Of course, such a chip should work on ChipRAM only.

This sort of DSP, in addition to supporting some existing Amiga software, would allow things such as realtime effects (ex: reverb), wavetracing, hrtf or ambisonic sound positionning or other types of sound editing and processing.

Adding an enhanced FPU to the 68K Core or adding enhacnements as SIMD instructions to the 68K Core do make the 68K even more superior.

A 2nd 68000 is much easier to program and to support than a extra alien DSP core. Therefore if there is a desire for a dedicated Audio core the most sensible solution, in my opinion, is adding an extra 68K core.

we would need a library i think they got supplied with the cards that had them.
and using either the main MPU or a SPU is too generic, IMHO. (Sub Processing Unit)

Parse through such a library and think how would it m68k only.

oh one more thing take the high road on this one guys.

Am i right ? (i assume that question is also valid for the SIMD thread).

Why do you think a 070 would have less performance than such a DSP?

Yes, i understood that "sub cores" are copies of the main processor (050/070).

But i have some (stupids :) ) interrogations in mind (i don't know HW and i'm totally unaware of how is made a CPU).

Maybe "sub cores" don't need some specific parts that only the main processor will use (i don't know at all) ? That way, maybe "sub cores" could be enhanced with that free space and being more specialized.

If they are still 90% (random number) like the main processor, they are still easy to use like any 68k. And the enhancements will not complicate too much the use of the new functions, because coders can still ignore them and use "sub cores" in the 68k way.

But i agree that an easy to code NatAmi will still be better than a Natami who need 3 years to be tamed.

I hope that I'm understandable.

It is a generally accepted fact specialized hardware out preforms general hardware.

I am under the assumption a CPU fall under the classification general hardware, DSP's are optimized for a specific function.

So could you explain if both had the same speed and same bandwidth, why would the CPU out perform specialized/optimized hardware on a so called 'home ground advantage'?

This is simple to explain. :-)

Lets repeat our development timeline:
1) Get 68050 out of the door
2) Bring 68050E out
3) Bring 68070 (the best 68K core ever) out.

At this point the 68070 will be the best 68K CPU ever.

Now cloning the 070 to get some MAID-Cores will take a few days work only. In a few days work you can NOT develop a DSP.

This mean you have the choice between
A) The best 68K core ever as maid
B) No DSP core as the development is not done.

Developing a dedicated DSP will take long time.
Maybe 1 year, maybe 2 years.
End even after 2 years development time its not sure that the DSP will be as fast as the 68070 core.

Instead spending time on develping "island solutions" it would make much more sense in enhancing the 68070 into an 68090 ...
Any general enhancements to the 68K will benefit all cores.
The main CPU core, and all Maid-Cores.

I assume you all agree with me.

OK, I accept that the 070 will be faster than a non-existent DSP.
But, as Marcel said specialised hardware is generally faster.

Now is this prejustice or fact?

Please take the time and try to explain why you think a DSP like the propused 98000 should be faster.

> I am under the assumption a CPU fall under the classification general hardware, DSP's are optimized for a specific function.
Exactly, the instruction set of the m96k was specifically chosen to speed up signal processing, on the other hand, the m68k instruction set is not specifically tuned for signal processing / audio processing. This is like saying:
forget the 3D core, we shall put a second 68k, it will be easier.
Components should be sepecialized for their dedicated task. Otherwise, we are just going to create a 68k version of the IBM Cell processor, which is a fair design (but no more).

We should not just have a second 68k for whatever use a programmer finds for it. We should have a dedicated sound processor.

I often find that on the Amiga platform, people treat audio as something of second-level importance compared to graphics, which is unfortunate. I consider sound to be of top importance, even more than graphics. There should be a dedicated audio processor just as there is a dedicated 3D core.

Moreover, the advantage of software already supporting the m56k (found on the Delfina family of sound cards) should not be overlooked, those programs can be patched to take advantage of the m96k and m56k compatible DSP. The m96k has several strong points, like it's ability to do 10 operations in parallel, which I already mentionned.

I recommand against the modern approach of doing everything in software on a generic processor. Things such as audio should be done in hardware, in realtime on a dedicated co-processor.

> Please take the time and try to explain why you think a DSP like the propused 98000 should be faster.
It is a 96000, not an inexistant 98000. The reason why it would be faster is obvious, the m96k can do three arithmetic operations in parallel (versus 1 for the 68k), and a total of 10 operations in parallel. It works on it's local SRAM, rather than on the ChipRAM to which it is connected, this lessens the load on the ChipRAM bus, it can do DMA transfers between it's SRAM and the external RAM, in this case, the NatAmi ChipRAM. These transfers can be done in parallel to the DSP unit doing other instructions. More over, it has on board lookup tables for sine and log operations (I do however wonder how they managed to put such a table in 2048 memory words).

And i think 68080 would still be between the 68070 and the 68090.

DSP is a little bit to general but Gunnar you do agree blitter is faster than the CPU in coping information, right?
And your example is a bit unfair, there are plenty of available DSP in VHDL. ;)

But as i said before in this post it's a little too general still.

You're right that a special piece of hardware *can* work faster than a less dumbed down CPU. But each special piece of hardware introduces a new unorthogonality into the system. This makes it a lot harder to code for it. If you, say, add 3 new special hardwares, each with its own instruction set: Who will want to learn all that? This makes it 3x harder to get 100% power out of the hardware, as each new architecture has its own intricacies.

At first, I was reluctant about Gunnar's idea, too ;-) . But I arrived the conclusion that in a FPGA based system, it makes much more sense to add generic computing power first. We have a hard clockrate limit to obey. And FPGAs are perfect for doing 100x the same thing in parallel. The N050 is great, but still a lot less powerful than modern-day standard CPUs.

If we want to make Natami a powerful system, we first have to overcome the most obvious lack of Amiga nowadays: Generic computing power. This does not mean at all that there will never be some new specialised hardware additions to the system. But for now, they are really totally low priority. We need to multiply the generic power first, and make it easily accessible.

Cheers