Aros68k viability on real Amiga hardware

[Sinphaltimus]

AROS 68k

I understand that the goal is to run on 68k whether that is:
An Original Amiga.
An Original Amiga with upgrades (Any accelerator you want - including Vampire)
Emulation (pick any host PC)

Is that a correct assumption to make?

The same version of AROS 68k should run on everything mentioned.
Sure, you will need to develop drivers of course.
Sounds like it works fine with emulated drivers.
Does that mean it runs fine on original hardware also?
As for Vampire, I suppose it's a matter of writing all the SAGA drivers and such?

Am I understanding this correctly?

[grond]

Originally Posted by Sinphaltimus

Does that mean it runs fine on original hardware also?

Not as well as it should. "Slower than molasses on a winter day" was a description I heard.

As for Vampire, I suppose it's a matter of writing all the SAGA drivers and such?

The accelerated SAGA-driver has had a great impact on usability of AROS on the Vampire. It went from unbearably slow to pretty much usable. The AROS IDE-driver needs fixing. It used to only work on A1200 and has been fixed to also run on the A600 and A4000. However, it is still very slow and currently reaches only 1.1 MB/s as compared to the at least 3.5 MB/s on the same hardware running AOS. So there is still much room for improvement. The slow IDE speed is probably the most important single remaining factor that makes AROS perceivably slower than AOS on the Vampire. Of course, on WinUAE things are totally different as it just uses the underlying OS infrastructure and thus needs no real driver but only wrappers. This is also why a thread "Aros68k viability on real Amiga hardware" was created in the first place...

[Sinphaltimus]

Thanks for that. Perfectly clear now.

[Marlon_]

Originally Posted by grond

Not as well as it should. "Slower than molasses on a winter day" was a description I heard. The accelerated SAGA-driver has had a great impact on usability of AROS on the Vampire. It went from unbearably slow to pretty much usable. The AROS IDE-driver needs fixing. It used to only work on A1200 and has been fixed to also run on the A600 and A4000. However, it is still very slow and currently reaches only 1.1 MB/s as compared to the at least 3.5 MB/s on the same hardware running AOS. So there is still much room for improvement. The slow IDE speed is probably the most important single remaining factor that makes AROS perceivably slower than AOS on the Vampire. Of course, on WinUAE things are totally different as it just uses the underlying OS infrastructure and thus needs no real driver but only wrappers. This is also why a thread "Aros68k viability on real Amiga hardware" was created in the first place...

From apollo-team irc channel last night:

[23:14:38] ( wawa) on internal ide of an a4000/060/50 i have 2528k/s.
[00:16:34] ( wawa) 000 unfortunatelly result of the very same cf on a 1200 is worse: 1309kb/s

And today:

[14:05:12] ( wawa) interestingly rscp gives me 1416kb/s on the same cf on the same amiga 1200 under os3
[14:05:22] ( wawa) never thought it was that slow
[14:05:40] ( wawa) so aros is oly like 10% slower in comparison

So, they're working on it!

[wawa]

no, i simply checked on it with the same tool the vampire team is using and it occures that aros ide (ata) driver speed is about to be on pair with the original os3 counterpart.
the penalty might be in a range of 10%..

[grond]

Originally Posted by wawa

aros ide (ata) driver speed is about to be on pair with the original os3 counterpart.
the penalty might be in a range of 10%..

Perhaps this is the case on relatively slow hardware. With a Vampire the speed comparison is something like 3.5 MB/s with AOS vs. 1.1 MB/s with AROS. Or that's at least how I understood it. ShK should know the detailed numbers.

[wawa]

i cant review or confirm it. on irc gunnar suspected that the apollo ide fix may have influenced this.

[grond]

Originally Posted by wawa

i cant review or confirm it. on irc gunnar suspected that the apollo ide fix may have influenced this.

Yes, this is very well possible. It seems that the fix is just curing some symptoms. It works well when running the core at 78 MHz (x11) but not so well at 85 MHz (x12). Thus, it looks like it is timing sensitive which is not a good sign.

Simo did a testrun with RSCP (128KB blocksize) which showed 0.984 MB/s with AROS's ata.device as opposed to 2.9 MB/s with AOS's scsi.device running on the same hardware, an A600 equipped with a v600.

[ShK]

Here's vid of that quick test: http://apollo-accelerators.com/files/shk/AROS_RSCP.mp4

Aros ROM is aros-amiga-m68k-ext + aros-amiga-m68k-rom joined to the 1MB rom. It contains IDE fix by Mike Ness. Nightly build ROM works about 1 times of 10 (or less) with the A600 IDE. The faster the core, the worse IDE works.

[wawa]

we have tested with the default 512kb instead of 128 but perhaps it doesnt matter. on amigas i have (a1200, a4000) aros provides similar results as the original scsi.device.