View Single Post
Old 28 May 2010, 13:58   #39
Registered User
Join Date: Jul 2009
Location: UK
Posts: 74
Originally Posted by Paul_s View Post

it would be better to just recreate the A1200 PCB... like one of those GB1000 boards
Updated for 68030/just slot your custom chips in etc... far nicer for the hobbyist and you could probably get them under 100 if enough people chipped in.
it would only be temporary as parts will get unavailable and no new cases are being made.

here is more info about the changes : a quote about the DDR
Due to the way DDR memory works I think they may be using the rising signal flank of the pulse for custom chip access and the falling one for cpu accesses.

You seem to conceptionally misunderstand how DDR memory is working.

Ill try to explain it with an easy example:

With DDR you send 1 address and then a burst of data belonging to this address will be transmitted.
It would similar to like the 68040 or 68060 did burst in their 16byte cache lines.

Typical burst size is 4 words.
If your memory interface is 32bit this means every memory access READ or WRITE will transfer 128bit in one go - ALWAYS!

The AMIGA chipset originally was not prepared for bursting.
When AGA was developed a short burst of 64bit was added for the Video and Sprite DMA.
But bursting and buffering was never added for the Blitter or Audio or Copper or other DMA channels.

Finally now the NATAMI does correct this.

Now all AMIGA DMA is bursting.
And the DMA is not only bursting but also doing pipelined burst.

Example of a pipelined burst and "old school" memory interface.
Todays memory has a latency of roughly 20 clocks this means if your FPGA clocks with 200 MHz then you "old school" blitter can fetch every 20 clocks a word of 16bit.
This give you a blitter bandwidth of 20 MB/sec.
Better than old AGA but not very exciting.

Using the same Hardware you could burst every 20 clocks a 28 bit word.
To be able to do this you need to add the right buffers to your Blitter.
This gives you a bandwidth of 160 MB/sec.
That is a lot better isnt it?

Now when your DMA will also proper pipeline the burst also then you can have 5-6 burst in flight in parallel.
This will finally you a bandwidth in the order of 800 MB/sec
Now we are talking.

The NATAMI did has working AGA already 2 years ago.
For those that wondered what was changed / developed since these 2 years.
I think you know the difference know.
natami is futureproof, faster and cheaper FPGAs will come :-)

remember the estimated performance for Natami is x150 - x200 cpu speed of an A600 and a x50 DMA performance.

But one must code a program to fully take advantage of the new features.
asymetrix is offline  
Page generated in 0.05505 seconds with 9 queries