"New" amiga hardware

[NorthWay]

Jens has not officially killed Clone-A. Though I am not certain what stage of development it is at.

From what I have read he seemed to be more or less done, but he had some bright idea he wanted to incorporate to make it sell better/wider that was so "obvious" that he wouldn't want to reveal it so that others could do the same. Plus he is probably not sure if it is an economically viable project - I would guess it it expensive to produce.

(As for "obvious" - think Tetris: Once people had seen it they could make a version in a day or in 512 bytes and it was still great stuff. That is the kind of idea you don't want to escape.)

[NorthWay]

The result of 3.5 MB/sec is actually the best possible result on a 16bit AMIGA.
I'm not sure if this is normal to reach.
What do other cards reach on 16bit AMIGAs?

Can't speak for 16-bit (I don't remember what my borrowed 040 did), but my CS060 MK1 was consistent 3.5/7 in my 4000. It was considered a good implementation AFAIR.
I think some of the 1200 accelerators were struggling...

[NorthWay]

What do you think about this value?

I think this is great news, but I do question the validity of the values when it reports 3.17(or even 4.00) times the chipspeed of an A600.

Anything more than 1.xx from an accelerator does not make sense. A cache can eliminate the instruction fetch, and the non-bus using logic can be sped up, but the rest should be pure chipmem cycles of which you can't magically conjure any more than there already are.

(Well, I can see a possible trick in your implementation if you replace one write with a more recent one if the old hasn't yet gone out to the bus, but then we are in bad code territory and should find some other testing tool - which is interesting in itself of course.)

[NorthWay]

/\
Sysinfo reports "Chip Speed vs A600" at 7.20 on my A1200 030 50MHz.
Is it because it's AGA?

Sysinfo was never my choice of testing tool - and for good reason it seems.
There are only two bandwidth speeds in Amigas produced by C=
1) (7,160,000 / 4) * 2 giving you 3.5 megabytes per second - OCS/ECS
2) (7,160,000 / 4) * 4 giving you 7 megabytes per second - AGA

The second one requires longword access, the first one word access. The bandwidth halves for each time you halve your access width (unless the cpu is clever enough to glue two accesses together if they are close enough in time - I don't think any of them are?).

So 7.20 might be a number to compare other machines with as long as they use a consistent scale, but taken as a measure of speed it is totally FUBAR.
If you want to check your speed I suggest bustest: http://aminet.net/util/moni/bustest.lha

[NorthWay]

This is in a nutshell how the calculation of SYSINFO works.

Thanks.

So in other words it has very little to do with the chipmem speed at all really, it is a compound of cpu, cache, and chipmem speed - bad benchmark, bad!

[NorthWay]

Therefore your can up to 4 times the bandwidth of the chipmem

I missed that part - then I expect you are targeting the full 4.00 as you had in one of your builds?