Slo-Tech | nova perspektiva računalništva

Interview with Richard Huddy

Avtor: Marko Dolenc
Datum: 23.06.2003

Slo-Tech: Could you first tell our readers a little about yourself and what you do at ATI?

Richard Huddy: I work in Europe as part of the group which supports games developers - I watch the development of our new technologies and evangelise them to various technical staff who work in games development, and I work with a number of external technical resources who are also involved in developing graphics theory and practice.

I also represent ATI at a significant number of technical and press events around the world. For example I was flattered to be involved in the Radeon 9800 Pro launch at GDC this year, and by now I've visited most European countries in some form of PR capacity but this is my first work for Slovenia.

Slo-Tech: ATI definitely came on top in the last year especially with introduction of Radeon 9700 and Radeon 9500 series. How long do you think current situation with two strong players can last?

Richard Huddy: It surely can't last long as this situation is fundamentally unstable. Sooner or later one company will go wrong twice in a row - and that will cost them a huge chunk of market share. NVIDIA slipped badly with the NV30 series - and they've publicly admitted as much. Now they urgently need to regain the high ground - and NV35 is supposed to achieve that. As we've seen from the recent furore over driver cheats it seems likely that they don't plan to let their own inferior hardware come between them and first place. In such a competitive atmosphere it's hard to see the situation lasting for a long time.

On the other hand, a year is a long time in the graphics business. And in two year's time the market will have moved on a long way. It's also clear that the financial barriers to entry in the graphics market are very substantial. Creating a new company to compete in this market was feasible 5 years ago, but now it looks almost impossible.

Richard Huddy
Richard Huddy

Slo-Tech: ATI has been growing in popularity in Slovenia mainly because of Radeon 9500 as it provides all the latest technology at affordable prices. However as prices for high-end cards seem to be on the rising will ATI continue to introduce new cards with all the latest technology into mid-end and low-end?

Richard Huddy: The mid-range cards from ATI (the 9600 and 9600 Pro) are proving very popular. So popular in fact that we just can't make them fast enough to satisfy all the demand which we have. The price of an average PC has fallen dramatically in the last few years, but in the mean time the high end graphics cards have managed to maintain their prices at around $400 or even $500. Those two trends can't last forever - sooner or later even the top end graphics cards will have to fall in price.

All of this means that ATI is highly focussed on mid-range and low end graphics. The top-end is great for bragging rights, but we sell roughly 30 times as many boards into the low-end of the market. That's our main market - and there you can expect a strong feature set at affordable prices.

Slo-Tech: What do you think about attempts from companies like S3 and Trident trying to break into DirectX 9.0 club? Even bigger companies like 3D Labs and Matrox don't consider DirectX 9.0 parts as their top priority and NVIDIA obviously had a lot of trouble bringing their GeForce FX on market.

Richard Huddy: The companies which are trying to produce entry level graphics with high-end features seem unlikely ever to have much impact in the add-in board arena. Those companies need to show the existence of their technology to convince people that they're capable of producing fully functional integrated chipsets, but their boards tend to be seen only a low-end technology provers. No serious gamer would choose to go for a Trident graphics board...

The cost of producing a functional piece of high end DirectX 9 hardware from scratch is now reckoned at about 300 to 400 million dollars. None of the small players can afford to get involved in that kind of investment. It's not that they don't have the money (though many of them do not). The problem is that it's such a gamble. We've seen some extraordinary attrition in the graphics market over recent years - and there's no reason to believe that the market will become more forgiving in the near future.

Slo-Tech: What do you think went wrong at NVIDIA with its much hyped "NV30"? Was it supposed to look like it looks today back when you worked at NVIDIA?

Richard Huddy: Obviously I'm still bound by my non-disclosure agreements which I agreed to when I joined NVIDIA - and that means I can't say very much in response to this question. The most obvious thing that I can add is that NV30 was very, very late. And even from outside NVIDIA it's been clear that when it arrived it was badly broken. It's taken them a long time to address these problems - and it'll be interesting to see if they've solved the biggest issues. Their very slow full-precision performance is still a major weakness of the NV3x design, and nothing I've seen so far says that NV35 will solve that.

Slo-Tech: Do you think NVIDIA driver team is actually focusing on the right aspect of the driver? It seems like finding a nice place to add a clipping plane in 3D Mark ranks higher then making some DirectX 9 features like floating point textures to actually work.

Richard Huddy: NVIDIA's driver team does what their management tells them to do. Someone right at the top of NVIDIA has clearly decided that making benchmarks run faster by ignoring some of the requested drawing operations is more important than running games quickly. And that's a shame for two reasons. Firstly the enhanced benchmark score has no merit. I read it as a blatant attempt to persuade buyers that NVIDIA hardware is better than it actually is. Secondly there are features such as those you mention which are still missing and general performance which would give significant benefits to users. It's sad to see a company preferring to spend it's energies on these wacky pieces of code, which they describe as "Application Specific Optimizations", when they should be producing better drivers which could enhance the value of all their existing hardware in the market.

Slo-Tech: Will ATI continue to push its way into chipset market and seriously challenge NForce 2?

Richard Huddy: Yes, if you've seen the recent press releases from ATI then you'll know that our chipset business in both desktop and laptop areas remains strong. The fact that we have a Pentium 4 bus licence allows us access to a very substantial market. The nForce seems to be justifiably popular in the AMD enthusiast area, but ATI has some hardware coming soon for the Pentium 4 which will really impress!

Slo-Tech: Can you perhaps tell us how many people are migrating between ATI and NVIDIA?

Richard Huddy: We're seeing a significant move, particularly in the enthusiast segment, towards ATI. Of course the fact that we had the only available DX9 hardware for almost nine months is a large part of the explanation. But it's also impressive how well the 9700 design (internally code named "R300") has stood up to the competitive pressure in the last few months. We've updated the clock speeds, and made only a few small changes to the logic and we still have a really excellent competitive part. If you compare performance on existing games (which are typically using DX7 or DX8 class hardware well) then you see NVIDIA's GeForce FX comes in about level with the 9700 and 9800 - so which is better remains unclear from this test... But when you compare both companies' top end hardware on any DX9 benchmarks and games then the Radeon 9800 is in a class of its own. That's a staggering testimony to the quality of the R300 architecture.

Slo-Tech: Do you have any information about how many people are switching their jobs from NVIDIA to ATI or vice versa?

Richard Huddy: I'm afraid I don't have any information on this. As some of your readers will know I used to work for NVIDIA and now I work for ATI. The best thing about ATI now is that it is now the company that NVIDIA should and could have been. ATI is the technology leader and it's shown great integrity and determination. I'm one very happy person who is happy precisely because I changed sides - and I'm sure there must be more.

Slo-Tech: When will ATI provide open source (Linux) developers with information about yours hardware so that they will be able to write drivers instead of reverse engineer them?

Richard Huddy: ATI gives Linux drivers quite a high priority - but there's just way too much intellectual property exposed in the low level chip interfaces so we don't put that into the public domain. I'm amazed that people can really reverse engineer drivers from our binaries - but I guess that shows just how keen the Linux community is to get the best out of their machines. Sorry I can't offer more on this!

Slo-Tech: Do you think VR headsets that were viewed as future of 3D in the nineties will make a comeback now that graphics cards are fast enough?

Richard Huddy: I don't plan to spend much time with my head in a bucket, and I'm sorry to say that these headsets always made me feel that way - pretty ludicrous. I guess if manufacturers are able to build a small enough, light enough alternative then there may be a future in it. But the main stream is still totally dominated by people with just one monitor roughly 40 to 50 centimetres from their face. And we have a lot to do before we've satisfied the demands that those regular gamers tend to make. VR headsets are destined to stay out of the mainstream for quite a while...

Slo-Tech: How do you think will DirectX and OpenGL continue to evolve in next couple of years? What are they still missing and how fast will they improve?

Richard Huddy: Right now both API's have integrated the shader programming model and both are moving to high level languages. The shaders will continue to become more powerful over the next few years - because it's obvious that we need more power. But by the time 5 years is up we should see several new programming technologies in place. Graphics programmers should be freely manipulating subdivision surfaces, generalised displacement maps and global illumination. It should also be possible to perform Radiosity lighting entirely in the graphics chip at that time too. All at high frame rates, and with uncompromising quality.

What is perhaps most impressive is how fast Microsoft have innovated with DirectX. They're produced a new version about once per year for the last six years, and they've now achieved the impressive step of releasing DirectX 9 which has support for all of next year's hardware already built in! That's quite something given that it contains pretty much everything that's expected in OpenGL 2.0. Microsoft may have their failings - and they certainly have their detractors, but I'm impressed with what they've done this time around.

Slo-Tech: Did TruForm technology fulfil ATI's expectations and what could still be done in this area?

Richard Huddy: Truform is an interesting technology because it added quality to game rendering without requiring a great deal of intervention from either the artist or the programmer. But realistically it's a stop-gap. What artists really want to manipulate is subdivision surfaces. These allow fine grained control of curved surfaces and they remain compact in a way that high detail polygon models do not. You can think of a subdivision surface as like the result of getting an artist to tessellate a model with infinite detail. And the nice thing about the SD surface is that it doesn't require as much time to generate as the artists would need. Instead the artist defines the interesting bulges and twists in a mathematically smooth surface and then the hardware creates the polygons from that mathematical description. That's the way of the future...

Slo-Tech: Could you tell us what are perhaps 3 most common mistakes that game developers are making and how they affect performance or visual quality?

Richard Huddy: First of all, nine out of ten games under-use the graphics card. That's amazing, and it's been true for the last three or four years. With the future generations of graphics hardware that I've been talking about in response to the last question the situation will be corrected. Then it suddenly becomes easier to use the power of the hardware because the hardware, the artist and the programmer all 'think' in the same terms.

Secondly not enough games take advantage of the power of the shaders which are available now. We're still having to work hard to push games developers to move from writing pixel shader assembler to HLSL (Microsoft's High Level Shading Language which is new to DirectX 9) or GL2 (which is OpenGL's high level shading language).

And thirdly almost every games developer I know drinks too much caffeine and works too many hours in the ordinary working day. We need to work smarter not harder. Using the high level shading languages are a major part of this, but generally the life of a games developer is pretty tough.

Slo-Tech: What could (should) game developers push more on current generation hardware? More triangles, bigger textures,...?

Richard Huddy: Bigger textures only help if the textured objects can get _really_ close to the player. And more polygons usually only improve the silhouette of an object. Most game developers get these things about right these days... The thing they should push harder is their artists. If the programmers showed the artists how much could be done with modern shader hardware then the artists could really start using their hidden creativity. I'm also disappointed that quite a few games still don't allow the player to enable anti-aliasing. Anti-aliasing is such a big quality win that I recommend that if a game is running on DirectX 9 hardware then it should enable anti-aliasing by default.

Slo-Tech: Games are obviously quite far behind graphics cards technology. How do you think will Doom 3 affect this considering that it looks like there will be graphics cards capable of running it in 100 fps range when the game ships?

Richard Huddy: Doom 3 won't require anything more than DX8 class hardware to run the game, and while there's no doubt that it'll be a tremendous game a great success there are several other games that I'm also looking forward to this year. Top of my list are Half Life 2 from Valve, and Far Cry from Crytek. Both look amazingly good and they take the outdoor games engines to a whole new level.

And Doom3 will still not run at 100's of frames a second if you crank up the anti-aliasing settings, enable anisotropic texture filtering and run at the highest resolutions. We still have plenty more raw power to deliver to make that happen.

Slo-Tech: Do you think ray tracing will ever become viable solution for games? What do you think about OpenRT project (

Richard Huddy: Ray tracing is still a long way off from being a viable way to write games - but as graphics chips extend their programmability it's clear that at some stage you should be able to do all the calculations required for ray tracing inside consumer graphics hardware. But this isn't a principal aim of the games graphics business at present. Ray tracing is rather different from the new buzzword techniques of "spherical harmonics" and "global illumination".

But any research into ray tracing has to be a good thing. There are people these days in the games business who are using amazing tricks of lateral thinking to make graphics cards solve complex physics simulations - so I wouldn't want to rule anything out.

Slo-Tech: Any extra message for all the ATI fans here in Slovenia?

Richard Huddy: For the existing ATI fans I'd say thanks for your support - you should know that our commitment to you is a serious long-term commitment to providing the best hardware and drivers that can be built.

For the future ATI fans I'd ask that you look closely at what ATI has achieved in recent years. We have the best hardware available, we delivered DirectX 9 hardware months before Microsoft completed DirectX 9, and we have a driver quality program which is consistently delivering high quality WHQL certified drivers on a roughly monthly basis. ATI is what every graphics company would like to be. We're intensely competitive, delivering great products to tough schedules, and we do this with honesty and integrity. In a business where arrogance, cheating at benchmarks and relying on product hype has become the norm ATI is setting great new standards for giving you what you pay for - high quality, high performance graphics available to everyone - at an amazing price. I'm proud to work for ATI. I hope one day you'll be proud to own one of our graphics cards.


Richard "7 of 9" Huddy.

Slo-Tech: Thank you Richard for answering our questions. In the name of Slo-Tech crew and our readers I'd also like to wish you and ATI all the best in the future.