What's significant beyond DX9?

[dbeard]

Let me see if I can explain my question. Since I have been interested in 3D hardware from a gaming perspective, I have seen 4 industry changes take place.

1. Hardware acceleration
2. Consumer T&L
3. Shading Technology
4. General Programability

From my perspective, hardware acceleration was such a benefit that the games in development quickly switched over to making this a requirement to play the game. This directly impacted me by having to upgrade as a price of admission for these new games.

Apparently, consumer Hardware T&L was not a great direct benefit to game developers. I suspect this is due to the rapid increase in speed of CPUs. Unless I am mistaken, the CPU was fast enough to adequately do the T&L since we were fighting fill rate and bandwidth issues throughout this period. Once the industry got insane FPS at 1024*768, I lost interest on that front and started watching FSAA/ansio numbers at this resolution. This hasn't directly impacted me as I haven't been forced to upgrade to HT&L.

Shaders have only been in the landscape for a short period. They must offer some fairly significant benefits to game developers since they are already being used to limited extents in several games. Unless I'm mistaken their uptake has been faster than for Hardware T&L.

As an extention of the shading comments, general shader programability seems to me to be almost as important as hardware acceleration was in the begginning. To me, this offers numerous benefits to game developers. These new languages allow a game developer to create a game with a single code paths that should provide wide support for the market as a whole. The first would be the path that points to the runtime compliation version. The game developer would probably add a second or third for specific top of the line graphic hardware but I suspect that this would be more about improved performance rather than additional feature set. That seems to be what JC is doing in DIII.

So assuming all of this is somewhere in the neighborhood of accurate, what comes next in graphics? Higher order surfaces? Do they offer a step change in advantages on the order of hardware acceleration? Any other items come to mind that would make game developers move towards incompatibility with DX9 hardware?

Since I can't think of any, it seems like a DX9 card will be compatible for a LONG time. It might not be as fast as some would like but compatible.

Thoughts?

 

[Tahir2]

1,2,3 and 4 all link together pretty tight. You can't have progressed from one without the other. Now especially it seems to me that 3 and 4 are linked in such a way that this is where the next level of advnaces will come from. I believe for the moment we have reached a stage where we do not need to worry about fillrate or bandwidth as much as we used to. Compression technology and other resource saving techniques have made FSAA truly feasible with amazing quality.

I agree I do not think we will be going into the realms of pure general programmability in the next 12 months (that is what the CPU is there for afterall!)... but GPU's/VPU's are taking over some of the tasks of the CPU slowly but surely. The next step is more speed in shader execution rather than anything else. After that, I have no idea - but the goal is photo-realism and whatever hack makes this a closer step in realtime will be the next big thing.

 

[Jare]

Apparently, consumer Hardware T&L was not a great direct benefit to game developers. [...] Unless I'm mistaken their uptake has been faster than for Hardware T&L.

I disagree. The thing is, it is not terribly hard to fallback from hardware T&L to the emulation provided by Direct3D, aside from the increased amount of geometry you can expect T&&L cards to cope with. With shaders, especially pixel shaders, things are different because they can do significantly different things and a software fallback path with identical features is not so simple.

 

[JF_Aidan_Pryde]

Hardware acceleration gave us simple 3D texture mapped objects. With DX7 and TnL came the possibility of more polygons and faster transformation. With DX8 came shaders that simulated solid materials faily well. The shaders generate pretty good effects for metal, plastic, chrome - basically the hard stuff in the world. DX9 with higher precision, full programmability will give us the ability to render the 'soft' stuff with realism. Skin, fur, light sensitive materials will be much more realistic with DX9. Other improvements noteworthy are HDR rendering, tone mapping and better camera simulation effects (depth of field, motion blur etc.)

Beyond DX9 the improvements will be focused on lighting. Since most of what we see are lit indirectly, ray-tracing will be essential. All the buzzwords radiosity/photon mapping etc will be comming closer. Basically, beyond DX9 we should see greater improvements by improving hardware lighting capabilites from simple direction lighting to indirect, reflected/refracted lights.

 

[epicstruggle]

other than lighting, what else can we see might get added on to dx9+. Anything related to longhorn (next windows os). I hear they want to make a big jump to a 3d desktop/experience. I would assume that they will use DX to power this new GUI. hmm wonder if they will require you to have a graphics card upgrade for it.

later,

 

[KimB]

Longhorn won't be pushing 3D graphics technology. It will just be pushing minimum 3D hardware specs for peoples' PCs, and will reduce the effectiveness of windowed 3D rendering.

 

[ET]

Apparently, consumer Hardware T&L was not a great direct benefit to game developers. I suspect this is due to the rapid increase in speed of CPUs. Unless I am mistaken, the CPU was fast enough to adequately do the T&L since we were fighting fill rate and bandwidth issues throughout this period.

The CPU is never fast enough. The direct benefit of T&L is that it allows more polygons to be used, and reduces the desire to program the math directly for performance sake. Fill rate is rarely a real issue. You can always increase speed by lowering resolution, but you can't do that as easily with geometry. This has been true, IMO, the standard game resolution became more than 640x480. The only place where you really need high resolutions is for detailed user interfaces, and games that have these typically don't require high frame rates.

 

[Maverick]

If you're doing software T&L then you have to send all of your geometry over AGP, or even *shudder* PCI, rather than keeping it in vidmem...

 

[SpellSinger]

PCIExpress will change things quite a bit. IHV's will have to design their ASICs for this new bus. This will be costly for the industry and IHVs will need Longhorn to help drive sales for these new graphics boards and motherboards. Longhorn better push the envelope cause old boards and ASICs won't work in 3GIO motherboards. The whole industry will need to upgrade fairly quickly. AGP will be considered legacy about the end of '05. Just in time for Longhorn and DX10 8)

 

[ET]

If you're doing software T&L then you have to send all of your geometry over AGP, or even *shudder* PCI, rather than keeping it in vidmem...

I don't think it'd be a problem for AGP, considering CPU T&L speed vs. AGP bandwidth. Even with PCI, what's the bandwidth? 132MB max? You need, say, 40 bytes per vertex? That's over 3 million vertices per second. High end CPUs will be faster than this, especially with simple lighting, but I can't see many users with high end CPUs using PCI cards.

 

[alexsok]

Most definetly, the prospects for further progress are pretty exciting indeed...

As has been mentioned here already, Raytracing, Radiosity Lighting, MonteCarlo alghoritm type techniques, Photon Mapping, Global Illumination (see http://www.openrt.de/Gallery/), etc are all the features the industry will strive for in the following years.

While many people are firm believers that higher than 128-bit precision is needed, my stance regarding this is along the lines of "don't fix what ain't broken", since 128-bit precision is pretty much enough for everything, unless we're talking about offline rendering, where there are pretty much no borders or boundaries...

The most important thing that should happen though, and one that would definetly boost the quality in video games by several inches is a faster adoption of the never ending wave of technologies each vendor offers, and there are really plenty of tools at the developer's disposal now to make this a reality (i'm talking about HLSL of course).