Blitter vs Cpu on Aga

[sandruzzo]

I know that Aga Blitter is the same as ECS, but due to more memory bandwidth
Blitter can go full speed.

What CPU is Blitter worth? 020, 030?

[Samurai_Crow]

According to the GameSmith documentation, for making bobs it is equivalent to an 040.

[Galahad/FLT]

030 will be faster than blitter in most things, especially if real fast memory is available as well.

[sandruzzo]

Originally Posted by Galahad/FLT

030 will be faster than blitter in most things, especially if real fast memory is available as well.

030 50mhz or even 28 mhz?

[ReadOnlyCat]

Originally Posted by sandruzzo

030 50mhz or even 28 mhz?

Assuming the 030 accesses RAM using the same number of clock cycles as the 68k (which is false, it uses less), at 28MHz it will have a RAM bandwidth around four times higher than the 68k (28 = 7 * 4, 7mHz = original clock frequency).

If you run a mem copy in a tight loop which fits into the cache, this means that it is not slowed down too much by instruction fetches and will almost have this full bandwidth available for data reads and writes. This is grossly the same speed as the blitter with its four (3 read, 1 write) channels.
(If you write to chip RAM you would have to take into account that that its bandwidth is limited in addition to being used by the chips as the Galahad kitten hinted.)

Sure, this is just a back of the envelope approximation but this gives a reasonable idea of what to expect.

Hope this helps.

[sandruzzo]

@ReadOnlyCat

I'm not talking about Memory acces only, but real operation like direct copy D = A, and cookie cut blit operation, Assuming that blitter and cpu can run at full memory speed

[Samurai_Crow]

Originally Posted by sandruzzo

@ReadOnlyCat

I'm not talking about Memory acces only, but real operation like direct copy D = A, and cookie cut blit operation, Assuming that blitter and cpu can run at full memory speed

In a direct copy the 020 is the same speed if Fast RAM is available. Cookie cutting is more like an 040 unless you can avoid shifting bits.

[zero]

Originally Posted by sandruzzo

I'm not talking about Memory acces only, but real operation like direct copy D = A, and cookie cut blit operation, Assuming that blitter and cpu can run at full memory speed

Depends on the operation... The blitter can shift bits left/right faster than an 030 I think.

However, if you are reading graphic data from fast memory and only writing to chip, the 030 can be faster because fast RAM has more bandwidth than chip.

The ideal scenario would be to use both the blitter and CPU, with the CPU reading from fast memory and only doing the final write into chip.

[Photon]

"Normal" code: (max-optimized general-purpose function) Linedraw and cookie-cut should still beat the 020/030. All other operations, or small blits, say 32 words or less, use the CPU.

In special cases there will always be some tricks to pull to at least give the visual impression that the CPU has beaten the Blitter.

In all cases it will be up to screen size+depth and intelligent fighting with the DMA for the coveted slow chip RAM.

[sandruzzo]

Originally Posted by Photon

"Normal" code: (max-optimized general-purpose function) Linedraw and cookie-cut should still beat the 020/030. All other operations, or small blits, say 32 words or less, use the CPU.

In special cases there will always be some tricks to pull to at least give the visual impression that the CPU has beaten the Blitter.

In all cases it will be up to screen size+depth and intelligent fighting with the DMA for the coveted slow chip RAM.

So blitter wasn't so bad. If we had 32bit 14mhz on aga....

[TEG]

Originally Posted by sandruzzo

So blitter wasn't so bad. If we had 32bit 14mhz on aga....

[Samurai_Crow]

Originally Posted by sandruzzo

So blitter wasn't so bad. If we had 32bit 14mhz on aga....

Actually, 16 bits at 14 MHz would saturate the fast page access if buffered properly. Some parts of the AGA Alice were clocked at 28 MHz so the blitter could have been miniaturized to 8 bits for a smaller die and a single chip solution at that clock speed.

[sandruzzo]

Originally Posted by Samurai_Crow

Actually, 16 bits at 14 MHz would saturate the fast page access if buffered properly. Some parts of the AGA Alice were clocked at 28 MHz so the blitter could have been miniaturized to 8 bits for a smaller die and a single chip solution at that clock speed.

Yes mate! Blitter with few changes would had ben so powerfull!

[ReadOnlyCat]

Originally Posted by sandruzzo

Yes mate! Blitter with few changes would had ben so powerfull!

It needed more than a few changes. As has been hinted, the bus bandwidth is not that great and even a slightly faster blitter would have been hampered by that. The system needed a whole redesign rather than an evolution here or there.

[sandruzzo]

Originally Posted by ReadOnlyCat

It needed more than a few changes. As has been hinted, the bus bandwidth is not that great and even a slightly faster blitter would have been hampered by that. The system needed a whole redesign rather than an evolution here or there.

Of course. The problem is the bus bandwidth. It should be doubled on Aga machine. At least have 2 bus: one for bitplanes and another for the other chip

[AnimaInCorpore]

The Amiga Blitter haven't been improved for the Amiga 1200 for only very reason to keep up the compatibility to the Amiga 500.

However, the time since 1990 was an era for a paradigm shift. The advent of new and more powerful CPU generations wasn't a matter of decades anymore but more like years and months; so it was the "system upgrade" time. In fact, supporting obsolete custom hardware, just to justify the compatibility to older software, was simply too expensive.

The CPU was the new "replacement" for proprietary graphics devices and so the MS-DOS system had a clear advantage, especially due to the cheap hardware components and the fact that the 3D games were on the rise.

Also, in respect of this matter, having a clear but simple graphics interface was economically better than relying on old and too complicated devices only for compatibility reasons.

So in conclusion there was no reason to improve the Amiga Blitter, except to provoke problems for later generations.

[Miggy4eva]

Aga blitter should be faster than '040