Save Scening with AAA

[downix]

Bodie wrote:
Why would you want AAA when you could have a brand spanking new video and sound card?  

AAA would have been nice back in 93 and 94,  but a bit out of date  now ;-) .

Cheers
 :pint:  :pint:

Well... yes and no.  AAA does have some advantages in design, and could actually be brought up to spec nowadays.  Fortinately, it was designed using an HDL, which means you can update the design far easier than the classical schematic approach.  Replace that VRAM with DDR-II RAM, increase the clock, you've got options.

However, it is still a 2D chipset.  Sound support kicks anything Creative Labs has in the teeth (99.6khz audio here guys) and with the higher-speed, it would actually not be too difficult to modify the disk controller to Serial ATA support.  

But, it's 2D... unless you pull what Commodore was attempting with Hombre.  Integrate a high-end processor into the chipset to act as a 3D accelerator, add in a texture pipeline, etc.  But then, you're talking a whole-scale re-design which would take 12-14 months to accomplish, providing you use an odd-the-shelf processor core.  (I could recommend a few)

So, a year+, we'd end up with a very programmable chipset that would be able to compete with the big boys in a real way.  Maybe not on polygon count, but definately in programmability and quality of output.

Pity Amiga, Inc is only interested in software.

[downix]

gregthecanuck wrote:
Hi there - this is my first post so be gentle with me... might as well dive right in ... :-P .... what are you guys thinking???

The graphics card market is so cut-throat and rapidly evolving who would want to waste their time developing a custom graphics chip set?  This kind of R&D money is better spent developing abstractions for off-the-shelf cards.  Way, way cheaper and a better end-result.

Same idea goes for sound cards.  Creative's newest Audigy 2 has pretty good specs (192Khz, 24bit) and even supports 5.1/6.1 sound.  Also includes a built-in firewire port - all for 100 bucks on eBay.

Custom hardware is dead, dead, dead.

The better path is to take the best hardware out there and do a great job making it all work together.

OK, flame away - I can't wait for my first flame....  ;-)

You say custom hardware is dead... then go on to mention custom hardware as an alternative.  Can you please explain this to me?

The Audigy2 is such a custom chip solution.

"But I can buy it off the shelf, or on eBay."

So?  Doesn't mean that it was not custom-designed by Creative Labs.

Custom chips are designed and sold everyday.  In fact, they out-weigh non-custom designs by an order of magnitude.  ATI, nVidia, even Intel and AMD make a living designing their own lineup of custom designed chips.

Now, if you ment "a chipset custom-made for the Amiga market" then I might agree with you, that it might be a bad option, unless you can guarantee a competitive edge that way.

See, hardware has an absurd amount of overhead nowadays.  Do you honestly think that a G4 costs $200+ to make out of a $45 wafer and $10 per chip fabbing cost?  If you custom design a solution to eliminate the high-overhead components in a system, you can actually make a competitive solution.  The game console makers do this all of the time.  What prevents a PS2 from becoming a desktop computer?  The software.

So, someone could make a custom solution that could actually deliver something of an edge to the Amiga market.  Note, I said could.  Will it occur, not likely.

[downix]

Have you ever actually designed chips?  10 million to design a chipset, are you kidding?  Any engineer worth his mettle can make re-usable modular components that can be re-implimented in 6 months to a new process and gain the full performance improvement.  Even the initial design wouldn't cost beyond $400k unless you're being robbed blind.

Not saying Amiga should, but saying that it is possible and there are advantages to it that are more than off-set, including being able to branch out into new markets, thereby growing the Amiga community.

[downix]

Actually AAA allows just such an arrangement.  Remember, modern 3D gfx chips, as a rule, do *not* output to the screen.  They need an intermediate, called a RAMDAC, to do this.  AAA could act as this RAMDAC, allowing new functions and ability to be inserted *into* the final framebuffer.

Just a small note of possibilities with alternative arrangements with AAA.  Could even make a 3D-only chip, allowing for good specialization, rather than the 3D/2D split-mode chips availible today.  Then when you don't need 3D, turn the chip off for maximum power savings.

[downix]

@JJ
Untrue, I can name 4 people with access to the designs.  I can even name 2 with access to the AAA and Hombre designs.  The issue is IP rights to these designs, and Amiga's unwillingness to license them to 3rd parties.

[downix]

@Athiest

Of course there are.  AGA and even AAA could be put into FPGA quite easily nowadays and not got an arm and a leg.  I've got a trio of Xilinx FPGA's sitting here emulating components of my own chipset (which is larger than AAA, would need larger FPGA's than I have today in order to impliment fully) so I know that they can get the speed needed if you're willing to pay for it.

A smarter idea, of course, is to mask-form a chip using ASIC techniques.  Get the low volume production cost savings of FPGA, but the lower cost per-chip of a custom chip.

[downix]

A fab will run you about $250 million for one "good enough" for what we're talking about here.

And the cost for a mask is not $2 million.  That's the cost of a whole assembly.  The mask itself is only $400k or so.

But, let's address the situation.  We don't *need* 0.13u processes here.  Hell, the Alpha was kicking everyones posterior and taking names using a process 4 years old.... namely dirt-cheap to get chips made at.  Let's use 0.18u or even 0.20u here, suddenly the cost for a whole assembly is under $150k.

You're suddenly in a price-affordable range and the technology is in-reach of the big-boys.

[downix]

Well, let's go into detail here, if you'd like Jose:

roughly 2.5 million transistors total for the design as/is.  Design done in a HDL called "M", no longer in major use it seems.  However, Mentor has conversion tools to render these to Verilog in short order, so $120k for the tools needed.

now, 2.5 million transistors comes down to roughly 500k gates, depending on register handling, any SRAM included, etc, so 500k is a good estimate.  half of the gates are in 1 chip, Andrew.  The other half are split among the remaining 3 chips.  Now, need to double 2 of the chips, to get the 64-bit configuration.

So, we need a single 250k gate FPGA's and 5 50k gate FPGA's.  

Altera has several FPGA's that fit the bill.  For the 5 smaller ones, I'd recommend Flex10k's.  For Andrea, use an Apex20KC, and throw in a DDR-RAM or sDRAM controller to replace the legacy model as well as a PCI bridge over the off-chip logic that's there now, effectively for free.  

Total cost for the chips in bulk:  $18.

[downix]

@alx

Sorry about that, I tend to forget not everyone knows my habits.

I always go on price for 1000 units when pricing out any solution.  if I price it out differently, I always let people know.

[downix]

Jose wrote:
With the price current classic Amiga hardware sells 18$ is ridiculously nothing. I bet some people would give more than 100$ for it.

But I remember AAA wasn't finished yet.
If it was finished and you're completely sure of what you're saying this could really happen. I'd buy it. At this price many people would buy it for nostalgia alone.
Is EVERTHING included in the estimation you made? Like you forgot the D/A converters or something?

Also, that $18 was for just the chips.  I also mentioned the cost of the software needed at just under $200k.  As for work left, nobody can be sure how much is left to do so it is impossible to estimate.  Dave Haynie claims one value, but he's also not a chip designer.  

Also, I was including the DAC's and PLD's needed, as with modern technology in FPGA's you'd no longer need these supprting chips.