I’ve been worried as of late. No, it’s not the war in Iraq, or the scratch in my car; the cats and kids are fine. No, I’m worried about the AIB business, particularly the game AIB business.
Don’t get me wrong—the new breed of AIBs are fantastic. Unbelievable amounts of power at reasonable (albeit high) prices, on a BOPS/dollar comparison the new GPUs and VPUs are much less expensive than the current litter of x86 offerings. But the AIB suppliers have hit that white water that the CPU suppliers have been in for about two years—the users are saying, so what? Up until this year the GPU suppliers (and for ease of writing I’m using “GPU” to mean GPU and VPU) had an anxious and enthusiastic community of users who couldn’t wait to get the latest fastest baddest AIB. Now they’ve got them but they are still playing their old games. Uh-oh.
But not to worry; DX9 Shader Model 3.0 games are on the way. And a patched version of “Far Cry” is already available. “Doom 3” and “Half-Life” will be ready really soon. No, really. Honest. And others are sure to follow. But is it enough? Good enough; that’s been Intel’s explanation for the mediocre 865G IGCs they shipped 78.3 million of in Q1’04. Soon the tune will change, and that spells trouble in game land (and a few other places).
Intel’s new IGC, the i915G, is a solution for DX9 Shader Model 2.0 (SM2) level games. The Intel Graphics Media Accelerator 900 (or GMA 900) has four pixel shaders, supports 24-bit color, and can export 32-bit values. Basically, GMA 900 supports all the same formats as the 865G chipset with the addition of 2-10-10-10 format. Max Poly/Sec is 17.5MV/s. Given that it uses the CPU for geometry processing, it is obviously bound by what the CPU can deliver. However, the CPU is so damn good now, and the memory is so much faster today, that the arguments about memory bandwidth bounding on the graphics are almost moot, and truly are moot for DX8 games and earlier.
OK, so the 915G is wonderful—so what? So this: it’s good enough to wipe out low-end AIB sales no matter how cheap those old AIBs are. That leaves the high-end market for the GPU and AIB suppliers. This isn’t a news flash, either, but the GPU suppliers and their AIB partners need an expanding market to support their PEs. Where’s it going to come from?
Well, the game market has been the driving force and demo darling of the GPU suppliers for the past five to six years. (And from now on I’m going to use GPU to also include AIB.) With the advent of SM3.0 things have changed. Just one example is Mark Daly at Nvidia. Mark, in addition to being a really great guy, is now a VP at Nvidia and has a team of over 24 inside people plus who knows how many outside contactors. Doing what? Demos. A demo team that’s bigger than the entire engineering team at some of these companies just a few years ago has become just another cost of doing business. That’s truly impressive, but it’s expensive—you gotta sell a lot of GPUs to pay for a team like that plus the architects and engineers who design and build these wondrous machines. And so the commanding generals at ATI, Nvidia, XGI, and other GPU places spend inordinate amounts of time leaning over their planning tables trying to get the right balance between ASP and unit sales based on all the intelligence their agents can bring in.
But their plans may be flawed; they may have underestimated the enemy and overestimated their allies.
The game developer’s challenge
If the game developers do not embrace and accelerate their use of the powerful new features in today's GPUs, the expansion of the high-end gaming market will never materialize. New high-end users need to be brought into the market. They will come from casual gamers, some console gamers growing up and looking for something faster with better graphics, and new users who are attracted to the new cinematic interactive high-quality games (we've all been expecting, waiting for, and promised).If the high-end game market does not expand, then the GPU suppliers will lose market share to less expensive but ever improving IGCs combined with fast CPUs. These all-Intel systems will provide more than a “good enough” experience for the gamer, and especially the casual gamer and the console graduate.
And, just to make matters worse, shocking news came out recently from Hardocp that CPU counts more than GPU. In a recent series of tests, it was demonstrated that the CPU has a greater impact on the game image quality and consequentially the frame rate than the GPU does. This suggests the game developers are not taking advantage of the GPU’s capabilities as much as they could and should and that high-end boards from ATI and Nvidia don’t contribute that much.
The game developers say, in their defense, that they have to target a certain CPU level of the installed base of PCs in order to reach the widest possible audience (customers) so they can get the best return on their investment (ROI)—which has also grown and rivals movie production in terms of cost, staff, and time. Game developers, however, have to understand the danger they face if they “dumb down” their games for the lowest hardware. It risks consumers becoming indifferent to their efforts.
Having said that, it would seem the game developers no choice but to go to a multi-level (in terms of performance) game design. The chart on page 22 shows the issues as I see them. Basically, it shows the new Grantsdale IGC (i915G–GMA) as a solution for DX9 SM2–level games.
That being true, then the differentiation and payoff for the consumer who has purchased the latest PC and the latest AIB has to be dazzling cinematic effects, which may not have any effect on game play—just eye candy (shadows, reflections, great smoke, hair, etc.).
If that is true, then the market of consumers, who will be willing to pay for the latest PC and AIB, are reduced to the Connoisseur, and not expanded to draw in the casual player. This shrinks the market for the high-end AIB suppliers (like ATI and Nvidia), and the game developers are not doing anything to alter the situation.
Two games in one?
If a developer wants to make fundamentally different game play with more advanced parts, he essentially needs to design two games! He would need to make the game play for the lower (least common denominator) parts, and a different version for the more powerful parts. This isn’t really practical for the whole game, but it could certainly be done with a mission pack or extra bonus level that is super cool and only runs on the more advanced parts.That is easy to do with artists/pro-grammers generating two or more tiers of rendering—for example, an SM3.0 path, and an SM2.0 path with multipass. Arguably, the game play is different if the realism and “immersion” are better.
Game developers look at it differently, in terms of “Immersion” and “In-volvement.” Involvement is all about the gaming modes, the scenario, the AI, the interaction system. Immersion is about the quality of animations, graphics, sound, physics, the coherency of the world (the better it is, the more you believe in the world that has been created, the more you spend time in this world). Game players tend to choose their games according to the Involvement factor, and then are willing to spend a lot of mon-ey in order to get the Immersion factor as high as pos-sible. In a way we can say that SM3.0, higher CPU configuration impact the Immersion factor, not the Involvement, but that it can still have a great impact on a player’s attachment to a game.
Some players are more attracted to Involvement; others are more sensitive to Immersion. Immersion-sensitive players on a PC—as well as the ones that are competing online—are driving our industry as they are the ones who buy the latest hardware. But can we attract new ones to the category?
The game developers will wake up and produce new genres and new types of game play to take advantage of the new hardware and compilers. But they are going to have to get hurt (again). I expect to see a new wave of indie games come along and show the way, just like the movie industry has to be retrained from time to time. They won’t have the mega-budgets and be three years late just to show shadows and hair; they’ll have interesting and challenging stories, and not just bigger, louder guns. And in the meantime the GPU suppliers will have to look for new market opportunities to carry them through this trough of inadequate games.
