Original Link: https://www.anandtech.com/show/695



Although the CPU industry has definitely increased in competitive levels over the past year it cannot compare to the undoubtedly cutthroat competition that is present in the graphics core logic market.

The graphics industry was not always this bad - if you remember, it was not too long ago that 3dfx, Matrox, NVIDIA and S3 all had competing solutions out in the market, all of which provided a fairly competitive, yet livable environment for both the consumers and the manufacturers.

However a major change occurred, driven by the extremely aggressive nature of one manufacturer in particular, that has changed the face of the graphics industry going into the new year. Leading up to that point, was a very bumpy road for many of these companies.

The transition of the market from 2D accelerators to actual 3D accelerators was what caught the initial kings of the industry off guard. Seemingly overnight, we saw the fall of ATI, Matrox and S3 from the top of the market, to the very bottom. Not necessarily in market share, but in the eyes of the public which would lead to the loss of control.

The company that came in out of nowhere and took the industry by surprise was 3dfx, a manufacturer with no real track record and a single product under their belt, the Voodoo graphics chip. 3dfx's solution wasn't even capable of 2D acceleration, an early indication of where the market was concerned with performance. What the original Voodoo did provide was exactly what the gaming market wanted - 3D performance at $300 or less.

Shortly after the introduction of the Voodoo came competing solutions from the little known creator of the NV1, NVIDIA. And before the end of 1998, Matrox and S3 had also released their comeback-kid parts, aimed at getting themselves back into the graphics market.

Eventually, ATI joined the quest for the fastest 3D accelerator on the planet, however we feared that they had entered the game a little too late, with their Rage 128 solution not hitting the streets until early 1999.

Looking at the way in which this 3D revolution was started in the brief overview we just presented, you would assume that 3dfx and maybe S3 and Matrox would be on top of the market currently. As we very well know, that is definitely not the case. 3dfx is on the path to closing their doors, S3 has pulled out of the add-in graphics card market, and Matrox has publicly stated that they are not pursuing the 3D gaming scene currently.

So what led up to the situation we are in today? And what will this New Year hold for the market? Let's find out as we conclude our look back at the year 2000 with an analysis of the graphics industry.



The performance lead goes to: ATI?

Here's a picture we'd never thought we would see. At the start of 2000, ATI had one of the fastest graphics accelerators available, the ATI Rage Fury MAXX. Just a year earlier we were wondering how ATI would remain competitive with their Rage 128 part offering sub-par performance around three months after we expected to see it.

About six months prior to the official launch of the Rage Fury MAXX, we had the opportunity to talk to ATI and discuss the potential of their Rage 128 Pro chip as a high-end gaming solution. In an interesting response, ATI mentioned that they would be able to produce a Rage 128 Pro that offered performance similar to NVIDIA's previous king of the hill, the TNT2 Ultra. Being very enthusiastic about having another TNT2 Ultra class performer in the market, we gave ATI our encouragement to pursue such a project, and while the card never did surface it did allow us to realize something very important about ATI: they wanted to compete in the gaming market, even more so, they wanted to compete with NVIDIA.

At the time we didn't think it was too likely, however out of all of the graphics chip manufacturers that we had talked to, that was the first time that we actually saw a competitor to NVIDIA realize exactly what it would take to compete with them.

Late in 1999, ATI dropped a similar bombshell on the public, the announcement of the Rage Fury MAXX. The Rage Fury MAXX was more of a pilot run for ATI, designed to test the waters of the 3D gaming market and at the same time, open the doors for what was destined to come.

The Rage Fury MAXX was a brute force solution to competing with the fastest card on the market at the time, the GeForce 256 from NVIDIA. Since ATI's flagship, the Rage 128 Pro, was only capable of offering TNT2 class performance at best, they had no hope of competing with NVIDIA's current generation product based on a single chip design.

The solution they came up with was to combine two of their Rage 128 Pro chips onto a single board design, they had good enough yields on the chips to keep the cost of adding two to a single board reasonable. With that, ATI released the Rage Fury MAXX, their first dual chip solution, and their first product that was able to directly compete with a current generation NVIDIA solution and sometimes even outperform it.

Going into 2000, ATI was all of the sudden back in the game, but they still had a very long way to go before they could be taken as seriously as they needed to be in order to be a leader in the market. At the end of the day, the Rage Fury MAXX was still based on old technology, while both 3dfx and NVIDIA were boasting new features such as Full Scene Anti-Aliasing and Hardware T&L.



Losing the battle without any weapons

By the end of the first quarter of 2000, the GeForce DDR became the semi-affordable card of choice for all gamers. Card manufacturers stopped producing GeForce 256 based cards that used regular SDRAM since the price of DDR SDRAM was not much more yet the performance gains were astronomical.

This posed a major problem for 3dfx. NVIDIA's flagship product was getting faster, and 3dfx still had no new products other than the aging Voodoo3, which was almost a year old at that point. The Voodoo4 and Voodoo5 cards that were announced at Fall Comdex 1999 had still yet to appear, and there was no word as to when they would appear. At the same time, 3dfx had to build up public desire for these products, otherwise the launch would definitely be a flop.

The major feature that 3dfx was touting that would be in this next-generation of Voodoo cards was the notorious T-Buffer, with FSAA being a major part of the feature set offered by this T-Buffer. Unfortunately, 3dfx encountered two major problems with convincing the population that FSAA was worth the wait.

First of all, it is very difficult to show off a feature that really follows the old axiom, seeing is believing. The still screenshots that were presented online of what dramatic increase FSAA can provide just didn't do it for most users and when combined with this next problem, made FSAA more of a laughable feature than something that was desired.

The second and most difficult problem 3dfx encountered was that enabling FSAA, using the current methods, resulted in a 50 - 75% drop in peak fill rate, which means performance that is equal to or less than the Voodoo3 3000 in many situations.

So while NVIDIA encountered similar resistance to their boasting that Hardware T&L was the way of the future, the users didn't complain since having Hardware T&L didn't degrade performance any - it was simply not used enough by developers.

In the end, the market was waiting for 3dfx to produce a competitor to NVIDIA's GeForce 256 while NVIDIA was getting ready to release its successor, the GeForce2 GTS.

April Showers

Ever since the release of the TNT2, NVIDIA had been operating on a 6-month product cycle. This was the aggressive NVIDIA that had risen from a little known producer of the NV1 to a force to be reckoned with.

NVIDIA's goal was to release a new technology every Fall, and then "refresh" it in the following Spring. The refresh usually meant a smaller manufacturing process, higher clock speeds, and as many additional features that could be included without dramatically changing the core. For example, the TNT2 was the Spring refresh of the TNT that was released the previous Fall. And the GeForce 256 which was released in the Fall of 1999, was due to be "refreshed" in the Spring of 2000, and it was.

At the end of April 2000, NVIDIA decided to launch their GeForce2 GTS product, the GTS being an acronym for Giga Texel Shader. The GTS was actually ready to go much earlier, but it didn't make sense to release it any sooner as the GeForce DDR was still selling very strong since 3dfx had no competing offering.

Five months after their incredible announcement at Fall Comdex 1999, 3dfx also brought their first next-generation card to the public view, the same week as NVIDIA's GeForce2 GTS launch. Unfortunately for 3dfx, while the GeForce2 GTS would be available just a couple of weeks later, the only card to make it out of 3dfx's doors since the Voodoo3 3500TV would be the Voodoo5 5500 and it would not be available until June. Even worse for 3dfx was the fact that the 5500 was noticeably slower than the GeForce2 GTS.

In a surprise move by ATI, the launch-week of the GTS and the 5500 was concluded with the presentation of ATI's next-generation core at WinHEC, the Radeon. While we were very skeptical of ATI's ability to deliver the Radeon core on time, we couldn't help but think that the Radeon's core was quite impressive on paper at least.

Three months later, ATI delivered the Radeon to the public. Well ahead of when we expected to see it and very unexpected by NVIDIA, ATI was back in the game.



The value of a dollar

If April was the month of high priced video card announcements, June was the month where the more reasonably priced products were announced. Towards the end of June AMD announced what would turn out to be our pick and your pick for best Value of 2000, the Duron processor. At the same time, almost in harmony with AMD's release of a high performing, value priced solution, NVIDIA brought the GeForce2 MX to the public's eye.

The MX offered greater than GeForce SDR performance at a cost that would be close to $120. Combine that with the ability to purchase a $100 CPU that offered 90% of the performance of an Athlon, and you had a killer system for a very small investment.

The GeForce2 MX was also NVIDIA's somewhat unsuccessful attempt to get into the corporate market. By offering TwinView (dual monitor outputs) as a GeForce2 MX feature, NVIDIA was attempting to steal some of the attention that Matrox's DualHead gained in the corporate market. The problem NVIDIA ran into was that they were too closely linked to the gaming market, which has its pros and cons.

On the positive side, they were making quite a bit of money off of the gaming community. On the negative side of things, however, that's not a good selling point when you're pitching a product to an IT manager that is looking for a productive video solution for the company's employees.

NVIDIA's Fall Refresh

As Fall 2000 approached we feared the worst for 3dfx and ATI. In spite of their most recent efforts, there would be no way they could compete with NVIDIA's next-generation core, the NV20. NVIDIA had yet to miss a product cycle, and as they had publicly stated, every Fall they would have a new architecture and the following Spring would mark the release of a refreshed version of that architecture.

As August rolled around, NVIDIA made their second release for 2000 as expected, however it wasn't the NV20. Instead, NVIDIA surprised us all with a second refresh of the GeForce core that was released 12 months earlier - the GeForce2 Ultra. The Ultra was a $500 version of the GeForce2 GTS that was nothing more than a hand picked GTS core that could hit higher clock speeds combined with some very expensive and very fast DDR SDRAM.

The performance of the GeForce2 Ultra was incredible, however the market didn't need another $500 card, the market was expecting the NV20. While some expected the NV20 to be released before the year's end, as we indicated in our GeForce2 Ultra review, the NV20 won't hit the streets until the Spring of 2001, or 6 months from the release of the Ultra. Given that the Ultra was released in August of 2000, you can expect the NV20 to debut very soon, in the next month or two.

The NV20's delay would have been the perfect opportunity for 3dfx, ATI or Matrox to step forth with a higher performing solution that would essentially kick NVIDIA while they were down. At the time, we alluded to the possibility that one of the aforementioned manufacturers was capable of doing just that. Many incorrectly interpreted that to be Matrox, when we were actually hinting at ATI. Indeed ATI had plans to have a fairly large show at Fall Comdex 2000, potentially because of the announcement of either a Radeon MAXX or the Radeon's successor.

To our disappointment neither the Radeon MAXX nor the Radeon's successor made their debut at Comdex, a lost opportunity for ATI to steal some of NVIDIA's thunder, however their reasons for not doing so were quite valid. The Radeon MAXX never surfaced simply because of cost, it would've ended up being a $500 - $600 board that wouldn't have garnered much other than the title of fastest card for a couple of months. The next-generation Radeon was most likely held back for reasons similar to the NV20's delay, the lack of DirectX 8 among other factors.

The one redeeming quality of the GeForce2 Ultra's release was the release of NVIDIA's Detonator3 drivers, which gave a free performance boost across the board to all GeForce based cards, especially those that were severely memory bandwidth limited - most significantly, the GeForce SDR and GeForce2 MX.

Also in the August timeframe were the introductions of the All-in-Wonder Radeon and the quiet introduction of the GeForce2 Pro, a slower, OEM version of the GeForce2 Ultra that board manufacturers eventually made into a pretty attractive solution for the retail market as well.



After 10 months, the Voodoo4 is released

As if the year 2000 wasn't already disappointing enough for 3dfx, a full 10 months after it was first announced, the Voodoo4 4500 finally made its way onto store shelves in September without a chance of outperforming NVIDIA's GeForce2 MX.

Just one month later 3dfx would make what would become their last product release, so much for going out with a bang as 3dfx released the VoodooTV. The VoodooTV was the only "good" to come out of 3dfx's acquisition of STB in 1999. If you haven't guessed already, this officially made 3dfx's decision to acquire STB worthless which they confirmed by later announcing that they would allow third party board manufacturers to make Voodoo cards.

Giving the Voodoo4 4500 an even harder time was ATI's release of the Radeon SDR, which like the GeForce2 MX, was without a doubt faster than the Voodoo4.

A king is lost

After announcing that the Voodoo5 6000 was essentially a dead project that would be licensed out to Quantum3D and never see the retail market, 3dfx dropped the bomb on us in December of 2000. 3dfx, one of the fathers of the 3D graphics revolution, was to sell its core assets to its biggest competitor, NVIDIA. And, pending shareholder approval, 3dfx would shut its doors completely - no technical support, no driver updates, just completely shut down.

I once told one of AnandTech's Senior Editors that I would hate to be in 3dfx's shoes. NVIDIA is a very aggressive competitor and has definitely gone to some very great lengths to succeed. Their success does not come without hard work though, as no other manufacturer in the graphics industry has been able to meet release after release on such a tight product cycle. However in the end, it was NVIDIA's ability to meet such incredible product release deadlines that contributed to the death of 3dfx.

This, combined with S3's pulling out of the high-performance graphics market and Matrox's focus on a purely corporate agenda for now, leaves NVIDIA virtually alone in this market. Their next prey? ATI.

But unlike 3dfx, ATI may have what it takes to put up quite a fight. You can expect this New Year to be a very intense battle between ATI and NVIDIA, and as much as it pains us to say this, you shouldn't expect to see any real competition from any of the remaining manufacturers mentioned in this article.

Right now, it's a race to see which company will slip up first. As we have seen in the past, just missing a single product cycle can lead to a downward spiral that is borderline impossible to recover from.

The heat is on.

Log in

Don't have an account? Sign up now