Atari Jaguar architecture discussion

swaaye

Entirely Suboptimal
Legend
Supporter
Alrighty then. I'd like to see some educated discussion about the Atari Jaguar.

It seemed to be designed for Gouraud shaded 3D and 2D games, but of course everyone wanted texture mapped 3D. It had a bizarre architecture with a 68k CPU that wasn't supposed to be used to run the games. Unified memory. The gamepad was weird and horrible.

John Carmack ported Doom to it and wrote about the experience, but I can't find that at the moment...
 
I owned one! Dunno about the architecture, but in practice many of the games paled in comparison to even the lowly SNES.
 
Alrighty then. I'd like to see some educated discussion about the Atari Jaguar.

It seemed to be designed for Gouraud shaded 3D and 2D games, but of course everyone wanted texture mapped 3D.

Don't take this the wrong way but back when 3D0, Jaguar, Saturn and PlayStation came out (with PlayStation being the dark horse) 3d hardware technology that was designed and developed just could not really deliver or wasn't really designed for texture mapped 3d unless advanced software development by top programmers were involved.

Jaguar's life cycle was very brief...I can't recall if it barely lasted as long as the 32X which in part was initiated by Sega of America to fight off Jaguar competition wide.

What I'm trying to say is "everyone wanted texture mapped 3d" is most likely not accurate until way after PlayStation launched in North America and when most people realized what that was.

Unfortunately my plans to get a Jaguar fell apart so I only purchased Tempest 2000 despite having many print magazines with some detailed information and some references to notable devs like Jeff Minter (hope I spelled that right) who made a new version of Defender before making Tempest 2000...then he made a media player effect that was supposed to be for the Jaguar CD...

The media at the time did have some coverage but as far as the actual games...ehh did I mention I actually purchased Tempest 2000?

I believe I still have two print magazines that mention and might have an interview with John Carmack about Doom Jaguar.

On a technology level it's not really fair to use media hype because the media didn't make hardware...and hardware was going through leaps and evolution from 1993 to 1994 alone before several leagues were jumped from 1995 to 1996.

Edit: iirc the impression I got from JC making doom was that he had fun or enjoyed it...or something.
 
Jaguar had texture mapping. It wasn't designed in a way that was especially good at it, but it could do it. Gouraud shading did kind of set it apart from some other contemporary hardware that didn't have it (3DO, Sega Model 1). Although it didn't offer colored lighting, the way it did shading was pretty interesting nonetheless - it had a luminance/chrominance sort of color space and varied the luminance for shading. This gave a lot of shading resolution while only changing one 8-bit value.

John Carmack seemed less enthusiastic about the Jaguar after he actually completed Doom. Here's what he had to say in 2000:

The Jaguar was definitely significantly hampered by its technical flaws, which kept me from ever being too big of a Jaguar booster. I was proud of my work on Wolf and DOOM (more so than just about any of the other console work Id has been involved in until just recently), but in the end, the better consoles won the war.

The tragic thing about the Jaguar is that it could have been pretty significantly better with a few minor changes/fixes. Mainly adding in some more buffers. Like a tiny instruction buffer to keep the 68k from hogging the bus so much. And a tiny buffer on the blitter to allow it to reuse extra textures loaded with 64-bit reads without having to hit DRAM every time (Carmack himself said it should have had this). Or small FIFO to allow loading the blitter registers in advance, so the GPU and blitter could be decoupled much better.

This wouldn't have made Jaguar competitive with the later consoles, but it would have helped. But then, I guess having better libraries and technical support for developers was another major thing Atari lacked.

The funny thing is, all four of the Atari consoles seem to have a lot in common even though they were developed under different companies. They seem to favor unified memories and a lot of CPU driven operations which makes things more flexible but ultimately slower. They like display primitives that are based on fine grain linked lists that can be a hassle to manage. And they all have terrible or no audio synthesis hardware (where you have to do it in software). I guess it's no real wonder that they were failures after the 2600.

One thing that I do find impressive about the Jaguar is how much they fit things into a relatively tight custom design. Two main custom chips did most of the heavy lifting (plus DRAM and an ancient 68k), manufactured on a good IBM process. That allowed Atari to keep the thing pretty cheap, and would have facilitated a shrink that combined the two chips at some point if they were successful.
 
still have a lynx 1, amazing machine for the time.

I also got a Lynx 1 near the end of its life when I was a kid. It was used and cheap. Sadly I never got more than Blue Lightning or Toki for it. It looks like I missed out on some pretty good titles.

Lynx was a very interesting hardware design. The other handhelds back then were either similar in design philosophy to consoles or were actual console shrinks. Lynx on the other hand had a design that seemed well tailored to the low resolution display it and other contemporary handhelds were stuck with.

The hardware could even sort of do flat shaded triangles, but since you'd have to calculate the slopes on the CPU and store lengthy sprite headers for them (two per triangle, usually) it was probably not worth a lot.

Fun fact: the two engineers responsible for the Lynx went on to work on the 3DO, and if you look at the designs it shows.
 
The Jaguar was by far my favourite console and I still have a few but the majority of the games are rubbish. One of the problems was the lack of storage space on the cartridges. It really should have come with the CD add-on as default.
 
I don't know... Those 1-2x CD-ROM drives were not overly useful. But I guess they could have at least joined the horrible FMV fad era. :)

I think the only games on the platform that interested me were Doom and AVP... But the tech seems quite underutilized. Probably because it was another very challenging architecture.
 
http://www.yaronet.com/topics/17703-john-carmack-parle-de-la-la-jag-2000

"The little risc engined were decent processors. I was surprised that they didn't use off the shelf
designs, but they basically worked ok. They had some design hazards (write after write) that didn't get
fixed, but the only thing truly wrong with them was that they had scratchpad memory instead of
caches, and couldn't execute code from main memory. I had to chunk the DOOM renderer into nine
sequentially loaded overlays to get it working (with hindsight, I would have done it differently in about
three...)."

JC ..... he did say a bunch of other stuff too, so click the link.. apparently he also liked the 16bit CLUT the Jaguar had and wished it was available on PC (that is from memory....read about it in an edition of Edge magazine).
 
http://www.yaronet.com/topics/17703-john-carmack-parle-de-la-la-jag-2000

"The little risc engined were decent processors. I was surprised that they didn't use off the shelf
designs, but they basically worked ok. They had some design hazards (write after write) that didn't get
fixed, but the only thing truly wrong with them was that they had scratchpad memory instead of
caches, and couldn't execute code from main memory. I had to chunk the DOOM renderer into nine
sequentially loaded overlays to get it working (with hindsight, I would have done it differently in about
three...)."

JC ..... he did say a bunch of other stuff too, so click the link.. apparently he also liked the 16bit CLUT the Jaguar had and wished it was available on PC (that is from memory....read about it in an edition of Edge magazine).

This is actually what I was looking for at the beginning of the thread but couldn't find it. Thanks for posting. :)

I wonder how many games just ran on the 68k.
 
Last edited:
"I wonder how many games just ran on the 68k."
Quite possibly zero unless you wanted Atari ST style graphics!

See for yourself.. https://atariage.com/system_items.php?SystemID=JAGUAR&ItemTypeID=SCREENSHOT

A lot of games look purely sprite based 2D, utilizing the object processor but not the blitter, or at least not in any obvious way. They probably also don't use the GPU, thereby running most of the game code on the 68k. Presumably the DSP is still used for music, at least for games that actually had music.

Maybe not exactly Atari ST level but reminiscent of Amiga games. Including a decent number of actual Amiga ports.
 
That's my bad, I was only interested in the 3D games of the time!
I loved racing games but even I did not dare touch Rebellion Software's Chequered Flag.

Another great website with some pretty rose tinted glasses:

Skilled programmer Steven Scavone, key member of 3D Stooges which released Gorf, still develops for Jaguar. Comparing it to systems he’s worked on, Scavone elaborated on tech-specs, also explaining in laymen’s terms. “It should be coded in as much assembler as possible. This machine flies when fuelled by assembler. The RISCs in proper concert with the 68k will do some absolutely amazing graphics. The Jaguar could [utterly] crush any 2D system. It’s a lot easier to program 2D for than the PSX or N64. You can thank the Tramiels for it being ‘underpowered’. The chips were not complete and had bugs. The designers, who weren’t experts in silicon design, missed fundamentals. Just one more register and [it could have run without stalling all the time]! If they [had fixed this], the Jag would have blown away the PSX. Later 3D titles like Battlesphere proved that systems at the time were no match for it.”

The failure of the Jaguar Europe launch.. get the violins out guys:
“It was frustrating, because there was 12-15 of us TOTAL, doing a Europe-wide launch of a major electronic commodity with absolutely zero budget, getting pages upon pages of press coverage and building an enormous demand. And we were hearing that we were rubbish at marketing, from journalists who knew absolutely nothing about the reality of the situation. You felt like screaming at them ‘C’mon then, you come and see if you could do any better with our finances.’ But of course we couldn’t say a word. We just had to keep on going.”
- Darryl Still - Atari UK

http://www.retrogamer.net/profiles/hardware/atari-jaguar-2/

I did an interview over the phone with one of the founders of Rebellion Software which I was going to use for a school project - in my excitement (I was about 15 years old at the time) I forgot to record it! Those guys loved the machine and were very excited at making proper 3D games. I was looking forward to a Jaguar 2 with the bug fixes and increased clockspeed.... PSX killer indeed!
 
I can't remember seeing anything of the jaguar during the dominance of the Megadrive/Snes in the 16 bit era. I only know of one person who actually bought one of these mythical beasts. I'm sure they did exist though.
 
A naked Atari Jaguar:
Atari-Jaguar-Motherboard-L.jpg
 
It had a really REALLY bad catalog, marketing and press. Just check this SCES 1994 coverage.


Few tech demos, non-gaming programs and a few poorly optimized versions. And it came out 1 year before.
 
For any of you interested in what a "fixed" and enhanced Jaguar would look like you should check out the specification for Jaguar 2, aka Midsummer:

https://github.com/jaguar64/devdocs/blob/master/Midsummer Tech Ref Version 6.pdf

Development was far enough along that prototypes were built, you can see one here http://www.atarimuseum.com/videogames/consoles/jaguar/jag2.html Some others have surfaced, but from what I hear a lot of them don't work. I don't think anyone has managed to run anything on them other than original Jaguar games, but I bet it's possible to run Jaguar 2 code on it with a flashcart.

The technical reference document is interesting because it makes a lot of comparisons with the original Jaguar and includes some good insights on where the system was most flawed. In particular the texture mapping memory considerations section on page 64, which flat out admits that the Jaguar's texturing design was terrible. It took a whole 11 cycles to render a textured pixel, which results in a paltry 2.4 Mpixel/sec peak textured fill rate. In reality the actual attainable numbers would be far less because of everything else in the system that drained the bus, including the obligatory transfer from the framebuffer to the line buffers which output to the screen. Some people like to talk about all the untapped 3D prowess the system had but titles like Doom and Alien vs Predator were probably already pushing its limitations in textured 3D graphics. AFAIK AvP's developers even talked about how they started with a much faster demo but the frame rate went down and down as they implemented actual gameplay features that ate into the bus time.

Midsummer is an interesting design that would have been a lot more competitive than Jaguar, particularly if it came out in 1995 (although that was probably never realistic). It bears a lot of resemblance to the N64. On both systems, a RISC processor is needed to calculate slope gradients to feed to the rasterizer, although N64 can at least render whole triangles while Midsummer's blitter will usually need to render them as two trapezoids. Both systems perform texturing from a small integrated buffer, and both systems render to a unified system RAM. On the other hand, the N64's CPU and RSP are a lot more powerful than Midsummer's RCPU and GPU respectively, and Midsummer splits audio tasks out to a separate processor which can (and probably should) also do other tasks. And the N64's RDP has a lot more features like perspective correct texturing and mip-mapping. Midsummer's blitter can perform bilinear filtering, but it comes at a huge performance cost so probably generally not worth it.

The Midsummer documentation lists a 750k triangle specification, but it's a number derived purely from a fillrate calculation plus fixed polygon sizes and is therefore pretty meaningless. To sustain such a rate the GPU would have to setup triangles and program the blitter (usually twice) in a mere 44 cycles which doesn't seem remotely realistic, even if the other CPUs are leveraged to perform T&L.
 
Last edited:
Back
Top