@Karlos

Well, that was different to wiki. what a crazy design to require cpu intervention like that

_________________
It's not the question that's the problem, it's the problem that's the question.

@SHADES

the main problem in my view was commodore and its management and the lack of any longterm plan where to go. Commodore was active in a lot of markets like PC and the home computer market with amiga and C64. C64 even was the main cash cow for them for some time. That all splitted attention and rare resources, f.e. they wasted lots of money with PC, a market where they were too small to compete. Managers and CEO did not use and understand the most valuable technology they had, the amiga. They not invested enough in R&D of their technologies but spent a lot of money on useless and clueless managers. Every new manager stopped the projects and started new one in a new direction. With a clear vision where to go and concentration of all resources (marketing, R&D) on the most promising technology they had (amiga) they "might" survived. But even then it is not certain that the company survived. The base of the demise was in the 80s when most of the amiga developers left because of frustration. That should have been a warn signal to the management but was not because commodore management did not understand the market and where it goes. For some time they sold amigas because the technology was ahead of competition but that changed in 90s. And then they had nothing left, no money and no technology.

Last edited by OlafS25 on 09-Feb-2023 at 10:15 AM.

@SHADES

Probably the truth is somewhere between, but in retrospect, I don't think Akiko c2p was well thought out. Putting it inside the display controller would require a complete redesign so I can see why they decided not to go that route given where things were headed. When you consider the round trip read/write bandwidth of chip memory from the 020 alone, even if the Akiko was infinitely fast and had no additional overhead or bus interaction at all, the performance is limited.

I don't have the figures to hand exactly but I'd say around 3MB/s is a best case chip to chip longword copy for the stock 020, and that assumes the loop fits in the CPU cache. So, even if you had a copyspeed c2p implementation, the highest possible frame rate you could get there is about 48fps. That figure assumes that there is nothing else happening other than the c2p. Which doesn't make a very good game: you actually want to draw something too!

However for Akiko we have many additional bus cycles on top for writing and reading the chip registers so that 3MB/s goes down.

The latest dev build of Alien Breed 3D 2 now has instrumentation that reports the milliseconds per frame draw ans c2p time. From it we can see that Kalm's c2p is genuinely running at copy speed based on a fast to chip bandwidth of around 4.5MB/s for 040/060 class machines with local fast memory. Interestingly, on the PiStorm, where the fast to chip speed is lower, the effect on frame rate is more obvious. Despite drawing the frames more quickly the overhead of the C2P time immediately dominates.

And here is where ppcamiga's argument dies in darkness once again: the C2P routine being used is already copyspeed. Even if there were chunky pixels in chip memory, the chip ram itself is too slow and the result would be identical. You would still want to render in fast ram and copy that finished buffer to chip because you absolutely never, ever want to have to draw on uncached memory, especially when doing transparency and other effects that require reading the framebuffer. The fast to chip copy bandwidth limits the performance even if you stick the fastest 68K imaginable under it.

At least it's fair to say Gunnar fully understands that.

_________________
Doing stupid things for fun...

@Karlos

Quote:

Interestingly, on the PiStorm, where the fast to chip speed is lower, the effect on frame rate is more obvious. Despite drawing the frames more quickly the overhead of the C2P time immediately dominates.

So are the Vampire V4 machine/accelerators better than the PiStirm32 for Alien Breed 3D 2 because they give higher Chip to Fast Ram speeds and hence higher FPS?

_________________
"Art challenges technology. Technology inspires the art."
John Lasseter, Co-Founder of Pixar Animation Studios

@BigD

the newest V4 have all onboard including ram and chipset. RPi uses the original slower ram and chipset so as long a game depends on the chipset V4 should be faster. V2 (that worked similar to PiStorm regarding chipset) should be more comparable to PiStorm.

@BigD

It's more nuanced than that. Essentially the chip mem bandwidth for the PiStorm today will limit the maximum framerate, whereas the vampire doesn't have that limitation. However as scene complexity increases the PiStorm will just eat that up and continue delivering the same or similar framerate whereas I expect the vampire to vary more.

This doesn't matter much because you can "adjust" the maximum frame rate now, so you might choose to restrict the vampire to 25fps to cut out the worst variance.

Once RTG is done, it's over to who has the fastest 68k execution and memory bandwidth.

Last edited by Karlos on 09-Feb-2023 at 11:21 AM.

_________________
Doing stupid things for fun...

As an addendum, it's also possible the PiStorm chip memory bandwidth will improve but I expect a lot of it is the faff of interfacing everything together.

_________________
Doing stupid things for fun...

@Karlos

It’s also possible PiStorm will… go places. It already works with Minimig 1.8x/1.9x that has 68000 socket, but maybe one day someone designs a “PiStorm” board that simply plugs into any raspberry pi, with DRAM and FPGA for AGA minimig core and perhaps some legacy DB9 ports and optional analogue video out for non-RTG video. A “stand-alone” PiStorm Amiga ;)

Last edited by kolla on 09-Feb-2023 at 02:50 PM.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@kolla

Do you recall there was some modified AGA system that pretty much fit into a 5.25 inch drive bay? The designers had radically improved the chip ram speed. I don't recall the details though.

_________________
Doing stupid things for fun...

@Karlos

Yes, the Access system.

@Karlos

060 was released after Commodore bankruptcy.
So it is still valid Commodore bankrupt because AGA has not chunky pixel.

In 1992 developers on pc/mac/atari do not have to waste time on writing c2p procedures.
Rest of world just use 256, hi or true color in chunky pixel mode.
AGA was too slow and too outdated I'm happy I don't have to use it anymore.

@ppcamiga1

Are you completely ****ing retarded, or are some parts missing?

It's obvious to every actual developer here and probably to plenty of non developers too, that once you reach the write bandwidth limit of the chip ram, it ceases to matter if you are using chunky pixels or planar. We have seen, categorically on real hardware, that Kalm's 040+ C2P routines take around 18ms to convert a 320x256 8-bit display. If you do the arithmetic, that's 81920 bytes converted in 0.018 seconds, equivalent to 4.5MB/s. Which happens to be the same as the Fast2Chip copy benchmark.

I'd draw you a picture if I thought it would help, but chunky 2 planar is not the limit to performance on 040 processors or higher. Which are exactly the sort of CPU you will already be using of you want to play 320*256 8-bit style doom clones

Once and for all, it's the chip ram write bandwidth that is the final determining factor for the frame rate of your FPS game. It's too slow. You should be able to relate to that.

To benefit from chunky pixels properly we'd want faster write speed from the CPU to go with it. This is why AGA outperforms many Zorro 2 graphics cards that have chunky pixels because they have even slower buses to contend with.

Clear enough?

Last edited by Karlos on 09-Feb-2023 at 09:14 PM.

_________________
Doing stupid things for fun...

@Karlos

Quote:

Clear enough?

Dear Karlos, apparently you forgot you are talking to a troll who is having a lot of fun now because of your reply. He can reply to you with the same bullshit he is writing every time, most likely using copy and paste.

Please, do not feed the troll...

@michalsc

I know he's a troll. I just like to vent occasionally.

_________________
Doing stupid things for fun...

@Karlos

Quote:

Do you recall there was some modified AGA system that pretty much fit into a 5.25 inch drive bay? The designers had radically improved the chip ram speed. I don't recall the details though.

I think you may be mistaken. The Access is claimed to be 2.3 times faster than an unexpanded A1200. That's the same as an A1200 with fast ram and Access had fast ram soldered to the board.

@Rob

No, I specifically remember that Chip RAM bandwidth was increased too. I can't recall the numbers now but it was significant. I'll have to try and find it.

_________________
Doing stupid things for fun...

@michalsc

szulc I wrote you many times what you should do instead of trolling
instead of trolling you should spend time on making zune working
it is still not compatible with mui 38 from 1997

@Karlos

I'm not interested in how fast is c2p on 68060.
Commodore never use 68060.
68060 come after Commodore. So it is NG.
I you switch cpu to not used by Commodore
just use something better like ppc and use graphics card.
68060 was and is too expensive and too underpowered.
In midle 1998 68060 card for a1200 cost only few % less than cheapest ppc card for a1200 plus bvision. So I buy ppc because 040 with gfx card was faster and I get ppc extra.

In winter 1996 I sell first time my a1200 because upgrade to 68030 cost as much as switch to pentium 75 MHz.

I'm not interested in c2p copy speed on 68040 or 68060.
It was too little too late.

@ppcamiga1

What you are or are not interested in is wholly irrelevant. Commodore could not have switched to PPC and punky chixels in 1992. Chunky pixels may have existed but the PPC 601 was not released until 1993 and, in hindsight, it was absolute shit. They were the in thing in our department at university at the time until people realised they ran all the scientific software we needed much slower than the 040 machines they were supposed to replace. And most of that software did not have a PowerPC native replacement.

Commodore did intend to move to HP RISC which was somewhat proven at the time. We had HP work stations running chemical simulation software powered by the same processor line and they were on a completely different level of performance.

You switched your A1200 in 1996 to a P75? Your loss. A year later (I think) I upgraded mine with the BlizzPPC with 240MHz 603, which unlike the 601, wasn't complete turd. I had a lot of fun with it. Even more with the BVision later. I ran on two monitors, one on AGA the other on VGA.

I had no real need to switch to a PC. I drank the PPC kool aid for some time, but it soon became clear it wasn't the right choice, as PPC only systems began to emerge, riddled with flaws and sky high expensive.

Last edited by Karlos on 10-Feb-2023 at 09:23 AM.

_________________
Doing stupid things for fun...