@crumb
Excuse me, how have you got these numbers?
Give a look here:
http://list-archive.gin.cz/amiga/0010/msg00426.htmlthere are results of many speed tests. With AGA I think your numbers are wrong.
In fact the AGA bus runs at 7.19Mhz, which is a theoretical 28.8MBytes sec.
(with AGA 1 access=4 bytes, with OCS 2 bytes only hence you have to halve all
numbers).
Unfortunately the cpu is only allowed to acces half of these cycles, which lower
it to 14.2MB/sec. But sadly no CPU can do a write in chip for each cycle.
It depends a lot on what are DMA channels doing but with a standard 320*250*8
screen they don't interfere too much. Since the CPU have to syncronize its clock
with the 7.19Mhz clock more or less it write 1 every 2 cyvles, which is
7.1MB/sec. In practice, as these test show it's a little bit less, more or less
6.8MB/sec (for AGA machines). It's completely false that writing is 4 times
slower than reading, they have the same speed. Maybe you got confused thinking
to write 1 byte at the time, but good c2p as the Azure's on write 1 long at a
time.
Anyway these tests also show that on EACH AGA machine, regardless of the
CPU you have more or less a 6.8 MBsec writes to chipram. That is what I call
a *clear*, *fixed* limit.
With gfx boards things as you can see are different: the conclusions of the long
test are (I consider not-overclocked results):
Card Orginal Overclocked
CV3D 7.2 Mb/sec -> 10.1 Mb/sec
PIV 9.3 Mb/sec -> 12.4 Mb/sec
CV64 13.3 Mb/sec -> 18.6 Mb/sec
well it seems to me that THERE IS a difference! And these are all ZIII boards!
Then there is CVPPC and Bvision which I think are different because they are
directly connected to the CPU with a custom bus (in fact they require CybPPC /
BlizzardPPC). And now, with A1 and Peggy, there are also PCI and AGP cards!
I guess there will be more differences.
So it is a complete mess, each card is different.
You say the bottelneck is the CPU. Maybe, it depends on the effect.
There the situation is less clear, but anyway I think with AGA you have
more stability. (BTW in this whole post with AGA Amiga I was speaking of
680x0 CPU. I only code 680x0, no PPC fo me).
It is true that also the (classic) Amiga can
have a wide range of CPU. But basically (with AGA machines) we only have:
1) 020 at 14Mhz (1200)
2) 030 at 50Mhz (many boards)
3) 030 at 25Mhz (4000/30, not many)
4) 040 at 25Mhz 4000/40
5) 040 at 40Mhz some boards but very rare
6) 060 at 50Mhz CybStorm
In practice, when you code a demo you choose one of these, usually 1) or 2)
and use it as the limit. And since the speed access to chipram is the same
regardless of the CPU (only very small differences, as you can check) many
effects look the same on all CPU. With RTG you have all these possibilities,
and since here the speed is not limited by the chipram, Practical differences
are BIGGER. Then you have 603, 603e, 604, 604e, G3 and G4 in many clock
variations. maybe in the future even more CPUs...
I think with RTG you can have 90% optimization, not more. With AGA 99.9%. :-)
But if you have fun doing 90% optimization, you are fine with RTG!
See Ya!
The Dark Coder