Why no FPGA accelerator cards?

[commodorejohn]

I would guess because an FPGA-based accelerator offers essentially no advantage over one using an actual 680x0 chip and requires much more time and effort to design (look at how much effort has gone into NatAmi's CPU core.) If you're designing an FPGA-based system anyway, sure, it makes sense, but if you're just trying to get an accelerator for an existing Amiga system, it's a lot simpler, cheaper, and less time-consuming to just use an actual chip.

[commodorejohn]

True, but FPGA is not as limited in terms of potential performance.  It wouldn't necessarily be "just an accelerator," it could be the "mother of all 68K accelerators."  Could sell better than PPC accelerators IMO.  Imagine an incredible power-boost for your classic Amiga that is still largely compatible with old software.

I suppose so; if the estimates on NatAmi's CPU core hold up in actual use, you could get it a good bit faster than any of the actual 68k chips (ColdFire excepted.) Still, at that point you're harnessing a 150MHz CPU to a bus and chipset that still run at 3.5MHz; it's kind of impressive, but in terms of performance boost it's like those early Pentium machines that hooked a 90MHz+ CPU to ISA-bus peripherals. You'd be better off just doing what NatAmi's doing and creating a compatible, expanded reimplementation that runs the whole system at a consistently high speed.

[commodorejohn]

You'd put ram on the CPU bus, so you wouldn't need to touch the motherboard until you access the customer chip registers. With write queuing then it wouldn't impact performance much. Reads are more of a problem unless you can execute instructions out of order.
 
The bus on AGA amigas runs faster than that as well.

Yes, but still. Raw CPU horsepower plus fast RAM is all well and good, but you're still going to have that bottleneck. It's not just the chip registers, either. Not only does the chipset rely on chip RAM (which can't be accelerated,) any Zorro cards are running at a mere 2x the chipset speed. It's still better than an unaccelerated Amiga, it's just not as good a solution as a uniformly faster system.

[commodorejohn]

It's not all that slow, IIRC, as the 68000 only accesses memory every other cycle anyway (I could be wrong on that, but I believe that's the case.) And the chipset only steals CPU cycles when the software specifically gives permission (and maybe in 6bpp video modes, but I can't quite recall.) The Amiga hardware reference manual has a lot of good information on chipset/CPU timing.