Electric Escape (Haven 04 Cyber'tronix'')
Jaguar Performance Quotes

Table of Contents
David Taylor (Id Software)
Jez San (Argonaut Software)
John Carmak (Id Software)
Jeff Minter (Llamasoft)
Anonymous

David Taylor
Id Software: Wolfenstein 3D, Doom

December 16, 1993

The Jaguar is a cooler, much cheaper machine with more flexible hardware. The only thing the 3DO has over it is a CD ROM. Not sure about y'all, but Doom never struck me as a CD ROM game. As an amateur businessman, I don't understand how the 3DO will be ablt to survive. It costs almost 3x as much as a Jaguar and promotes CD ROM games with their painfully slow load-times and forever-development-schedules.

If you look at just the hardware, with the exception of the 68k, the Jag has more horsepower if you're willing to do some hacking and parallel design. Besides, it'll be a fun project, and the final Jag version will have no compatability problems (ah, bliss) while being as fast or faster than Doom on high-end PC's.

If y'all get a chance to check out a Jag, try to disregard the current speed/coolness of the games. They're utilizing a fraction of the Jag's horsepower. It can do some amazing stuff.

        =-ddt->

* * *

January 1994

We've got this crappy Atari TT-series thing. It's a piece. Waiting for the PC development system. Will prolly end up porting all the development stuff to UNIX anyway as the game will be simulated under NeXTStep before it ever sees the Jag.

In my humble opinion, the 3DO doesn't have a chance. It's expensive, and it's CD ROM, which has a bizarre effect on developers. It makes them want to put all this ghastly footage in there (see Crash and Burn characters -- barf!), ray-trace everything (very time-consuming and only minimally useful as most cinematics are), and basically slow things down. "We have 600Mb! Yay! Let's use all of it and force everyone to watch/hear every second!!! They'll love it!"

On the other hand, the Jaguar is a very hip machine -- bizarre, but hip. The approach we'll take is simply to get it running on the 68k (it'll crawl, but that isn't the point) and then to offload pieces of it to other processors. Although it'd be nicer if the other processors were just symmetric with no fancy app-specific stuff, they're a lot more flexible and in many ways more powerful than the 3DO hardware. And whereas there's no mass storage that comes standard with it, the cards can hold up to 6Mb(ytes), and they have 2Mb RAM inside. That's more than we'll need.

Atari also doesn't have the anal censorship policies of Nintendo which has raped SNES Wolf 3D and repeatedly stalled it for ridiculous "fixes". As you know, Doom is just an incy-wincy bit more graphic than Wolf, and it wouldn't get through the Nintendo censors until Hell froze over.

I think that a $250 home console which you can play Doom on at roughly the same speeds as you can on a high-end PC will make people just darn happy. :)

	=-ddt->


Jez San
Argonaut Software: Starglider, Starfox

November 1993

The Jag has a 64 bit bus, but its RISC chips are 32 bit. That doesn't belittle it in the slightest... the Jag does indeed outperform the 3DO by a reasonable margin!  -- Jez.

* * *

December 14, 1993

All this is leading to the question: Is the Jaguar a 64 bit machine?

It is. No question about it. It has several parts that are 64 bits big and it has a 64 bit memory architecture, so it is a 64 bit system. It's also a pretty nifty system... I think it's a very copmetitive system, and the power of the hardware is not even touched by the present batch of the first games.

I feel that Atari are justified in calling their machine a 64 bit machine regardless of how many bits their cpu contains, since it is the overall system which is talked about, and not simply the CPUs.

Case in point, the TurboGrafx 16 is known as a 16 bit system, even though it has an 8 bit processor. That's because the rest of the system is 16 bits and the aura that the machine projects, by virtue of its superior graphics to 8 bit systems deemed that it should be called a 16 bit system.

I'm a big advocate of demystifying these videogame systems. All the marketing hype surrounding every single machine (eg: 3DO, jaguar, and many others) is really unnecessary. These 'magic numbers' marketing quotes that everyone puts in their press releases or even on their boxes only serves to confuse the consumer, when what they really care about is the quality of the games, not how many 'bits' or 'pixels' per nanosecond these systems can move.

-- Jez/.


John Carmak
Id Software: Wolfenstein 3D, Doom, Quake

December 24, 1993

Feel free to quote this.


We have a few reasons for not developing on the 3DO, but development machine bigotry isn't one of them. I used an Apple IIGS for SNES development (I am never, ever, going to work with Nintendo again), and I am suffering with an Atari Falcon for Jaguar work until I can port the tools to NeXTStep. I wouldn't turn away a Mac-based environment.

The biggest reason is that I doubt that 3DO is going to become a huge success. $750 is way out of line for a pure entertainment machine. Was the Neo-Geo a success two years ago? We bought one, but we don't know anoyone else that did. I doubt there will be all that many units sold.

To make matters worse, there are over one hundred third party licensees suposedly developing on 3DO. If there were only a couple companies developing for it, they might make money. I predict there is going to be some serious lossage going on in the 3DO developer community.

The other major argument is somewhat philosphical. I don't like what people expect out of CD games. Does anyone think that the cheeseball dialog in Crash and Burn is a good addition? It turns my stomach. People expect CD games to have tons of digitized speech and video, and the 3DO is going to be strongly associated with it. The joke here is that if we ever do a CD version of Doom, you are going to get the game and "The Making of Doom," a one-hour feature film. Companies spend hundreds of thousands of dollars putting all this media into their games, and it often actually detracts from it. We don't want to be part of this crowd.

I would rather cut down to the essentials and fit on a cartridge than uselessly bulk up on a CD. I have a minimallist sense of aesthetics in game design.

Many developers are planning on waiting out the eary 32 bit hardware wars, but I want to do a cool product even if it doesn't make tons of money. Sandy (our map designer) semi-derisively calls Doom Jaguar my "reward" for writing Doom PC. "Good job, you can go play with your new toys." :-)

Our initial appraisal of the Jaguar was "nice system, but Atari probably can't make it a success". But when I got the technical documentation, I was very impressed. This is the system I want to see become a standard platform.

I was slated to do a cut down version of Doom for the Super Nintendo SFX chip, but I kept thinking about how cool a Jaguar version of Doom would be, and nintendo kept rejecting Wolfenstein SNES for bullshit reasons (a golden cross bonus item might offend christians. right.).

We finaly decided that we didn't want to be a part of the chicken-and-the-egg problem of new systems not attracting customers because developers haven't written for the platform because there are no customers. The Jag is cool, I think it has a shot at success, and I am going to put my time where my mouth is.

Why the Jag is cooler than the 3DO (from my point of view): It only costs $250. The bulk of its processing power is user programmable. The 3DO has a capable main processor (a couple times better than the weak 68k in the Jag), but most of its power is in custom hardware that has narrow functionality for affine transformations. The Jag has some stupid hardware for z buffering and gouraud shading, but I can just ignore it and tell the two 27mhz risc chips to do exactly what I want. A 64-bit bus with multiple independant processors may not be the easiest thing to optimize for, but there is a lot of potential.

There will probably be a version of Doom for 3DO. We are talking with a few companies about licensing out the port. It would be kind of fun to do it here, but I am eager to get to work on the next generation game engine that will make Doom look puny...

John Carmack
Technical Director
Id Software

* * *

May 1994 (from an interview in Edge magazine)

Q: What do you think of the Jaguar's hardware overall?

A: I honestly think it's the best designed videogame hardware around. The object processor gives you all the beneifts of a framebuffer and a sprite/background engine, without any hard limits. 16-bit CRY colour is the absolute best colour model for games I've come across. I wish I could get it on a PC!

The RISC chips are really great. When i first heard about the Jaguar, I groaned at the thought of Atari designing its own processors, thinking that they would mess it up for sure. It turns out that i like the chip architecture better than anything I've worked with: they're great to program, and very efficient. The only drawbacks are that they can only execute a few K of code at a time, so you can't just run the entire game on them. There are a few minor things wrong with the hardware: the RISC chips have a couple of bugs you need to work around, and the blitter has some special properties that seem the wrong way round to me. the only real mistake I think was made is in the keeping of the 68000 as the central procressor. If you just write something the easy way and run it on the 68000 it'll be very slow; if you go to the trouble of writing optimised RISC code to run on the other processors. It'll be over 20 times faster. This lets good programmers really stand out, but it makes it more difficult to get the performance the system is capable of. I think they should have used another RISC processor with a dynamic chache as the CPU.

* * *

September 16, 1994

Wow, there is a lot of rabid advocacy in this group... I hope this isn't a really bad idea to poke my head in here. Please, please, please do not send me mail arguing something I say here.

Doom is allllmost done. Music and modem code is about all that's left. Its good.

To address the two main topics of discussion:

Is the Jaguar doomed?:

This Christmas will tell. If Atari sells close to their estimates, they will be a serious market target for next year.

I really don't think 3DO will bury the Jag. It's too expensive, and it doesn't have a technical edge to make up for it.

The sega 32x is a nice machine, and they are shipping good numbers for Christmas. It is less powerful than the Jaguar is when really pushed, but it is easier to get things going at a decent speed.

The Sony PSX and the Sega Saturn will both cost $100 to $200 more than the Jaguar (CD machines). They are both more powerful (to a greater or lesser degree), but neither one will have a wealth of games when they debut.

The Ultra 64 is over a year away. It will probably be damn good, but a year is a long time.


3D engines:

For 24 bit parallax scrolling graphics, the Jag will outperform a Pentium, but it is only about as powerful as a low end 486 for texture mapped games. It's not really an apples to apples comparison because of the parallel nature of the Jag, but that is a fair aproximation.

The Jaguar cannot make a fully textured, full screen, full resolution game that runs at 30 fps. The bus will simply not take that many accesses. The 64 bit bus will let you do really fast shaded polygons, but texture mapping is done a single pixel at a time.

Doom had to be significantly reworked to get good performance, but it wasn't designed from the ground up to take advantage of the Jaguar. If I was designing a game from scratch for the Jag (I'm not), I would target 20 fps with a 256*180 view window in 16 bit color as a reachable goal. Doom runs 15 fps at 160*180 because the basic design is non-optimal for the Jag's characteristics. I wrote it for the PC.

There are a lot of tradeoffs you can choose. Alien vs. Predator made very different choices than I did. They have a lot more pixels on the screen, but it runs slower (about 12 fps) and the engine is a lot more limited. The engine is essentially the level of Shadowcaster on the PC (90 degree walls, transparent segments, floor/ceiling texture mapping, strictly diminishing lighting and a rear clipping plane). They chose to use higher resolution bitmaps, so they have less variety.


John Carmack
Id Software


Jeff Minter
Llamasoft: Attack of the Mutant Camels, Llamatron, Tempest 2000, Defender 2000

April 18, 1994

Q: Since T2K seems to be one of the first really great games for the Jag, it occurred to me whether Jeff might like to hint to us exactly how much of the code is 68k based and how much is Tom and Jerry doing the work. What about fps of T2K? I realise that Jeff is restrained by the NDA, but perhaps some vague idea would be nice?

Can you help us Yak?


- Certainly. First off JERRY is doing all the audio. TOM gets to draw all the calculated starfields by himself; he gets in league with the blitter to do polygon drawing, transforming and rendering game objects, and pixelshatter (where things break into a zillion pixels like the '2000' bonus). Melt-O-Vision is almost entirely a Blitter operation. The 68K runs all the game logic and orchestrates the other co-pros and builds the (very simple) Object List. The OLP isn't really doing much at all, it's sitting there twiddling its electronic thumbs and just displaying the main screen and an overlaid sprite for the score and ships info.

- Frame rate is variable, depending on how much drawing is to be done. Logically, the game engine never slows down, as it's slaved to a regular interrupt. The draw engine is decoupled from this, so when the going gets tough it drops frames where necessary, which results in a degradation in the frame rate but does not affect gameplay speed. It is possible to clog it up if you let a lot of stuff build up on the web, but the dynamics of normal gameplay mean that this happens sufficiently rarely to be a problem in gameplay. The alternative would have been to maintain a constant framerate by limiting the number of objects onscreen, or having them 'disappear' (arcade Defender used to do that!) when the load is high, but I don't like that as it spoils the consistancy of the game environment. Don't assume that the rendering of polygons in T2K is necessarily as fast as it can be done -- T2K represents my first attempt at a polygon-based game, and the first cut of my polyrenderer is unlikely to be optimal. Everyone optimises over time, and as we get into the Jag we'll learn cheats, techniques and shortcuts to do more stuff faster and smoother than ever...

\
(:-)
/


Anonymous

I haven't run any benchmarks one our latest polygon engine, but it should end up taking about 30-35% more time (assuming the same lighting model is used to shade the textures, and depth queue them).

Not that I'd prefer a 100% texture mapped Cybermorph... after some of the stuff I've seen recently, it's painfully obvious that Cybermorph suffers from extremely poor use of the Gouraud shading abilities of the Jaguar. Cybermorph also suffers from very bad color choices as well. I'd really like to see a Cybermorph which tastefully combines Gouraud, Textures, and Flat shading where appropriate. Gouraud done right is gorgeous! Look at what it does for the Tempest 2000 playfields! Combine gouraud with textured stuff, and you'll get killer effects.

An interesting side note... Jez San recently spewed out some facts about how hard it is to translate points in 3D and that the 3D math was the limiting factor in the speed of his polygon engines, therefore "Pixels-Per-Second" didn't matter much. I have heard from several 3DO developers statements to the effect that the 3DO's cel engines spend most of their time idle, waiting for the ARM60 to pass them some translated points to render (this is further exacerbated by the fact that the ARM60 is stopped in its tracks when a cel engine is working on a polygon). So, from a 3DO programming point of view, Jez is 100% correct.

From a Jaguar programming point of view, Jez is wayyyyyy wrong. The Jaguar's math capabilities kick @$$!!! Our 3D engine can stream off translated points faster than any Intel based PC clone currently out. Even with the Jaguar's 64-bit blitter working it's butt off, (in 64-bit mode, running in parallel on the system bus, at full speed) we had to put in code to prevent it from getting backed up with polygon data from the GPU! The reason for this huge advantage is that the Jaguar's parallel architecture allows the GPU to be "doing the math" while the Blitter is rendering polys. Even if the PC had a Jag blitter, it wouldn't be as fast, 'cause the Jag's math is indeed faster. The PC is further slowed by the fact that it has to move the pixels with the CPU, just as the 3DO's ARM60 is 'offline' during a polygon render.

Cybermorph was written using a very early polygon engine, which (I estimate) is about 1/3 the speed of the latest stuff (which can be improved upon if you're going to make it less 'generic'). Battlemorph should rock!


[ Back to the Atari Archives ]