Texture Mapping

[BeefMan]

Sorry if this has been covered before but I was just curious what it is about the Jaguar's hardware that makes texture mapping so hard for it to handle? Most games that try to use it extensively seem to suffer from pretty bad slowdown like SkyHammer and HoverStrike.

 

On paper the Game Boy Advance doesnt seem any more powerful than the Jaguar but Ive seen a couple games that seem to be pushing texture maps pretty quickly(V-Rally 3), although not at a very high resolution . Ive also read that the Quake and Quake2 engines have been ported for use for the GBA.

 

http://www.bobbeetec.com/drg_media_v.htm

 

So my next question is, if anyone can answer, is that what makes the GBA so much more powerful that it could handle Quake. Or maybe the Jaguar could handle quake at the same resolution of GBA 240x160 I think?

 

Thanks for any replys!

[atarifan49]

It's mainly due to RAM limitations and CPU speed power.

 

Let me try and give some background as to why I think this.

 

Battlesphere is a good example of the proper blending of Gourand shading and some texture mapping (logos on side of ships). Fight For Life shows the struggle with heavier texture mapping. Haven't been able to look at the source for it yet, so I'm not sure how much they relied on the 68000, compared to using GPU/DSP. Battlesphere is also an example of what Atari had envisioned Jaguar programmers to one day progress to, writing games that completely rely on the GPU/DSP processors for all operations.

 

Now Atari Games did show in Area 51 and Maximum Force the improvement you could get with the Jaguar chipset by beefing the system up with a better processor and some more RAM. The Jaguar is only using a 68000 processor, which internally can do 32bit, but externally communicates at 16bit. Atari Games used a 68020 initially and a R3000? RISC processor for later Co-Jag systems. 68020 is a better processor than the 68000. But it was costs that drove the decision to use the 68000.

 

Also, you're trying to compare game systems from different stages of technology advancements. Jaguar chipsets were desgined and worked on in the very early '90s. Around that time, Gourand shading was one of the features being utilized by games at that time. By the time the Jaguar was on the market, things were advancing to texture mapping. Jaguar's design focus wasn't texture mapping.

 

How long has the GBA advance been around? GBA has the luxary of years of evolution of technology behind it. Plus, I would imagine it's R&D budget is a hell of a lot bigger than Jaguar ever got.

 

You just can't compare apple's and oranges like this. You have to look at the persepective of the time the system was designed and marketed. The systems to properly compare with of that time is SNES, Sega Genesis, Turbo Graphics, and 3DO. And if you compare it against the PSX (which came out in 1995), remember, Sony invested around $500 million dollars in the R&D of that system.

 

Sorry for the bouncing around on the answers. But your questions don't necessarily have simple answers.

 

 

Glenn

[ZylonBane]

Sorry if this has been covered before but I was just curious what it is about the Jaguar's hardware that makes texture mapping so hard for it to handle?

It's because of the Jag's texture-mapping hardware.

 

There isn't any.

[Thunderbird]

Sorry if this has been covered before but I was just curious what it is about the Jaguar's hardware that makes texture mapping so hard for it to handle?  Most games that try to use it extensively seem to suffer from pretty bad slowdown like SkyHammer and HoverStrike.

 

On paper the Game Boy Advance doesnt seem any more powerful than the Jaguar but Ive seen a couple games that seem to be pushing texture maps pretty quickly(V-Rally 3), although not at a very high resolution .  Ive also read that the Quake and Quake2 engines have been ported for use for the GBA.

 

http://www.bobbeetec.com/drg_media_v.htm

 

So my next question is, if anyone can answer, is that what makes the GBA so much more powerful that it could handle Quake.  Or maybe the Jaguar could handle quake at the same resolution of GBA 240x160 I think?

 

Thanks for any replys!

 

 

The real difference here is that RAM is a lot cheaper and a lot faster now than it was in 1993. The ability to draw textures relies most heavily on the ability to copy pixel data from one area to another. The Jaguar has main RAM which is quite slow compared to newer RAM chips. This makes it hard to use a high framerate with a lot of textures. The Jaguar also has some faster RAM built into the GPU and DSP chips, but this fast RAM is so small in size that it really doesn't help much at all with texture mapping.

 

By the way, the Jaguar has hardware which is used to do Texture-Mapping, but it is not as autonomous as newer hardware that does texture-mapping. The Jaguar's texture-mapping hardware needs to be updated with new values for every line of pixels, while newer hardware can draw whole polygons without any updates.

 

Put all this together, and the Jaguar is going to be slower than modern hardware.

[Punisher5.0]

Sorry if this has been covered before but I was just curious what it is about the Jaguar's hardware that makes texture mapping so hard for it to handle?

It's because of the Jag's texture-mapping hardware.

 

There isn't any.

 

Neither does GBA. Heck, GBA doesn't even have 3D hardware.

[ZylonBane]

The GBA's CPU is a 32-bit ARM chip. Probably about as powerful as a high-end 486. Combine that with the low screen resolution and you have just enough juice to do software 3D.

[Gunstar]

The GBA's CPU is a 32-bit ARM chip. Probably about as powerful as a high-end 486. Combine that with the low screen resolution and you have just enough juice to do software 3D.

 

Is the GBA's ARM chip the same or from the same "line" of chips as the 3DO's?

[Thunderbird]

The GBA apparently has 96K of high speed RAM directly insie the CPU. The Jaguar only has 4K of RAM in the GPU, and this has to hold the GPU program, so little (if any) of it can be used for texture-mapping.

 

If the Jaguar had 96K of RAM on the GPU, you would have seen Texture-Mapping somewhat comparable to early PSX games.

[BeefMan]

Thanks for the feedback! I thought it might of had something to do with the amount of internal memory on the cpus. Too bad memory was so dang expensive around the time of the Jaguars production.

[Alex Czarnowski]

Speaking of cpu, interma memory size and texture mapping does any one know what happened to all those overclocking projects? I mean there are very interesting turbo cards for 8 and 16 bit machines - it's time to have something similar for Jag

Cheers

[Gunstar]

The GBA apparently has 96K of high speed RAM directly insie the CPU. The Jaguar only has 4K of RAM in the GPU, and this has to hold the GPU program, so little (if any) of it can be used for texture-mapping.

 

If the Jaguar had 96K of RAM on the GPU, you would have seen Texture-Mapping somewhat comparable to early PSX games.

 

Well, when I look at games like SW2k, BI/WN, Skyhammer, etc. I think we are seeing textures easily comparible to early PSX games, possibly even better, they sure don't seem to get a blocky when you get up right next to them as most PSX games do! IMHO.

[Thunderbird]

Speaking of cpu, interma memory size and texture mapping does any one know what happened to all those overclocking projects? I mean there are very interesting turbo cards for 8 and 16 bit machines - it's time to have something similar for Jag    

Cheers

 

It really can't work on the Jaguar. The Jaguar uses the same clock for sound and video that it uses for the CPU's. If you increase the CPU clock then you screw up all the timing for the video and audio.

 

I suppose you could make a new game that had all-new configuration routines that would put new values into the hardware registers for the timings, but no existing software would be compatible with a faster Jag.

[Thunderbird]

The GBA apparently has 96K of high speed RAM directly insie the CPU. The Jaguar only has 4K of RAM in the GPU, and this has to hold the GPU program, so little (if any) of it can be used for texture-mapping.

 

If the Jaguar had 96K of RAM on the GPU, you would have seen Texture-Mapping somewhat comparable to early PSX games.

 

Well, when I look at games like SW2k, BI/WN, Skyhammer, etc. I think we are seeing textures easily comparible to early PSX games, possibly even better, they sure don't seem to get a blocky when you get up right next to them as most PSX games do! IMHO.

 

The quality of the textures may be similar in those games to the PSX, but the speed at which they are rendered is inpaired by the lack of fast RAM on the Jaguar. If the Jaguar had more fast RAM, then it would have the nice textures and also be able to render them at a similar framerate as the PSX.

[Jetboot Jack]

Gunstar,

 

Unfortunatley even the earliest PSX games displayed frame rates, number of polygons and texture mapping many, many times better than even the best Jag demo or game...

 

At a purely functional hardware level the PS1 has many many times the pipeline bandwidth than the Jag and the system is designed from the ground up to produce texture mapped polys in large numbers.

 

sTeVE

[JagAttack]

The quality of the textures may be similar in those games to the PSX, but the speed at which they are rendered is inpaired by the lack of fast RAM on the Jaguar. If the Jaguar had more fast RAM, then it would have the nice textures and also be able to render them at a similar framerate as the PSX.

 

Would it have been cheaper to increase the clock speed to increase the MIPS instead of including more on chip SRAM? The playstation's graphics chip can do 60MIPS vs the Jags 26.6MIPS. In some calculations I did, the bottleneck was actually the cpu's ability to do the instructions to access the memory in the 33.3ms it has to do a frame (assuming 30fps) not the dram being slow. I agree that having ram that has zero access time is good but couldn't the problem have been solved by getting the MIPS up to say 60 like in the playstation.

[LinkoVitch]

The quality of the textures may be similar in those games to the PSX, but the speed at which they are rendered is inpaired by the lack of fast RAM on the Jaguar. If the Jaguar had more fast RAM, then it would have the nice textures and also be able to render them at a similar framerate as the PSX.

 

Would it have been cheaper to increase the clock speed to increase the MIPS instead of including more on chip SRAM? The playstation's graphics chip can do 60MIPS vs the Jags 26.6MIPS. In some calculations I did, the bottleneck was actually the cpu's ability to do the instructions to access the memory in the 33.3ms it has to do a frame (assuming 30fps) not the dram being slow. I agree that having ram that has zero access time is good but couldn't the problem have been solved by getting the MIPS up to say 60 like in the playstation.

 

But if the CPU is made a lot faster with the same RAM, it will be kinda pointless as the CPU will just be sat around waiting for data to be read from RAM. So whilst it could process stuff it has in it's cache nice and quick, the data it needs to use is stored in the slow RAM which it will have to wait for.the data ro be read from main RAM.

 

So you would need to icrease RAM speed too. Plus I wouldn't be suprised if the PSX has more bus bandwidth enabling it's CPU's to talk to the memory in a better way than the JAG's.

 

More RAM on the GPU would be ace, or perhaps the ability to allocate space that would ONLY be accessible via the GPU (giveit it's own bus and use dual port ram like vram). and a shiney MIPS CPU instead of the 68K for that extra umph too..

 

Or give the 68K some cache so it's not chewing on the bus all the time it's running..

 

oh and some go faster stripes..... and a nodding dog... yeah and....

[Thunderbird]

The quality of the textures may be similar in those games to the PSX, but the speed at which they are rendered is inpaired by the lack of fast RAM on the Jaguar. If the Jaguar had more fast RAM, then it would have the nice textures and also be able to render them at a similar framerate as the PSX.

 

Would it have been cheaper to increase the clock speed to increase the MIPS instead of including more on chip SRAM? The playstation's graphics chip can do 60MIPS vs the Jags 26.6MIPS. In some calculations I did, the bottleneck was actually the cpu's ability to do the instructions to access the memory in the 33.3ms it has to do a frame (assuming 30fps) not the dram being slow. I agree that having ram that has zero access time is good but couldn't the problem have been solved by getting the MIPS up to say 60 like in the playstation.

 

No.

 

The Jag's two CPU's run at 26.6MIPS approximately equalling 53.2MIPS combined. this is totally independent from the Blitter chip which is doing the actual rendering onscreen. The bottleneck in the Jaguar is the DRAM, which is really slow and is even slower if you are moving data across a page boundary (page fault). Since it's impossible to store textures and framebuffers on the same page in DRAM, everything you blit is going to cause a page fault and slow things down more. Increasing the SRAM onboard the chips means you get fast access to the SRAM by nature of the higher speed of SRAM, AND you don't get page faults from the DRAM by accessing all over the place... the accesses to DRAM are confined to the framebuffer.

 

CPU MIPS has little to do with the speed og the texturemapping. In fact, the CPUs in most games on the Jaguar are doing nothing while the blitter is busy trying to copy data in the slow DRAM.

[Atari_Owl]

The Jag's two CPU's run at 26.6MIPS approximately equalling 53.2MIPS combined. this is totally independent from the Blitter chip which is doing the actual rendering onscreen. The bottleneck in the Jaguar is the DRAM, which is really slow and is even slower if you are moving data across a page boundary (page fault). Since it's impossible to store textures and framebuffers on the same page in DRAM, everything you blit is going to cause a page fault and slow things down more. Increasing the SRAM onboard the chips means you get fast access to the SRAM by nature of the higher speed of SRAM, AND you don't get page faults from the DRAM by accessing all over the place... the accesses to DRAM are confined to the framebuffer.

 

THis presumably the point of the section in the underground dox talking about using GPU RAM as a buffer.

 

Since the GPU RAM and DRAM aren't the same thing, then presumably the texture can be blitted into the GPU Ram in pixel mode (hopefully then there would be no page misses since the read from DRAM could be contained within a page and the GPU Ram is local to the chip) and therefore not affecting the DRAM page?

 

Then we would need to blit out the data in the GPU ram to the framebuffer?

 

Problematic if the GPU RAM is not dual port but not so much if the chips are, as Tbird says, mostly sitting idle waiting for the blitter.

 

Or am i badly missing the point?

 

Cheers

[Thunderbird]

The Jag's two CPU's run at 26.6MIPS approximately equalling 53.2MIPS combined. this is totally independent from the Blitter chip which is doing the actual rendering onscreen. The bottleneck in the Jaguar is the DRAM, which is really slow and is even slower if you are moving data across a page boundary (page fault). Since it's impossible to store textures and framebuffers on the same page in DRAM, everything you blit is going to cause a page fault and slow things down more. Increasing the SRAM onboard the chips means you get fast access to the SRAM by nature of the higher speed of SRAM, AND you don't get page faults from the DRAM by accessing all over the place... the accesses to DRAM are confined to the framebuffer.

 

THis presumably the point of the section in the underground dox talking about using GPU RAM as a buffer.

 

 

The idea would work great except for the fact that the GPU RAM is the only RAM that the GPU can execute programs from, so the GPU's RAM is usually completely filled to the brim with the CODE that the GPU is executing and there's no room to use it as a texture buffer... Besides the fact that 4K is pretty small for that use anyhow.

 

 

Since the GPU RAM and DRAM aren't the same thing, then presumably the texture can be blitted into the GPU Ram in pixel mode (hopefully then there would be no page misses since the read from DRAM could be contained within a page and the GPU Ram is local to the chip) and therefore not affecting the DRAM page?

 

Then we would need to blit out the data in the GPU ram to the framebuffer?

 

Problematic if the GPU RAM is not dual port but not so much if the chips are, as Tbird says, mostly sitting idle waiting for the blitter.

 

Or am i badly missing the point?

 

Cheers