|3dfx goes on a Rampage. Written by rashly. Edited by Atif Butt. 7/8/2002.|
|Editor’s Note: rashly is the owner of Rashly Productions, and creator of arguably the most potent 3dfx fan site on the internet. We recently contacted rashly for his opinion on 3dfx’s past product releases and business decisions. rashly took the liberty to go one-step forward and hypothesize on future nVidia offerings making use of 3dfx technology as well. The ensuing article is the product of our request, as presented by renowned 3dfx fan and aficionado rashly. It’s a very interesting read that we hope you thoroughly enjoy.
Ever since the demise of 3dfx on December 15, 2000, there has been great speculation on what 3dfx could have done if they were still in the market. What if they had another year? Would the infamous Rampage bailed them out of financial turmoil?
A little primer here: (for those of you that don’t know), 3dfx was the biggest pioneer in 3D gaming on the PC. Their Voodoo 3D accelerator revolutionized PC gaming, making it a viable platform able to compete with console systems of its day. Sure, 3dfx wasn’t the first to offer 3D acceleration on the PC, but they were able to make it more mainstream than ever before. With the release of the Voodoo Graphics chipset (and later the Voodoo2) 3dfx had put a headlock on the then-feeble competition. Everything in 3dfx-land was dandy until some questionable business decisions came into play following the release of the Voodoo2.
Many are not aware that 3dfx’s infamous Rampage (next-generation) graphics chip was scheduled to be released immediately following the Voodoo2. During this time 3dfx acquired STB, and was over-going a transition in which they would be working with new employees, and manufacturing video cards in-house (rather than selling graphics chips to third-party board manufacturers).
As a direct result of the pending changes at 3dfx, there were some delays in the production of the next-generation Rampage chips. With the ensuing threat of a 32-bit, marketing-pumped nVidia TNT2 release, 3dfx couldn’t sit out this product cycle entirely. They released the Voodoo3 (essentially as a placeholder) until Rampage was ready for the big-time.
It should be stated here that the Voodoo3 was a solid product. At the time of its immediate release, it had the industry’s fastest 3D, and arguably the best 2D image quality as well. What single-handedly hindered the Voodoo3’s success was the release of nVidia’s TNT2 several months after Voodoo3-based video cards hit store shelves.
The TNT2 wasn’t as fast as the Voodoo3 on paper, but it offered faster Direct3D and (many times) OpenGL framerates than 3dfx’s offering. Perhaps more importantly, 3dfx lost the marketing battle fought by nVidia in regards to 32-bit color support. Even though the TNT2 couldn’t necessarily run games at desirable framerates in 32-bit mode, many consumers saw "32-bit" and noticed the numerical value to be higher than 3dfx’s claimed "16-bit" rendering and based their buying decision upon this fact.
Despite 3dfx’s attempts to stress that 32-bit rendering was not necessary at that time in gaming, and their efforts to stress that the Voodoo3 made use of a quality 22-bit post filter, the general public bought into nVidia’s side and 3dfx lost substantial market share to nVidia. Thankfully for 3dfx, Voodoo3 video cards still sold well, and 3dfx wasn’t in any real trouble, yet.
3dfx's tumble downwards
In purchasing STB, 3dfx immediately lost the OEM backing they once had in the past. Seemingly every OEM that 3dfx had left behind placed an order with nVidia for their graphics chips. Indirectly, 3dfx had dug themselves into a hole. They would be forced to produce all of their newer video cards in-house or sell to third-parties at lower prices. They chose to stick with in-house production just as they’d done with their Voodoo3 cards.
Unfortunately, 3dfx was having trouble releasing their next line of chips, the VSA-100 series in a timely fashion. The original plan was to release the VSA-100 based Voodoo4 and Voodoo5 in early 2000, but they weren’t ready for the spotlight in January, or February, or March… (you get where this is going). Not until the summer season did consumers find VSA-100-based products on store shelves.
Instead of competing with nVidia first-generation GPU video cards, 3dfx was going head-to-head with nVidia’s refined GeForce2 product line. To 3dfx’s credit, the VSA-100-based video cards had significantly better image quality than their GeForce2 counterparts. (3dfx had vowed to improve the image quality beyond their competition following all the snubbing incurred from the 32-bit vs 16-bit debate months earlier). But the masses were looking for speed alone, and as such, the GeForce2 handily snatched the crown from 3dfx’s kingdom. 3dfx had lost the speed game they so handily dominated years before.
To add to the misery, 3dfx didn’t ship nearly as many Voodoo5 video cards as they had initially projected. They had a surplus of memory chips and circuitry left over because demand was not increasing for their products as time progressed.
Despite having purchased GigaPixel in March of 2000, the hyped Tile Based Rendering technology wasn’t implemented soon enough to help alleviate memory bandwidth issues. 3dfx continued to use the traditional "brute force" approach, loading their cards with an unheard of 64MB and (proposed) 128MB of video RAM. With sales for Voodoo5 cards lagging, 3dfx had purchased too many RAM chips and soon found themselves in debt to memory companies.
The $186 million dollar purchase of GigaPixel was, in my opinion, a bad move by 3dfx because they had already missed their proposed deadline on the Voodoo5 and were in no position to purchase another company’s assets at that point.
The Voodoo5 6000 as a commercial product was scrapped in the fall of 2000, and 3dfx made one last hurrah in the OEM market by licensing their chips to PowerColor. It was already too late and 3dfx had one last weapon that they hoped would save them from what appeared to be (and eventually was) a bleak future.
Going on a Rampage
Rampage. Say this word to any avid 3dfx fan, and they’ll surely tell you something along the lines of "it would have saved 3dfx."
Rampage was the internal name for 3dfx’s next-generation part intended for direct competition with nVidia’s GeForce3 product line. 3dfx learned from their mistakes and pulled out all stops while making this card. It was to have had the power and image quality to remain unmatched by the competition’s offerings. As far as performance is concerned, the Rampage would have been slightly speedier than a GeForce3 offering. However, the Rampage’s image quality and feature set would have easily separated it from the competition.
The high-end Spectre offering was to make use of two Rampage chips, and one Sage unit. The Sage unit wasn’t a "smart chip" of some sort, but rather, it was 3dfx’s external transformation and lighting unit that supported twenty-five hardware lights. Since the AGP specification can only detect one chip on any given video card, the AGP slot would detect the Sage chip, which would be supported by the Rampage chip(s). The DDR memory bandwidth would have been rated at 12.8GB/sec, easily slaughtering the GeForce3’s 7.36GB/sec.
One of the most-advertised features for Matrox’s new Parhelia offering is its 10-bit per RGBA component processing. Well, Rampage would have layed the smackdown on Matrox as well. Rampage was set to feature 13-bit per RGBA component, allowing it to have 52-bit internal color processing. It still made use of 32-bit output, but the quality, and precision of colors would have been of a much higher standard than anything offered by today’s commercial video card products.
3dfx’s beloved FSAA was planning a return as well. It would have still used RGMS (rotated grid multi sampling), but Rampage’s texturing abilities would have directly affected how anti-aliasing worked. The M-Buffer (similar to the Voodoo5’s T-Buffer), would have allowed for 4 sample AA per clock with no pixel rate loss, unlike the older T-Buffer, which took a 4x performance hit when AA is enabled.
The only downside would have been that MS takes the same texel coordinates and jitters them, thus there's no texture clean up. 3dfx implemented the advanced anisotropic filtering to remedy this once and for all. It could do up to 128-tap anisotropic filtering. If quad texturing was used, Rampage would provide for (performance-hit) free FSAA. When dual texturing was used with anisotropic filtering, Rampage again would have provided (performance-hit) free FSAA.
Several other features would have been brilliant as well. Rampage continued to feature support for the cinematic effects from the T-Buffer, and probably would have included additional effects not seen in Voodoo5 products. It would have featured photorealistic rendering as an extension of the Rampage’s VTA. Finally, it was planned to include hardware Photoshop effects (toon shading).
I honestly believe this would have revolutionized the industry because other chipmakers surely would have followed suit. Unfortunately, 3dfx went out of business just when Rampage-based products began initial testing in 3dfx’s labs.
Other chips were planned for production after Rampage’s release. Unfortunately, there is little known about them, and all what is heard about them are rumors. After Spectre was released, it has been said that the next product line would have been from a line known simply as Fear. Fear was basically a refresh to the Rampage, just as the GeForce4 is to the GeForce3. It would have made use of Sage II as its bus master. After Fear had its run in the market, 3dfx was said to have planned to release Mojo to the world. Mojo would have made use of a radically new architecture created in large part using technology developed by Gigapixel. It is supposed that Mojo would have had made use of an advanced tile based rendering technique and most likely kicked some serious b00ty.
3dfx's life after death
The NV30 from nVidia will be the first chip to incorporate 3dfx technology acquired from the 3dfx buyout. When nVidia got ahold of 3dfx’s technology from the buyout, they mixed nVidia and (former) 3dfx staff. They broke them up into two groups and had each group argue for different technologies to be included as a part of a radically new video card product. One group argued for technology nVidia had been culturing in-house while the other argued for technology created by 3dfx engineers. In this way, former 3dfx employees and longtime nVidia veterans collaborated with one another on creating the most revolutionary and arguably effective product possible.
My expectations for the NV30 do not include seeing the advanced tile based rendering (developed by GigaPixel) in action. The technology is great, but it’s too drastic a change from nVidia’s current chips/architecture. NV30 will likely feature an external hardware transformation and lighting (T&L) unit (like the proposed 3dfx Sage) and it will be as powerful, or more powerful than Sage would have ever been.
I would surely expect to see NV30 make use of some of 3dfx’s FSAA technology. The FSAA on the Voodoo5 is just beginning to be equaled today by the GeForce4. nVidia will have to figure out a way to use Rampage’s FSAA (optimized by nVidia algorithms of course), and get it to work with the NV30’s new texturing unit.
The NV30 will be SLI capable, but nVidia will surely leave the SLI solutions to the workstation market in the form of their Quadro followup. nVidia definitely won’t be keeping the “GeForce” name as it would imply that the NV30 is a new product based upon the same “GeForce” architecture (which it definitely is not). On a side note, no matter how cool it would be to see it happen, nVidia won’t be adding GLide support to the NV30 so don’t stock up on the original Unreal just yet.
These are of course things that nVidia should do. It’s entirely possible that nVidia will be stubborn, and not use any of the purchased 3dfx technology in the NV30. Of course, this would make the buyout all but pointless (oh wait, they eliminated their biggest competition) and I, for one, will be very disappointed, as will many 3dfx users, I’m sure.
3dfx lovers unite!
To this day, the 3dfx community is quite strong. More than users looking to solve Windows XP-Voodoo3 compatibility issues, these people have actually became close friends. The hub of the 3dfx community would have to be the x3dfx forums where any/all 3dfx-related questions can be answered. As far as drivers are concerned, have no fear. To this day, all new games work with the Voodoo5 (including the soon to-be-released Unreal Tournament 2003) and are playable as long as they are tweaked properly and are teamed with a fast CPU.
The team at 3dfx Underground has taken on the task of essentially re-writing new drivers from the ground up. Thus far, the beta drivers are amazing. There is full Windows XP support, and (soon) DirectX 9 and OpenGL 1.3 compatibility!
The 3dfx empire revolutionized 3D graphics forever. If it weren’t for them, graphic performance, and quality would have been at least two years behind where it is today. It’s just too bad that Rampage never made it out. For 3dfx its release was so close, and yet so far.