Its 1993, you're in charge of the Jag, what do you do?

[triverse]

EA was already behind the 3DO more than most of the gaming companies that supported it, they helped finance the thing for goodness sake, the 3DO could be considered EA's attempt at a console (makes you wonder what would have happened had EA come out with their own branded 3DO and threw the NBA and NHL licenses behind it to match the PGA, Madden and Fifa licenses already on the system).

[sd32]

Talking about some technical aspects of 3DO, i have a question for you tech-guys. How come some 3DO games like The Need For Speed and Star Fighter have way larger draw distance than your typical Saturn, Playstation and even Nintendo 64 game(fogging is cheating)? Hell, games like Total Eclipse look better on 3DO than on PS1, sure, the PS1 versions beat it in the frame rate department, but still, you would think that PS1 would clean its clock on every facet.

[kool kitty89]

Yeah, that's whay they'd be the natural 1st party developer. But as it was they weren't a 1st party developer, given the issues with the 3DO's strategy, it's probably good EA didn't end up like this, but had they gotten that part right for the 3DO, developing exclusives for the console would have been a good push for it. (EA wouldn't have to be completely exclusive, particularly for non-console platforms, like their PC stuff, but pushing this to a lage degree could have been good) Again though, they'd have been hurt pretty badly if the thing tanked like it did. (but with a change in marketing strategy who knows what might have happened)

[kskunk]

Hmm, that's interesting, I wonder why they went with quads, perhaps for the Same reason the Saturn did (for sprite tiles)

 

Quads might seem like a great idea if you're new to 3D. Triangles seem really wasteful at first -- especially since you need two triangles for every quad, and quads are everywhere in most 3D models.

 

Like I said before, quads really are more efficient than triangles. The downside is that the world uses triangles. Most algorithms are designed for triangles, not quads (or have considerable complexity when dealing with quads). As a result, most tools only work with triangles.

 

I just noticed that the 3DO actually already has 1 MB of VRAM separate from main, so why would it still have the bus shaing problems similar to the Jaguar? (except for the general 50 MB/s bandwidth limitation)

 

Correct, the 3DO has a main DRAM and separate VRAM. However, the bus sharing problems are the same. When the Cel Engine is executing, it uses DRAM as the source and VRAM as the destination. This disables the ARM until execution is complete. Thus, if you need 50% of the ARM for your game logic, you've forfeited 50% of the graphics processing power.

 

The Playstation uses VRAM as both a source and a destination. It can do that efficiently because its GPU has a cache. This leaves bandwidth available for the CPU (which also has a cache).

 

The 3DO didn't have any choice on VRAM, because their custom chips were using older technology than the Jaguar (1 micron versus 0.5 micron). As a result, there was no room for cache or high speed logic. They needed dual busses and VRAM to get the level of performance they achieved. Even the Jaguar relies on large on-chip 'caches' (aka line buffers) to allow the Object Processor work efficiently with cheap DRAM.

 

- KS

[Atari_Owl]

Hmm, that's interesting, I wonder why they went with quads, perhaps for the Same reason the Saturn did (for sprite tiles)

 

Quads might seem like a great idea if you're new to 3D. Triangles seem really wasteful at first -- especially since you need two triangles for every quad, and quads are everywhere in most 3D models.

 

Like I said before, quads really are more efficient than triangles. The downside is that the world uses triangles. Most algorithms are designed for triangles, not quads (or have considerable complexity when dealing with quads). As a result, most tools only work with triangles.

 

Quads ARE a great idea for 3D

I wasted a lot of time around a couple of years ago when somebody put a lot of effort into persuading me otherwise, but i came to my senses. Long Live Quads.

[kskunk]

Quads ARE a great idea for 3D

I wasted a lot of time around a couple of years ago when somebody put a lot of effort into persuading me otherwise, but i came to my senses. Long Live Quads.

 

The NVidia NV1 famously used quads instead of triangles. They put a ton of research into making them work pretty well and developers still cursed it as a toy incapable of Real 3D -- after all, SGI says Real 3D uses triangles! Eventually NVidia relented and offered only triangles. Some people just can't be convinced to think differently.

 

- KS

[kool kitty89]

If you're using a software renderer you could do it either way anyway (as would be that case with the Jag and PC until 3D acceleration became popular in the mid-late '90s).

Still, triangles have the advantage in flexibility and common use. Too bad they couldn't design a system that could switch between the 2...

 

kskunk, what about my prevuous comments on a CPU with cache on the 3DO, even an 020. (or an ARM610 which seesm to be the only other ARM CPU from around the time that might have reasonable cost, the older ones with cache would probably be even worse manufacturing wise, and the ARM600 looks like it would be a lot more expensive as well, and with some unnecessary features)

Anyway, it all goes back to the buisness strategy on the 3DO, had they planned to make profits from software slaes/licencing deals, they could have cut the price of the hardware (sell it at cost, or possibly even below).

 

Also, I'd gotten the impression that the PlayStation's R3000A featured scratchpad memory, not an actual cache.

Edited August 11, 2009 by kool kitty89

[Gorf]

Also, I'd gotten the impression that the PlayStation's R3000A featured scratchpad memory, not an actual cache.

 

I believe it goes up to 64k I and D caches so I do not know where you heard that from.

[sd32]

Hmm, that's interesting, I wonder why they went with quads, perhaps for the Same reason the Saturn did (for sprite tiles)

 

Quads might seem like a great idea if you're new to 3D. Triangles seem really wasteful at first -- especially since you need two triangles for every quad, and quads are everywhere in most 3D models.

 

Like I said before, quads really are more efficient than triangles. The downside is that the world uses triangles. Most algorithms are designed for triangles, not quads (or have considerable complexity when dealing with quads). As a result, most tools only work with triangles.

 

Quads ARE a great idea for 3D

I wasted a lot of time around a couple of years ago when somebody put a lot of effort into persuading me otherwise, but i came to my senses. Long Live Quads.

 

So, thats way i always read that Owls project is so impressive...its renderer works with quads. The secret is out!!

[kskunk]

kskunk, what about my prevuous comments on a CPU with cache on the 3DO, even an 020. (or an ARM610)

Maybe this will put the answer in perspective: _Just_ the 4KB cache alone in the ARM610 uses 245,000 transistors. The entire 68020 has 190,000 transistors. The entire ARM CPU in the 3DO has 35,000 transistors.

 

That's what I was talking about when I said the 3DO was limited by its generation of semiconductor technology. They would have had to hold back the release, or take a much more aggressive design approach, to access such large chips at affordable prices.

 

People forget how cutting edge the Jaguar's chipset was for its time. Tom was 750,000 transistors. At its birth, it was the biggest chip ever created by a 3rd party (non-foundry).

 

Sure, consoles that launched 12 months later had that kind of technology too. But in 1993 it was really something special. Basically nobody but Flare was nuts enough to push the envelope that hard. It had a very high probability of blowing up in Atari's face, but they pulled it off. Saner companies had to settle with less.

 

Also, I'd gotten the impression that the PlayStation's R3000A featured scratchpad memory, not an actual cache.

It has both: A 4KB cache plus a 1KB scratchpad.

 

- KS

Edited August 11, 2009 by kskunk

[Atari_Owl]

If you're using a software renderer you could do it either way anyway (as would be that case with the Jag and PC until 3D acceleration became popular in the mid-late '90s).

Still, triangles have the advantage in flexibility and common use. Too bad they couldn't design a system that could switch between the 2...

 

If one was thoughtful about it, one might consider using BOTH.

[kool kitty89]

kskunk, what about my prevuous comments on a CPU with cache on the 3DO, even an 020. (or an ARM610)

Maybe this will put the answer in perspective: _Just_ the 4KB cache alone in the ARM610 uses 245,000 transistors. The entire 68020 has 190,000 transistors. The entire ARM CPU in the 3DO has 35,000 transistors.

 

That's what I was talking about when I said the 3DO was limited by its generation of semiconductor technology. They would have had to hold back the release, or take a much more aggressive design approach, to access such large chips at affordable prices.

 

So none of these chips were produced with newer manufacturing processes than 1 micron? And even a 68EC020 would have been more expensive in 1993 than the ARM60 the 3DO used?

 

It has both: A 4KB cache plus a 1KB scratchpad.

 

- KS

 

Ah, thanks, what I remember reading must have been an over simplication then (either on my part or the article itsself).

Edited August 11, 2009 by kool kitty89

[kskunk]

So none of these chips were produced with newer manufacturing processes than 1 micron? And even a 68EC020 would have been more expensive in 1993 than the ARM60 the 3DO used?

 

Yes, the 68EC020 was a 1 micron part (the original '020 was 2 microns!). So, in short, yes, an EC020 was _much_ more expensive to produce than an ARM60 on a 1 micron process. So presuming 3DO was good at negotiating they should have gotten a nice deal on the ARM60. Same thing applies to the 68K in the Jaguar -- that's a small, cheap, chip compared to an 020.

 

Very few chips were smaller than 1 micron at the time. Mainly just very pricey CPUs like Pentiums (0.6 microns) and 68040s (0.8 micron).

 

- KS

[kool kitty89]

So an ARM60 would be a cheaper choice for the Jaguar to use as well (even with Atari Corp's relationship with Motorola), though the cache problem would still be there. (though you'd probably still be a lot better off than with the 68k)

 

Too bad there wasn't a version of the ARM60 with a small cache like the 020 has, that would seem to be an ideal choice.

 

 

The SH-1 was around by then too, though it's only got a scratchpad (4 kB) and I don't know how the price would compare, the SH-2 was brand new at the time (I think production started in late 1993, first silicon being in October), both wer 0.8 micron. (not sure on the transistor count for either, though given the onboard 4 kB each had, it was probably closer to the ARM610, granted the SH-1 would lack the cache logic, but it also had a 64 kB ROM onboard)

Edited August 12, 2009 by kool kitty89

[Doskias]

This thread has been a very illuminating read too bad 50% of it is a foreign language to me

[Mr Morden]

I'd fill the cartridge slot up to the top with epoxy resin, to absolutely 100% guarantee nobody ever ran anything unlicensed on it without the correct disolving formula that only Leonard had hidden in his locker at Atari Corp.

 

(At least, thats what he told the judge he used it for....)

[carmel_andrews]

were'nt motorola already doing the risc versions of the 68k series (88xxx i seem to recall) bk in the day....Perhaps atari should have opted for a CR version of the 88xxx series and ran rings round the competition

[Mr Morden]

were'nt motorola already doing the risc versions of the 68k series (88xxx i seem to recall) bk in the day....Perhaps atari should have opted for a CR version of the 88xxx series and ran rings round the competition

 

I see the words and the jargon is all in that post, but it's definatly a case of the lights are on but the property is long vacated.

[promethea]

were'nt motorola already doing the risc versions of the 68k series (88xxx i seem to recall) bk in the day....Perhaps atari should have opted for a CR version of the 88xxx series and ran rings round the competition

 

I see the words and the jargon is all in that post, but it's definatly a case of the lights are on but the property is long vacated.

 

There is an old saying that goes "If my grandmother had te*****es, she'd be my grandfather".

 

That said, Atari should have opted to use the Quad4 CPUS and those 1ghz memory modules and bus... it should have saved them a world of problems. What were they thinking.....

[kool kitty89]

were'nt motorola already doing the risc versions of the 68k series (88xxx i seem to recall) bk in the day....Perhaps atari should have opted for a CR version of the 88xxx series and ran rings round the competition

 

The 88000 architecture was never popular on its own, so that would be a problem right there, both in terms of volume production, available tools, and familiarity with programmers. Also I have no idea cost wise what any such chips would be like. (if they were similar in transistor count to the 68020 then there's basicly no point) Even if it was a smaller chip, the lack of popularity and volume production would likely make it more expensive, particularly as the 020 was an older chip by this time, an dwould be less pricy for that reason as well. The ARM60 is much samller as kskunk mentioned (less than 1/5 the number of transistors as an 020, barely more than than 1/2 of the 68000 even), though it was also newer and still had the problem of lacking cache. (though it's still much better than the 68k, with the 32-bit bus -also opening Jerry for this as well, and a lot more powerful than the 68k)

Edited August 12, 2009 by kool kitty89

[kool kitty89]

OK, I'm still not clear on one issue. How does might an ARM60 have been (practically speaking), for replacing the 68k. Particularly cost wise. (compared to both the 68k and EC020)

 

I understand it wouldn't have the advantage of the 68020's cache, so it would still be bogging things down, but it would still be doubling the bandwidth over the 68k (and by extention Jerry) and would be a lot more powerful than the 68k (or 020, but that's primary advantage is the cache), and also of a common, well supported architecture.

Manufacturing cost wise, the ARM would obviously be at a great advantage, but this doesn't necessarily directly represent the price of thse chips. (or Atari's favorable relationship with Motorola) If silicon was the only factor, than the ARM60 should have been considerably cheaper than the 68000, but there are obviously other contributing factors. (like age of the chips and volume production)

 

Could the older ARM250 have been a cheaper, albeit less performant, alternative as well. (was it available in 1 micron?)

 

Also, is the GPU RISC connected to a 64 bit external databus, or just 32-bits? (the FAQ seems to point to a 64-bit bus, but some comments in recent discussions made me wonder)

Edited September 4, 2009 by kool kitty89

[kskunk]

Also, is the GPU RISC connected to a 64 bit external databus, or just 32-bits? (the FAQ seems to point to a 64-bit bus, but some comments in recent discussions made me wonder)

 

The GPU RISC has limited 64-bit connectivity. Most of the 64-bit paths in Tom are for the blitter and object processor.

 

The only 64-bit path in the RISC is a direct connection to the main bus that can be accessed by a pair of special instruction, letting the GPU program load or store an aligned 64 bits at a time. This can help reduce GPU bandwidth in a few special circumstances but is not generally useful. It is almost always faster to use the blitter when you need to move data at peak bandwidth

 

All the other paths in the Jag RISC is 32-bits. The on-chip SRAM is 32-bits, and instructions are fetched 32-bits at a time even from main, leading to twice the bus cycles that you might predict.

 

I dunno much about the costs of ARMs in the early 90s, but it was a new business for Acorn at that time. Originally it was an Acorn-only part, so if you go back far enough it might not be possible to buy in volume at any reasonable price.

 

ARM history is very interesting. In the early days, ARM was a great little chip made by a very tiny team, much like Tom and Jerry!

 

- KS

[kool kitty89]

ARM history is very interesting. In the early days, ARM was a great little chip made by a very tiny team, much like Tom and Jerry!

 

- KS

 

But it they should have been available by the time the Jag was nearing release, considering the 3DO used it, plus there's the Apple Newton about the same time. (granted my comments about older chips like the 250 might not have been available) And they'd licenced some designs to Plessy and Sharp by then too. http://www.advanced-risc.com/sidebar1.htm http://www.ot1.com/arm/armchap1.html

 

Licencing one of the chips might have been an option too, and then use the smae chip vendor as was being used to produce Tom+Jerry.

 

I also just re-read this:

The bus architecture of Tom depends on a host processor. It can be a Motorola chip (68K, 68020), or an Intel chip (286, 386, or i960), or a MIPS R3000. They went to some lengths to support all these options, and it paid off since CoJag later switched to MIPS.

...snip

 

So ARM wouldn't have been one of th esupported Architectures of the existing design? (not that they couldn't have encorperated that had things moved in that direction)

[marigul]

Just skimmed through this thread..as someone that was there..lets say most of the people on the ground at Atari had simliar ideas. But the Tramiel's were autocrats and some of the siblings were there because they had to be. Plus they had no real joy for games, they loved computers(and money). But if they had spent money on marketing and got rid of a few bugs in the system, it would have been a cool little machine and even if not not a huge success wit would have had a fondness beyond this forum. I'm grateful for my time their because it got me in the biz...but unless the Tramiels had given up the reigns there is now way the Jag could have been successful.

[82-T/A]

EA was already behind the 3DO more than most of the gaming companies that supported it, they helped finance the thing for goodness sake, the 3DO could be considered EA's attempt at a console (makes you wonder what would have happened had EA come out with their own branded 3DO and threw the NBA and NHL licenses behind it to match the PGA, Madden and Fifa licenses already on the system).

 

 

Yeah, I guess that's true. I have one, but I haven't really played it much. Most of the good games I have for it are EA games... like Wing Commander III and Star Control II...