Good Day,

Something to take notice of: " http://eab.abime.net/showthread.php?t=111805 ", A nice improvement to all of the custom chips!!

Chris

@QuikSanz

Yes, very cool. I like the idea that you might one day be able to build a totally new machine without hunting for new-old-stock custom chips. Maybe even something Commodore had never considered. Between projects like this and people creating new motherboards, the potential for innovation is high.

_________________
"Knowledge is power. Power corrupts. Study hard, be evil."

Some things that caught my eye,

2x and 4x Blitter speed and Copper prefetch (non blocking).
non-blocking CPU cycles for up-to double the CPU bandwidth.
AGA 2x and 4x display modes, double playfield and HAM8.
Paula 2x and 4x audio and disk speeds.
AGA on an OCS machine.

This is Progress
Chris

Anyone know when these are shipping?

_________________
Sent from my Quantum Computer.

@QuikSanz

Quote:

QuikSanz wrote: Some things that caught my eye, 2x and 4x Blitter speed and Copper prefetch (non blocking).

This doesn't look realistic. From the project page: https://github.com/nonarkitten/amiga_replacement_project/wiki/Willoe-v0.5-(ReAgnus)

Willoe adds 2x and 4x fetch modes to every DMA channel. All of them.
This means a blitter that can run twice or four times as fast. A copper that can run twice or four times as fast.

So, it's all about the having the possibility to fetch more data for both Blitter and Copper. However this could only help caching some data to reduce accessing the memory for fetching data which is already available.

Another important thing, data might be misaligned with 32-bit and 64-bit fetches. So, you fetch 64-bit of data, but the pointer passed to the Blitter is referencing only the last 16-bit, so you cannot make use of all fetched data but only to a small part of it. It means that there's no gain here.

However this doesn't mean that the Blitter has 2x or 4x speed. In fact, it still processes data 16-bit at the time, for each ColorClock.

The Blitter needs to work at 2x or 4x frequency to make use of all fetched data, or be changed to process 32 and 64-bit data. But there's nothing on the above page mentioning it. So, I assume that they work the as usual.

Last, but not really least, on games usually there are small blits, which often are 16 or 32-bit for the width. It means that if, for example, 64-bit are fetched, then only 16 or 32-bit of it might effectively used by the Blitter, because graphics might be interleaved (so, the following 16-bit belong to the next bitplane, the other to next one, etc.).

Summing it up, the Blitter cannot be 2x or 4x faster only because there's the possibility now to fetch much more data: this could only help a bit caching, hence leaving some DMA slot free for other peripheral (Copper, CPU).

Regarding the Copper, there are similar considerations, because it looks exactly the same as the original one. The increased fetch can only help on same cases, because, for example, it doesn't need to wait for accessing memory if the data is already there.

However it still works internally using 2 or 3 DMA slots / ColorClocks, so its speed doesn't change from the PoV. It can improve a bit because it doesn't stall adding more wait states when it needs to fetch memory, which is good, but not 2x or 4x.

All of the custom chips can DMA from the 14Mb left on Agnus, this has to be what makes AGA possible, I think.

Chris

@cdimauro

Quote:

cdimauro wrote: However this doesn't mean that the Blitter has 2x or 4x speed. In fact, it still processes data 16-bit at the time, for each ColorClock.

You are assuming that's so. But a redesigned blitter could work faster.

Just in case you missed it;
Quote:

The on board RAM is fast enough to perform a 64-bit read per 7MHz clock cycle without caching. Regardless of the FMODE, Willoe will always read all (aligned) 64-bytes from the address requested if it is not already cached and add this to the cache. Thus, in 1x and 2x modes, Willoe automagically uses as little as 25% or 50% bus time.

The RAM is on the Willoe board (not the motherboard) and is always read 64 bits at a time. That means a 4x speed blitter could process up to four 16 bit words per 7MHz clock cycle.

If this works it will be awesome - the AGA chipset we should have got! More importantly, we will be able to build new Amigas (and keep old ones going) without being stymied by the shrinking supply of original chips. 2MB Agnus chips are selling for ridiculous prices now, which makes this enhanced replacement even more attractive.

@bhabbott

I welcome the clone AGA chipset update for my A500 rev6A.

_________________
Ryzen 9 7900X, DDR5-6000 32 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32lite/RPi 4B/Emu68)
Amiga 500 (rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Hammer

Same.

I hope this continues development into new boards with no 16bit limitations in the future.
I really hope she gets support for it too. She should.

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

@SHADES

It would be a perfect fit for A1200+ project!

Edit: updated link
http://www.amigaclub.be/projects/amiga1200plus

Last edited by pixie on 01-Dec-2022 at 12:40 PM.

_________________
Indigo 3D Lounge, my second home.
The Illusion of Choice | Am*ga

This will be interesting to see on my A2000. It should provide a much snappier response, even the buss could be packed a bit better...

@SHADES

32-bit 386DX capable PCs were able to upgrade their slow IBM EGA/VGA into fast VGA/SVGA clones e.g. ET4000AX.

Upgrading OCS/ECS Amigas with a full 32-bit 68K CPU with 32-bit Fast RAM to AGA would follow the PC's upgrade path.

Better late than never.

Lacking a cost-low AGA upgrade path for the full 32-bit CPU-equipped Amigas OCS/ECS install base has caused the "starting from ground zero" effect.

Ditching Amiga 500 with GVP A530 (68030/68EC030) or PPS 040-500 (Progressive Peripherals & Software's 68040) accelerators, Amiga 2500/030 and Amiga 3000/030 are wasteful.

Last edited by Hammer on 01-Dec-2022 at 03:25 AM.
Last edited by Hammer on 01-Dec-2022 at 03:24 AM.

_________________
Ryzen 9 7900X, DDR5-6000 32 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (rev 1D1, KS 3.2, PiStorm32lite/RPi 4B/Emu68)
Amiga 500 (rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

My G-Force 030 combo should compliment it nicely...

Chris

God Jul

Last edited by noXLar on 07-Dec-2022 at 09:21 AM.

_________________
nox's in the house!

These are most interesting new for long time. I little skeptical that they actually materialice as a buyable product??

@pixie

Gosh, if only that CF card was a properly buffered USB3or2 or remained PCMCIA.
Nice job though. Huge effort.

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

@SHADES

Well there's an USB for mouse and joystick, it's something...
Beside you already have these kind of CF Cards for regular Amigas.

Is that really slow using the Amiga chipset for controlling the cf card?
I actually find it kind of nice, looks like a floppy... on the wrong side of the Amiga though! xD

_________________
Indigo 3D Lounge, my second home.
The Illusion of Choice | Am*ga

Quote:

Hammer wrote: 32-bit 386DX capable PCs were able to upgrade their slow IBM EGA/VGA into fast VGA/SVGA clones e.g. ET4000AX.

ISA Bus VGA card benchmarks
Write speed on 11MHz ISA bus (37% overclocked), AMD K5-PR133 CPU:-
TVGA9000B 1.3 MB/s
Cirrus Logic GD5422 4.4 MB/s
ET4000AX 5.5 MB/s
S3 928 5.5 MB/s

A good ISA bus VGA card was faster than a crappy one, but none were that fast. The A1200 does 7 MB/s.

Quote:

Upgrading OCS/ECS Amigas with a full 32-bit 68K CPU with 32-bit Fast RAM to AGA would follow the PC's upgrade path. Better late than never.

It's never too late for the Amiga. We're just getting started!

@pixie

Yeh, it's slow.
CPU intensive too.
Maybe it would be better with Rene's chipset but yeah. We new stuff with less bottlenecks or better conectivity would be welcome. to adapt a CF card to an AMIGA isn't exactly difficult. I don't get it.
I guess the BUS runs are all 32bit where they can be. Perhaps if it was FPGA, it could be all re-designed better. No point on the old legacy chips when they can't be re-made with wider interfaces for connection.
Then again. The quad pumping stuff. Just really smart, that Rene.

I dunno, I remember AGA with blitter and copper coprocessors in Alice, are still working on 16-bit, that may be internal on Alice not traces on the board. Getting too long to remember all the details.

I just have pcmcia cards already, and USB 3/2 chipsets would be so much faster if done right. Although, drivers would need to be written as well.

It just opens up so much more scope to get things on when you apply such a well adopted spec like USB.
CF cards are still niche, like older cameras and embedded board. CF instead of IDE maybe but buffer it and attach to DMA would be nice. Not CPU polling modes.

Last edited by SHADES on 08-Dec-2022 at 09:00 PM.
Last edited by SHADES on 08-Dec-2022 at 09:00 PM.
Last edited by SHADES on 08-Dec-2022 at 08:58 PM.
Last edited by SHADES on 08-Dec-2022 at 08:58 PM.
Last edited by SHADES on 08-Dec-2022 at 08:57 PM.

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

@SHADES

He says "The IDE interface is buffered using the otherwise unused Bus Buffer in Budgie"
Still the board is open source, maybe someone can pick it and improvite it further. One interesting thing IMO is that it has no CPU nor memory on it, it's completely modular and dependent of accelerator cards you may throw at it

_________________
Indigo 3D Lounge, my second home.
The Illusion of Choice | Am*ga