slower than on my microA1, and after some investigations it turned out to be caused by two reasons: 1. The version of peg2ide.device which is included on the ISO contains a bug that it apparently falls back to PIO mode with transfers larger than 32k. This affects a lot of applications because pretty much everything on OS4 uses 64k buffers, even C:Copy or Workbench copy. The bug is already fixed and the fix will be available in an update. 2. The copy routine I used in DvPlayer to transfer the frames to videomemory was reaching the limitations of the AGP bus on the AmigaOne, however on the Pegasos2 a different technique is needed to reach the maximum speed. Using a different routine gave me a significant speed increase. I think there are probably quite a few OS components (such as P96) which could be improved to give the best performance also on the Pegasos2, not only on the AmigaOne, but that takes time. The same goes for the Sam440.

-Edit: removed irrelevant waffling-

Last edited by jahc on 18-Feb-2009 at 07:34 AM.

@itix

Quote:

There are no benchmarks where Amiga 1 could outperform Pegasos 2 G4

That depends on what CPU module you're using on the AmigaOne, e.g. an AmigaOne-XE with a G4 module upgraded to a 1.267 GHz G4 by ACube will surely outperform the 1Ghz G4 of the Pegasos in many tests.

@Chain-Q

Quote:

From what i've learned so far, "OS4 is optimized for A1" is not the case at all. On the contrary, *some* OS4 applications (and maybe system components) seem to miss some *generic* optimizations, which made close to no difference on A1s, but causes significant perormance loss on other hardware. Which means: the code tops out at A1 speeds, or little above, while other hardware could do a lot more.

That's just a different way of saying the same thing, actually. For example, when I started developing the RiVA MPEG Player, I had an A1200 with a 50MHz 030, and obviously I wrote the code so that it performs as fast as possible on my system. I also had no graphics card so initially I could not optimize for those. So I could say that it was optimized for 030. Then later I got an A4000 with a 25MHz 040, and as it turned out other techniques had to be used to get the most out of that processor, because of differences in cache, etc. However the optimizations which then made it faster on the 040 did not make it slower on the 030, so then I could say it's optimized for both 030 and 040. Obviously without access to an 040 I wouldn't have been able to optimize it for 030. Then later the 060 came and again that needed different tricks to be used because the 060 had parallel execution units the 040 didn't have, so re-ordering instructions, etc. made things faster on the 060, while not slowing the code down on the 040. So then you can say it's optimized for 060 as well.

In case of OS4 and the applications available for OS4, the developers could only optimize it for the hardware which they had available. Now that there's new hardware on which OS4 can run (Sam440, Pegasos2), the different performance-critical parts of the OS and applications can be fine-tuned to make the best of each of these, and in some cases a single code can be used which gives maximum performance on all hardware, in other cases different code may be needed (especially in case of the Sam) to get the most out of the hardware. What it takes is developers playing with the hardware, try different things and see what works best on that particular system.

@jahc
The 'different technique' Cobra mentions is the same thing i'm referring to as 'generic optimization' (generic in the meaning that it's not anything platform specific, really). The 'different technique' results the same speed as the 'A1 optimized code' on the A1, while on Pegasos2 it nearly doubles the speed of some operations. This is exactly the case i've been talking about.

_________________
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)

@COBRA
Quote:

That's just a different way of saying the same thing, actually.

It is. I've never said otherwise. :)

Quote:

In case of OS4 and the applications available for OS4, the developers could only optimize it for the hardware which they had available.

That is right also, and i've also stated it with *BOLD* letters. :)

Quote:

in other cases different code may be needed (especially in case of the Sam) to get the most out of the hardware

Right, but we both know that the Sam and other hardware we've got is very different. Even CPU-wise, Sam doesn't fit into the well known line of 603->604->G3->G4, and it's even more different when you think about the integration level of the board. While A1 and Pegasos share nearly the same CPU (only differs in G4 minor revision) and they both share the same "architectural layout", along with most of other PowerPC desktop hardware.

_________________
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)

@COBRA

Quote:

That depends on what CPU module you're using on the AmigaOne, e.g. an AmigaOne-XE with a G4 module upgraded to a 1.267 GHz G4 by ACube will surely outperform the 1Ghz G4 of the Pegasos in many tests.

It would perform 1 GHz G4 only in the raw CPU performance. It is still crippled with Articia which has no chances against the Marvell chip.

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

@Fab

Thank you for your efforts in testing!

Results like this:
"DVPlayer runs 10% faster on MorphOS (emulated) than on OS4.1 when playing from disk, and 27% faster from RAM."
indicate that AOS4 need some kind of optimization round at some point. IMHO.


(As a matter of fact, am I reading those test results correctly, AOS4 is slower in everything??)



about WB being outdated ... Ambient seems to be open source nowdays, anyone tried to port it to AOS4??

Last edited by KimmoK on 18-Feb-2009 at 08:41 AM.
Last edited by KimmoK on 18-Feb-2009 at 08:38 AM.

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

@itix

You said there are no benchmarks where AmigaOne could outperform Pegasos 2 G4, so I just pointed out that this is only the case when the Pegasos2 is fitted with a faster G4 than the AmigaOne.

@COBRA
Quote:

this is only the case when the Pegasos2 is fitted with a faster G4 than the AmigaOne.

[nitpicking]Faster or equal speed, you mean. [/nitpicking]

Last edited by Chain-Q on 18-Feb-2009 at 08:44 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)

@KimmoK

Quote:

DVPlayer runs 10% faster on MorphOS (emulated) than on OS4.1 when playing from disk

Read my post in which I explained how the peg2ide.device which made it to the OS4.1 ISO for the Pegasos contains a bug and that it falls back to PIO mode with most applications. Obviosuly video playback in PIO mode will be much slower than in UDMA mode. So, just wait for the fixed peg2ide.device to be released, then a fair comparison can be made.

Cobra, can you download that test video clip from Fab's site
http://fabportnawak.free.fr/benchmark.avi
and then test Dvplayer with the fixed Peg2 ide driver?

Last edited by jahc on 18-Feb-2009 at 08:50 AM.

@Chain-Q

Quote:

[nitpicking]Faster or equal speed, you mean. [/nitpicking]

ACube can also fit a 1GHz G4 to A1 CPU modules so if somebody has one, then you could compare the Peg2 and A1XE with the same kind of CPU and OS now, but that would be a different thread, this one is about OS4 and MorphOS on the SAME hardware.