Creative 3DLabs high-end 3D cards

COMPARED TO the desktop PC 3-D graphics, which basically started from nowhere after Microsoft got Direct3D for gaming, the OpenGL-based workstation 3-D graphics on the PC had a more tumultuous history.

You won't see many of the early players today. Remeber the likes of Intergraph and London-based 3DLabs? Both merged, and then submerged into Creative Technology. Accelgraphics with its excellent drivers and support? Well, bought by Evans&Sutherland, and quietly disappeared from the scene. And of course Silicon Graphics' ill-fated Wintel workstation effort by Micro$oft-oriented Rick Beluzzo - in fact, he was so well oriented to that direction that, after Silicon Graphics was ruined, he got a VP post at Microsoft right away.

Many things that only now gain importance in PC graphics were seen as critical decision-making features then. For instance, polygon and line performance was important, but users demanded to see its consistency with various sizes of triangles and lines as well. Handling hidden surface culling, precise geometry without malformed polygons due to FP errors, and proper high-quality antialiasing with as little as possible performance penalty were just as important as pure performance numbers or graphic memory size. Most importantly, since these were commercial and often mission-critical applications, reliability and error-free, expected results were (and very much still are) a must. That is why sooo much effort went into the driver development as well as specific application certification and optimisation, plus regular re-checks and updates for each app as it gets refreshed, compared to the gaming 3-D - and that is why Nvidia and ATI have to ask much more for their Quadro and FireGL cards, since the extra driver and support cost gets amortised over a small number of users.

Gam(bl)ing In Pro 3D
Today it is the PC gaming 3-D vendors Nvidia and ATI that, through sheer force of their sales volumes, dominate the workstation 3-D market as well. The recipe was simple - ensure that your 3-D chip handles OpenGL at least as well as DirectX, then only enable full OpenGL performance on specific card versions that would be sold to workstation users at a premium to cover the extra software costs, but still cheaper than what a niche workstation-only 3-D card vendor would ask. Would the 'pro' user give up on trusted brands and support these gaming giants yet newbies in the 'serious' arena?

Well yes, that gamble did work - despite their excellent products, the Intergraph/3DLabs combo was almost put out of business by the time 'white knight from Singapore' Creative's Sim Wong Hoo came in. Evans&Sutherland completely gave up on the generic workstation market even before that, making their very purchase of AccelGraphics pointless. Both of them couldn't compete with PC gaming 3-D vendors' frentic performance race and frequent graphics chip revisions, not to mention new major generations coming out yearly.

While there are no precise, really neutral, statistics due to the multitude of small whitebox vendors and 'self-assembly' power end users in this arena, as of now Nvidia Quadro line should be leading with over half of the market, ATI FireGL cards have around one third, and the reminder goes to Creative 3DLabs. Can this ratio be changed at all?

Well, ATI is trying to fix its OpenGL for quite a time by now - faster and better OpenGL run, plus more useful free applets like what Nvidia has with QuadroView, should help ATI regain some share since they are generally much cheaper than Nvidia in this segment for similar raw card performance. Yes, Nvidia now charges really quite a bit for their cards: the cheapest you can get the fairly old Quadro FX3000 in the open market is around US$ 1,300 (and over US$ 1,900 for Quadro FX4000 - if you can find it, that is). Compare this to the GeForce 6800 Ultra at around U$ 550, or 6800 GT (same GPU speed as FX4000) at US$ 450 - the premium becomes a bit too much to swallow.

New (old) kid on the block
So, did the market share gains give Nvidia too much ego boost? Hard to say, but Creative's boss, Sim Wong Hoo, has big plans in store for its 3DLabs operation. After all, it has decades of combined high-end 3-D graphics expertise of Intergraph and old UK-based 3DLabs, and could conceivably create 3-D chips that would compete (or exceed) the best of Nvidia and ATI not just in performance, but also in features and image quality. Creative is cash-rich and, if they want so, they can fight this game to win.

The first product line out of new entity are the recently launched Realizm cards - boy, do they look impressive. The high-end Realizm 800 makes Nvidia SLI look silly by comparison - all that SLI tries to accomplish with two cards and two PCI-E slots, and much more, is done within a single card (and single native PCI-E x16 slot - albeit the high-power 150 W one). In SPEC ViewPerf 8, even the lowest-end Realizm 100, retailing for around US$ 900, beats the Quadro FX4000 in all but one test, at less than half the price. The Realizm 200 gives double the RAM (512 MB - matching the new QuadroFX 4400) at around US$ 1200.

Realizm 800's architecture - with two graphics processors (VPU's) in parallel, connected via separate 4 GB/s paths to the dedicated geometry processing & PCI-E x16 interface unit (VSU), reminds us of the large heavily pipelined graphics geometry engines from old-time Silicon Graphics and 3DLabs days - after all, parallel rendering engines did exist well before Voodoo SLI.

Each VPU has its own 256-bit GDDR3 memory path with 256 MB memory, however, they function as a single logical 512-bit memory path and unified 512 MB memory, removing the need for texture and other buffer duplication that SLI has - more net memory is therefore available for higher multi-screen resolutions, finer antialiasing etc. Of course, the 512-bit path gives the card enormous total bandwidth and, with the VSU doing the geometry and keeping the display list in its own dedicated 128 MB GDDR3 RAM, the two engines can concentrate on the pixel and texture operations. Finally, this is the first 3D graphics pipeline I know of that can display 16-bit per-colour FP values for final display output pixels!

Can it fight the behemoths?
So, technically, these are great devices, and the path to upgrades is clear - for instance, if there is a need to upgrade the geometry performance, but just add extra memory, Creative only needs to revamp the VSU, while the two VPUs can stay the same, just have say 512 MB each instead of 256 MB - it saves a lot of new design costs and helps somewhat prolong the life of each chip. Remember that, in professional OpenGL 3-D, pure geometry performance is far more important than in games, since a huge portion of modeling and visualisation work, before job completion, is done on simple, often textureless but polygon-rich, 3-D models.

However, both Nvidia and ATI have working parallel GPU technologies as well. Nvidia SLI may not be as elegant and efficient, but it can work also on el cheapo PC game cards like the new GeForce 6600, for instance. ATI's work, on the other hand, has been proven in big SGI graphics engines where many ATI chips work in parallel. Their GPUs do come from the gaming background, but ultimately they can also scale to any required performance level.

In the opposite direction, Creative will most probably use some of its financial muscle to attempt to bring the Realizm technology to the mainstream PC market, too. After all, the chips supposed DirectX 9 just as well, and even the low-end single-VPU cards could probably hold their own in the games - assuming the drivers are tuned well. Since XGI is not really seen in this market segment, it will be the "triple threat" match in the luxury "workstation wrestling" ring. However, it may be the long-term success (or lack of it) of Creative's mainstream 3-D sales push to increase the overall volumes, more than how well the UK and US brains of 3DLabs fight the two behemoths, that determines the outcome of this competition.