Steve's H5 A3.14 Build (Chicken Pie edition)

[exxos]

@Badwolf blitter and no blitter already tested.

https://www.exxosforum.co.uk/atari/last/TF536/index.htm

In part it was why I started this topic https://www.exxosforum.co.uk/forum/viewt ... =29&t=4775

[stephen_usher]

A great deal of the acceleration that the Blitter gives is not the data transfer as much as the data transformation as pixels don't directly map to words.

The 68030 performs this transformation as fast or faster than Atari's custom silicon, at least for ST modes, so it's faster. The Blitter is constrained by the RAM data bus so it can't push words any faster to/from memory.

[exxos]

By the looks of it, the blitter is still like 50% faster than a 50Mhz 030 on the blitting test in GB6. But other tests are probably limited in RAM and ROM access speeds, so not much gain there. It is a pretty interesting topic really.

[Steve]

By the looks of it, the blitter is still like 50% faster than a 50Mhz 030 on the blitting test in GB6. But other tests are probably limited in RAM and ROM access speeds, so not much gain there. It is a pretty interesting topic really.

I know that HD-Driver supports 'blitter transfer acceleration' for disk transfer speed, but I am guessing that this would only make sense for something like the Trudie and an 8mhz 68k. I am assuming that with a TF536 it would probably reduce the speed, or just plainly not work at all due to the TF536 IDE working in fast-ram and the blitter being st-ram dependant?

[exxos]

I know that HD-Driver supports 'blitter transfer acceleration' for disk transfer speed, but I am guessing that this would only make sense for something like the Trudie and an 8mhz 68k. I am assuming that with a TF536 it would probably reduce the speed, or just plainly not work at all due to the TF536 IDE working in fast-ram and the blitter being st-ram dependant?

Not really sure how blitter speeds up a HDD. It is again why I open that other post because the blitter seemingly could deal with data transfers faster than the CPU.

If the blitter was doing byte swapping then I guess it would make sense if the blitter could do it faster than the CPU. But I really have no idea.

As for Trudie, it uses hardware byteswap. But again, no idea if the blitter would be faster or not. It certainly has not been tested.

EDIT:

Unless the blitter is used in HOG mode to steal cycles away from the CPU or something :shrug:

[Badwolf]

@Badwolf blitter and no blitter already tested.

https://www.exxosforum.co.uk/atari/last/TF536/index.htm

In part it was why I started this topic https://www.exxosforum.co.uk/forum/viewt ... =29&t=4775

The results are mixed until you get to the balls-out blitter test at which point it's a knockdown win for the blitter. Not surprising really.

The leveller there, in my opinion is the fast ROM. Most of the other display tests rely on the ROM (for the fonts, etc). I'll bet the results are skewed even more in the blitter's favour if the ROM were stock.

Anyway, I think this does just emphasise that it's a myth a 32MHz processor, by itself, obviates the benefits of the blitter.

Can we please stop saying it? :P

BW

[exxos]

The results are mixed until you get to the balls-out blitter test at which point it's a knockdown win for the blitter. Not surprising really.

The leveller there, in my opinion is the fast ROM. Most of the other display tests rely on the ROM (for the fonts, etc). I'll bet the results are skewed even more in the blitter's favour if the ROM were stock.

The results could actually be a bit different with the caches turned off. If some ROM was cached during tests, it would speed them up. So the blitter could by itself be doing better than it seems. CPU and blitter can access ROM at 50Mhz (well 32mhz effectively) .

Wait until we get a 32Mhz blitter ;)

Anyway, I think this does just emphasise that it's a myth a 32MHz processor, by itself, obviates the benefits of the blitter.

Can we please stop saying it? :P

ahhh Good luck there. I think one something gets "cast in stone" you might as well forget about it... Welcome to my world ;) :lol: :roll:

[Badwolf]

The results could actually be a bit different with the caches turned off. If some ROM was cached during tests, it would speed them up. So the blitter could by itself be doing better than it seems. CPU and blitter can access ROM at 50Mhz (well 32mhz effectively) .

I think caches are fair game as they're part of the 030.

I see the differences more starkly on my Falcon board with the AltRAM turned off. Because we're comparing apples with apples -- an 030 accessing both ST-RAM and ROM over a 16bit bus at ~5MB/s except with the processor doing 50MHz instead of 16. Enabling the blitter is a big boost to desktop performance & the benchmarks.

Wait until we get a 32Mhz blitter ;)

I genuinely think if you fulfil your goal of a full 32MHz ST bus people will be utterly amazed at how effective that is above and beyond processor speed.

I think it was Sophie (neé Roger) Wilson of Acorn who singled out memory bandwidth as the biggest issue with the jump to 16 bit computers. That was partly the catalyst that brought about ARM. Rarely a truer word spoken!

Can we please stop saying it? :P

ahhh Good luck there. I think one something gets "cast in stone" you might as well forget about it... Welcome to my world ;) :lol: :roll:

Wait! Blitter's a DMA device. Perhaps I have a 'bad' DMA..! :ugeek:

BW

[exxos]

Wait! Blitter's a DMA device. Perhaps I have a 'bad' DMA..! :ugeek:

:chairsmack:

Well, everyone expected it so... :lol:

[DoG]

I think the statement comes from the PAK3/030 manual. In it, it says to deactivate/remove the blitter since it doesn't matter in those speeds. Maybe just due to conflict it states that. I don't know. PAK have it own roms however.

At some stage (mhz) the CPU would be moving bits and bobs faster then the blitter, no? Both run at 8mhz bus speed though of course as you guys mentioned to access ram. Is it at 100mhz?