Performance of Akiko C2P on 030/50 CD32 systems

[Karlos]

Is there any good information as to the relative performance of Akiko C2P on 68030+FastRAM accelerated CD32 machines compared to well-optimised CPU-only routines on the same hardware?

[alexh]

Have a look around and you'll get some figures posted last year.

From memory I believed it showed that Akiko C2P is out-performed in SW as soon as you get fast-ram even with EC020@14MHz

But there were some unusual outliers

https://eab.abime.net/showpost.php?p...0&postcount=63

Any testing of Akiko with an 030 in the system must remember there is a bug in the 030 cache controller

https://eab.abime.net/showthread.php?t=56255
https://aminet.net/package/util/misc/AkikoTFFix

[pixie]

Originally Posted by alexh

From memory I believed it showed that Akiko C2P is out-performed in SW as soon as you get fast-ram even with EC020@14MHz

Damn! What a waste of silicon, then!

[Amigajay]

Originally Posted by pixie

Damn! What a waste of silicon, then!

Not at all! If you read the magazines at the time, it was added for no cost because they had the space on the chip to do it. Just an added bonus.

[alexh]

Originally Posted by Amigajay

it was added for no cost because they had the space on the chip to do it. Just an added bonus.

Exactly the Akiko was needed as it is the "glue" that mops up lots of A1200 motherboard functionality (Budgie + Gayle + CIAs) and it is the CD-ROM controller + NVRAM controller. It has a tremendous number of pins and that limits minimum die (chip) size. It is common for pad limited designs to be filled with extra stuff rather than left empty.

C2P must have been a very last minute addition because the Akiko includes a DMA master (for the CD-ROM) which "could have been" plumbed into the C2P had there been more time.

[Karlos]

I'm not especially looking to debate the rights and wrongs of the feature, I just want some cold hard metrics. My plan would be something like having a tight conversion loop that starts off on a dummy scan line we don't care about. At the end of that line only, we would theoretically disable the data cache and then try to use thr chip to complete the conversion. Not sure if that would be sufficient.

[alexh]

I swear I remember a thread from earlier this year (maybe late last year) that examined the various performances of C2P implementations (direct to chip mem, using fast mem with a copy to chipmem, using blitter assisted, akiko etc.) on 030, 040 and 060.

I wonder if it was in the thread that discussed the "new" chunky screen mode? (PED81C) maybe not?

I'll have a look tonight if you don't find it.

[pixie]

Originally Posted by Amigajay

Not at all! If you read the magazines at the time, it was added for no cost because they had the space on the chip to do it. Just an added bonus.

I was thinking that if Akiko went on detracting commodore from adding fast ram it would be a bad decision, but along this lines it's not actually that bad.

[Amigajay]

Originally Posted by pixie

I was thinking that if Akiko went on detracting commodore from adding fast ram it would be a bad decision, but along this lines it's not actually that bad.

Exactly, a 50% speed increase on 3D games was very good for basically nothing! Obviously compared to adding extra hardware it fails in comparison, but that's what always gets done!

Commodore would never have added fast-ram to the budget models, simply down to cost.

An '030 CD32 model (and plug in board for the 1993 model) was on the drawing board for late 1994 (possibly an updated A1200 too) but even then the Commodore papers never mentioned fast-ram, only the Hombre model planned for 1995 would include faster V-RAM.

[Karlos]

@alexh

It's the outlier that intrigues me. Perhaps the reason was the software 030 routine was not the best one, or perhaps it's the case that there are some common pitfalls in Akiko C2P to be avoided. Either way, if we are running in single-digit FPS, a 1-2fps increase from Akiko would be worth it for what doesn't seem like a huge amount of programming effort.

[hitchhikr]

From memory I believed it showed that Akiko C2P is out-performed in SW as soon as you get fast-ram even with EC020@14MHz

DoomAttack:

Amiga A1200 020/14Mhz 8Mb (Optimised 020 C2P) – 22976 realtics (3.3 fps)
Amiga A1200 020/14Mhz 8Mb (Blitter 020 C2P) – 20660 realtics (3.6 fps)
Amiga CD32 68020/14Mhz 8Mb (Optimised 020 C2P) – 18971 realtics (3.9 fps)
Amiga CD32 68020/14Mhz 8Mb (Akiko Optimised C2P) – 12872 realtics (5.8 fps)
Amiga A1200 030/28Mhz 64Mb (Optimised 020 C2P) – 17732 realtics (4.2 fps)
Amiga A1200 030/28Mhz 64Mb (Blitter 020 C2P) – 12727 realtics (5.8 fps)
Amiga A1200 030/50Mhz 128Mb (Optimised 020 C2P) – 8696 realtics (8.6 fps)
Amiga A1200 030/50Mhz 128Mb (Blitter 020 C2P) – 8296 realtics (9.0 fps)

[alexh]

Originally Posted by hitchhikr

Amiga A1200 020/14Mhz 8Mb (Optimised 020 C2P) – 22976 realtics (3.3 fps)
Amiga CD32 68020/14Mhz 8Mb (Optimised 020 C2P) – 18971 realtics (3.9 fps)

Interesting. Where did you get these results from? Did you generate them? Are these two HW configurations not supposed to be "the same"? Do they have a memory bandwidth difference or is there really a 0.3 fps variability in the test?

Someone will have to work what "Optimised 020 C2P" means. Which C2P implementation is it using? Is it really using FastRAM?

[Karlos]

I think the only practical option now is to just try it.

[StevenJGore]

Originally Posted by Amigajay

Hombre

You realise you've just turned this into a 200 page thread?

[Amigajay]

Originally Posted by StevenJGore

You realise you've just turned this into a 200 page thread?

Haha! There is certain people that can turn every thread into why the Amiga was bad/underpowered/outofdate

[hitchhikr]

Maybe you could start here: https://aminet.net/search?query=doomattack

[paraj]

Originally Posted by Karlos

I think the only practical option now is to just try it.

I know you already knew that I wish I had the hardware for testing since this seems like something that should have been "settled" long ago. C2P's haven't been changing for 020/030 for a long time, so I think we "just" need timings from one 030+fast ram user and a "good" Akiko C2P implementation. (IDK what "good" means here, but I guess at least chunky pixels coming from fast ram).

[derSammler]

Originally Posted by alexh

From memory I believed it showed that Akiko C2P is out-performed in SW as soon as you get fast-ram even with EC020@14MHz

Even if that would be true, it still does not mean Akiko is useless in that case. Note that when the CPU does C2P, processing time is wasted that would otherwise be available to execute code. It seems people often completely ignore that. The DoomAttack figures are a perfect example of that. The CPU is better used to render the 3d graphics. Even if Akiko is slower in doing C2P than the CPU would, you still get better speed in the end. It's different when the CPU would otherwise mostly be idle, but how often is that the case?

[Karlos]

Originally Posted by paraj

I know you already knew that I wish I had the hardware for testing since this seems like something that should have been "settled" long ago. C2P's haven't been changing for 020/030 for a long time, so I think we "just" need timings from one 030+fast ram user and a "good" Akiko C2P implementation. (IDK what "good" means here, but I guess at least chunky pixels coming from fast ram).

Right now, I'm going to be pitting it against Kalm's 030 routines. It seems self evident that the read/write throughput will be the limit for Akiko, but we aren't dealing with large resolution. Like you I have no HW to test on, so I'll be prototyping in UAE.

[hitchhikr]

Right now, I'm going to be pitting it against Kalm's 030 routines. It seems self evident that the read/write throughput will be the limit for Akiko, but we aren't dealing with large resolution. Like you I have no HW to test on, so I'll be prototyping in UAE.

I do have a cd32 with fast ram, i can make some tests if you need, eventually.