What is so special about the R500?

[Acert93]

The R500 is the successor of ATI's unreleased R400. We know that ATI is going toward a Unified Shader architecture. This seems to be partly motivated by the fact that everyone (MS, ATI, nVidia) agree that shader programming for Vertex and Pixel Shaders is going to coincide in the future (although nVidia believes the PS and VS units should remain separate). A unified shading language is an expected feature of SM 4.0. Another motivation is the potential for better efficiency and performance in games that are either PS or VS intensive.

While I think Unified Shaders is a really great idea, I would like to pontificate out loud: Is this all R500 will do that is "new"? What new developments will come about that make R500 look new and "next gen"?

As the Beyond3D DirectX 10 article mentions, "there are several instances where you want to do things at the vertex level that’s mostly a pixel level operation, or something at the pixel level that’s usually done at the vertex level"--what are these, and will consumer/developers notice?

What else will we be getting out of R500? As a consumer with a strong interest in 3D, I am curious what else ATI/MS may be bringing to the table. Like most gamers, I want to know how Xenon will be "next gen". e.g. A recent patent seemed to indicate that the Xenon may use one of the POWER processors for Vertex operations. Does this mean that the Unified Shaders will almost all be dedicated to Pixel Shading? Also, we recently saw neat features like Geometry Instancing introduced by ATI (R300) and nVidia (NV40). Stuff like that is exciting and will affect how games are made.

What features from the DX Next article should we expect to be usable in games with R500? Are we just going to get more of the same with more power? Or should we expect some Topology Processor, specific Tessellator Enhancements, Frame Buffer Access in the Pixel Shader, Integer Instruction Set, some new AA technique with no performance hit ??? Are we looking at Fast14 allowing clock speeds to be so high we can expect shader programs with hundreds, maybe thousands, of shader operations for games? (I believe FarCry used 50ish for the water and HL2 has some in the 70s). Will vertexes be able to be created and destroyed at will? How will flow control, branching, looping, etc... make a big difference?

As you can tell I am curious. I would hope R500 introduces some new features that are usable. So, what else will R500 do? R500, more of the same with more power, useless innovations, or new approaches that make a difference to developers/gamers? Any hints or educated guesses?


Ps- Any chance of ATI using the Xenon/Revolution launches to push some of their personal HW ideas out the door for support to leverage on the PC and possibly WGF 2.0? Will anything other than 3Dc be introduced and make waves?

 

[Lazy8s]

Between ATi's new solutions for the next Xbox and GameCube and nVidia's new product for PS3, I hope at least one will fundamentally move away from immediate mode rendering. It sounds like nVidia may be hinting at that with their new architecture.

 

[Dave Baumann]

It sounds like nVidia may be hinting at that with their new architecture.

What gives you that impression?

 

[London Geezer]

First of all, nice post!! In a question-y kind of way, they're all very good questions that need to be answered, although i'm not sure anyoen will at this point in time, anyway, so i won't forget anything, i'll go the annoying puoting way of replying... And ask more questions too.

The R500 is the successor of ATI's unreleased R400. We know that ATI is going toward a Unified Shader architecture. This seems to be partly motivated by the fact that everyone (MS, ATI, nVidia) agree that shader programming for Vertex and Pixel Shaders is going to coincide in the future (although nVidia believes the PS and VS units should remain separate). A unified shading language is an expected feature of SM 4.0. Another motivation is the potential for better efficiency and performance in games that are either PS or VS intensive.
While I think Unified Shaders is a really great idea, I would like to pontificate out loud: Is this all R500 will do that is "new"? What new developments will come about that make R500 look new and "next gen"?

There are many threads on the 3D Tech and Hard Forums, but basically SM4 is much more than "Unified Shaders", which in theory could already be done today, if only it were to be specified at a driver level. I think one of the most important thing that will make thigns look more next gen is the fact that polygon performance will go up a lot. Same for fillrate, if rumors are to be believed that most next gen consoles GPUs will have embedded VRAM.

As the Beyond3D DirectX 10 article mentions, "there are several instances where you want to do things at the vertex level that’s mostly a pixel level operation, or something at the pixel level that’s usually done at the vertex level"--what are these, and will consumer/developers notice?

Oh i'd love to know that, but i also believe that ultimately, the consumer won't really care if a calculation is done on a pixel or vertex level. 99% of the consumers hardly even know the difference between the 2.

I guess all your other questions will be answered in due time

 

[Guden Oden]

I don't think any of us know enough about the chip to answer this poll, but as this forum's created to contain our useless speculation about things we have no clue, I will add mine anyway.

I doubt there will be very many useless features in nextbox's GPU. Quite the opposite, the design seems to be a tight integration of both CPU and GPU to cut down on inefficiencies and waste. I am a little curious about the plan to dedicate two full CPU cores to geometry though, as GPU vertex shaders have shown themselves to be vastly more powerful for that at a considerably lesser transistor expenditure, but the nextbox seems to be flexible enough to allow using all of the CPU or GPU resources for whatever task the programmer decides...

 

[Dave Baumann]

I am a little curious about the plan to dedicate two full CPU cores to geometry though...

Thats not "the plan" that is a suggested usage scenario - if there is one thing about the overriding feature of the whole design it appears to be that there are pools of processing resource that its up to the developer to utilise in a manner he sees fit. As the patent mentions those CPU cores could all be handling the OS, or some could be handling the physics or OS, physics, suround sound and geometry all distributed across the CPUS cores. To some extent this same flexability goes over the R500 as well - if the game has a "typical graphics balance" and then the graphics processor can be used as they traditionally are with VS and PS ops running on it (but the hardware / software managing the shader load on the unified structure), however if the developer has a particularly fragment shader intensive title/scene then he can choose to have the graphics core just fragment processing and the CPU's vertex processing (there may even be some automatic load balancing between CPU and graphics for VS ops as well).

Fragment processing, and possibly Vertex Texturing, will probably be fixed to the graphics processor, but other than that the design looks to be one that gives the developer large pools of processing resource that they can either control themselves or let the system allocate resource.

 

[overclocked]

If someone could just clear up one thing, is the "leaked" spec 48 shaderALU´s for R500. oh and btw did R400 ever taped out?

 

[London Geezer]

If someone could just clear up one thing, is the "leaked" spec 48 shaderALU´s for R500.

Clear up, meaning, is it true? Possibly. Eventually we will get there and i expect either 32 or 48 from R500 depending on when it finally comes out.

 

[tEd]

i really wonder if it has the same AA engine as in r3(4)x0. Dave?

 

[Acert93]

Some great comments guys Not answers, but some great things to chew on! The relationship between the CPU and GPU should be interesting. It sounds like it will give a lot of versatility to the developers to use the system in ways that cater more to their general needs. Having a system with CPUs' dedicated to AI and Physics is probably the most exciting part about the next gen for me personally. Interactivity, especially when it improves the game immersion and gameplay, are big for me. I cannot wait for a Football game where the motioncapture stuff takes a back seat to actually dynamic interaction. It is so lame seeing a tackle animation where the other defenders cannot interact/effect the tackle animation once it is started. Making some kind of animation blending ragdoll with real physics would be great. And in games like football, you finally will see a real difference between big players and small players. The possibilities are endless... if they have the time and money to do this kind of stuff

Anyhow, keep the great comments and ideas coming. It seems so far most people expect the R500 to be pretty powerful and have some new features that are usefull. I hope they are right!

 

[pc999]

Since the patents things like procedural tech, and MRM are been talked, to me that looks very next gen, changing the way that games are create (hope that dev start making big games again), or less pointy heads (for example).

 

[cthellis42]

Poll's not too complete. I rather expect R500 to be "more of same but faster" but neither add a lot of useless innovations or things that will make utterly stand out. I rather expect it will go in the same directions we're able to easily see and predict now. Will it go straight to a unified shader achitecture? Doubtful. Give hard-ons to oh-so-many people on here by pulling in the likes of tile-based rendering?

I expect it to be "more of same," operating a lot faster, with a concentration on more efficiency (through clever design and smaller innovations no doubt) and likely chock full of new features we already expect--and probably some we don't, a few of which are useless, and a few of which are cool and interesting...

I don't really see yawning or slackjawed shock as the only two options, though.

 

[Acert93]

cthellis42,

That is why there is a "Other. Please explain" option I did not want to make the poll too complicated since we do not know a lot of details, just gauging the basic "feel" for what we expect/know. Thanks for your comments. You seem to be debating all 3 categories at the same time!

 

[cthellis42]

<laughs> Yeah, I tend to fence sit a lot and debate/debunk the current rather than fantasize the coming. On this particular issue I have no specific insight, so I can really just evaluate overall trends and go from the gut.

I think it'll be the generation that comes along-with/just-after Longhorn that may contain more surprises than we expect.

 

[MightyHedgehog]

Unconventional deferred rendering design, unified shaders, faster and more efficient capability to perform high-quality AA. That's pretty much what I expect. Not necessarily revolutionary, but certainly innovative in its own way. More-integrated with MS' choice of CPU side hardware than perhaps expected...enough to be able to very freely change and share roles very well. Maybe a bit Talisman derived and made into something more realistic and usable. I don't really know tech-stuff, though...just my layman's guessing.

 

[London Geezer]

Well, i think the most important fact about the R500, or whatever chip will be used on Xbox2, is that it will be a chip that the developers will be able to exploit to the maximum, unlike its PC counterpart.
It might be missing some unknown features that the PC brother might have, but in the end the XGPU2 will enjoy a status that is much more important than any other feature: it lives in a closed platform.

Because of that, games on Xbox2 will look far better than 99% of PC games released at the same time, and they will look better than PC games for a long time, until the PC parts are at least one generation beyond the R500.

Sadly PC games target the last 2 generations of hardware, so the latest offerings are hardly ever fully used.
That's what's interesting about R500. It's like having a high-end PC finally being used to a much bigger potential than it would usually be. I'm sure i'm not the only one to get a little frustrated at buying very expensive PC hardware that is never really used apart from having higher resolution or the odd bump in AA/AF speed...

 

[Dave Baumann]

Sadly PC games target the last 2 generations of hardware, so the latest offerings are hardly ever fully used.

Thats a generalisation. Try running Doom3 at reasonable resultions with the detail settings very high on hardware two generations back; likewise with Far Cry and HL2 (at the periods they were release). These targets are entirely developer and publisher dependant - some do target older hardware, other taret newer, high end hardware for full details and scale back various options from there for older hardware. Other (EA) are crossplatform titles hence their detail levels are very much dictated by the capabilities of the console platforms they are also targetting - I can't wait for the new consoles since I expect we'll begin to see a fairly large jump in geometry levels which should mean that cross platform titles will begin to utilise the level of geometry processing presently available on PC hardware.

 

[London Geezer]

Sadly PC games target the last 2 generations of hardware, so the latest offerings are hardly ever fully used.

Thats a generalisation. Try running Doom3 at reasonable resultions with the detail settings very high on hardware two generations back; likewise with Far Cry and HL2 (at the periods they were release). These targets are entirely developer and publisher dependant - some do target older hardware, other taret newer, high end hardware for full details and scale back various options from there for older hardware. Other (EA) are crossplatform titles hence their detail levels are very much dictated by the capabilities of the console platforms they are also targetting - I can't wait for the new consoles since I expect we'll begin to see a fairly large jump in geometry levels which should mean that cross platform titles will begin to utilise the level of geometry processing presently available on PC hardware.

Yes it was a generalisation, but i was really talking about geometry levels. Any GPU released in the last 2-3 years will be able to show a lot more geometry than in many games released today, including Doom3, shadows or not.
Geometry is what i was thinking about, there should be a lot more "stuff" on screen than there is in 99% of today's PC games.
Or maybe i'm expecting too much.

 

[AlNom]

Even a few of the latest PC games seem to look better only because of texture resolution. Doom 3 or Chronicles of Riddick for examples...

 

[jvd]

Dave , I really think developers design games with 2 generations back hardware and then scale foward and add one or two new features to the game .


Doom3 was made with the geforce 2 gts feature set in mind and scaled foward using shaders to speed up some of the things that the geforce 2 would need fillrate or other parts of the chip to do .

Half life 2 seems like a dx 8 game with dx 9 to speed up the game and make the effects look better .

Not to mention that anyway you cut it , having a few choices of hardware to develop for and having to have it work on a wide range of them is going to make for a lower quality engine / game than if you took 1 video card like a 9700pro , 1 cpu like a athlon 64 , 1 memory amount and 1 sound card and dveloped for it