By Charlie Demerjian: Monday 23 August 2004, 13:05,
I HAVE A KNACK for spotting trends, and the latest one I've spotted is in the graphics industry. For the last 2-3 years, basically since the first R300 benchmarks came out, it was clear that ATI was in the driver's seat. NVidia was simply not delivering at the high end, or just about anywhere else. The next refresh was more of the samr. It looked like ATI was unassailable.
Then a funny thing happened, I was walking around the Intel Developer Forum this spring, talking to people who 'knew' things, and asked them about the new crop of graphics chips. At the time, they were weeks away from coming out, the then new and yet only named by us X800 and 6800. I was fully expecting them to say that ATI was going to extend its lead as was the word on the street. Strange thing happened, they did not say that.
To paraphrase Yoda, intrigued I was. Really intrigued. Especially when ATI was first out of the gate, and the numbers showed a little advantage for ATI. I was starting to think I was wrong, and several people were more than kind enough to point this out.
When the chips launched, and real people got their hands on real parts, the story changed a little. Nvidia's response was to tout the features that it had, and ATI was lacking. People pointed and laughed, editorials were written, and the collective tech world failed to grasp what was going on. Who cares about PS3.0 when there are no games for it? Who cares about PS3.0 when MS does not have a version of DX that supports it? Geometry instancing? What is that?
ATI was rather smug, and pointed to benchmarks like Quake3. Yay, 450FPS if you spend $400 on ATI, and you only get 425FPS if you spend $400 on NVidia. Most of the people didn't really grasp the concept that the best monitors out there only hit 120Hz for a refresh rate, and both cards would more than double that with all the eye candy turned on. The point of another 25FPS in an old game with all the eye candy turned on is.......?
I was almost ready to believe, but then cracks started to show, and they validated my beliefs, ATI missed the boat, and missed it badly. Don't believe me, how about looking at it from the ATI perspective? ATI knew it was behind, and knew there was nothing it could do for a year. What do you do in this case, spin, dodge and dance.
You'll have to trust me when I say that it tried, and it tried hard, it danced up the proverbial storm. The funniest jig was when they told me with a straight face that there were situations that PS3.0 made programs slower than PS2.0, which ATI proudly featured. Yep, Ferarris get crappy gas mileage also, but I still would not trade one in for a 1990 vintage Yugo, turbo or not.
Why was ATI so afraid? It knew what was coming, and it was all in Nvidia's favor. On the face of it, ATI has HalfLife2 and NVidia has Doom3. Benchmarks show that Nvidia has a commanding lead in Doom3, bordering on the abusive. ATI has a noticeable lead in HL2, but from what I have seen, not as commanding as the lead Nvidia has in Doom.
Why is this so worrisome? Well, for two reasons, licensing and the future. Licensing is the worst one, Doom3 is an absolutely spectacular engine, head and shoulders above anything else out there. It is scalable, beautiful, has a marquee name, and is saddled with amateurish gameplay. Three out of four ain't bad, and luckily for Nvidia, gameplay is not its problem. In fact, only the first three are relevant to video.
So, one marquee game versus another, it would appear to be a tie. The thing is that the Doom3 engine will be licensed by just about everyone under the sun. The HL2 engine may be, or may not be, but people won't be clamouring for a Newell engine as rabidly as a Carmack one. While the proverbial 'people' may be wrong, they are the ones spending the money.
This leads us to the future. HL2, for all its gameplay is a last generation engine. It looks good, very good in fact, but it is a year old, and things like Far Cry and Doom have passed it by. This isn't a criticism, it is just time and evolution, I have been playing CounterStrike:Source for a few days now, and it is very very nice. Newer engines look nicer though.
In the older engines, HL2, Q3, and others, ATI has a slim advantage. On the newer ones, Nvidia rocks. Which one do you think will be more prevalent in the next few months? Which one would you license if you had to lay the money down? Which one do you think will power the next generation in greater number? Yup, ATI has a problem, and it will only grow worse the further out you look. If you want real comedy, look at the requirements for Longhorn.
But it gets worse. ATI has no PS3.0 part, and never will. The problem isn't that big if you look at it from the perspective of the highest end on the market, the pimped out Ultra-Platinum-Stupidlyexpensive Edition, those cards make up a miniscule amount of the market. They cast a great shadow, but monetarily, I would be surprised if they made money directly, but they sell a lot of mid and low end cards.
Here too, ATI is sucking wind, and has been for quite a while. ATI has a crappy line up other than the high end. This is nothing new, it started in the last generation. The capabilities of the Nvidia low end lineup hits most of the checkboxes that the high end 5950 cards do. ATI just rebadges the previous generation, and uses its numbering scheme to make it sound like they have something that it doesn't. It is more than enough to fool some of the consumers, but then again, there are a lot of machines sold with Celerons in them.
The point is not consumers, but OEMs. They are more than smart enough to actually look at the capabilities before they buy the chips by the hundreds of thousands. Worse yet, many of them sell to Joe Six-Pack by means of spec sheets more than performance. Nvidia could hit those checkboxes with the last generation, but ATI still can't. The current generation gets worse, the high end can't offer the features at all, and the mid and low end get worse from there.
As the games start coming out that use PS3.0, and people start looking for those checkboxes, one company will be there. OEMs know this, and buy accordingly. That is where the money is, and it is where ATI isn't and won't be. The last time someone missed like this badly, it took Nvidia about 2 years to recover, and 3DFx never did. I'm not intoning anything here, simply stating it, just watch and see.
Time will only make the situation worse and things are falling into NVidia's lap. If you were at E3, and looked at all the must have games, you would see that they actually do use the features that Nvidia offers. The biggest display in the largest hall was run by EA, and the game that wowed me the most was the Lord of the Rings game. Think massive army combat, effects, and loud music. It was a showcase for geometry instancing and PS3.0 effect. The new crop of games coming out in a few weeks will tell the story loud and clear.
Add in things that do not fall under the domain of the GPU itself, and it only gets worse. Nvidia is really good at making chipsets and platforms, and the next generation of NForce chipsets have some really nice looking features. ATI is still trying to figure out what the term South Bridge means. I am not trying to kick ATIs chipset efforts here, that would be too easy, no sport there. The crushing superiority of the Nforce platform allows Nvidia to do something that ATI can't, tune the chipset for the graphics card to get that extra bit of mileage.
While this alone would probably be enough to give Nvidia the crown, it has a secret weapon. Not a little pistol hidden in their boot, more like a tactical nuclear weapon called SLI. If you go to ATI headquarters, you can still see the confused look on their faces when they first heard the whistling of the incoming warhead.
SLI alone will be enough to make Nvidia own every benchmark under the sun by simply abusive margins. The early benchmarks indicate that it really is everything Nvidia said it would be. ATI's response? Well, nothing at first, I think it realises that HL2 benchmarks won't cut it this time.
It gets worse though. There has been no response at all since tha SLI announcement, and this is a classic sign of one side getting caught with its pants down. What we know of the R500 and NV50 is that they will be PS4.0 and DX(insert marketing term here) parts. ATI will never have a PS3.0 part. It probably won't have an SLI part in that generation either. How do I know?
Quite simple actually. When someone notices that they have their pants yanked down like ATI just did, they tend to deflect things to any place but where they are and to any time but now. Standard PR operating procedure here is that ATI would deflect the question, and talk about how even though it is not a big deal, their next gen parts would have it. Piffle, tee-hee, there is no man behind the curtain.
When Nvidia announced its parts, you heard ATI crowing about how the R500 would have SLI also, right? Me neither. None of my sources heard it, and no OEMs are talking about it. Said checkbox will probably be blank until R600 time. That is a long long time to wait while you lose every single benchmark on the planet.
That may make you lose some high end sales, but as I mentioned earlier, those sales really don't amount to all that much. The OEM sales matter a lot, and everything looks to be in Nvidia's favour here as well. Of late, a lot of companies have defected from the Nvidia camp to the ATI camp. I think the OEMs will be quite receptive to the marketing pitch of 'You sell ATI, no SLI'.
If you want to make OEMs more receptive to the enemy's tender advances, there is a nice way to do this, screw your partners. While ATI trumpets the number of PCI Express parts it is selling, the OEMs tell a different story, namely that ATI is selling all that it can make, just not to them. When you are competing with your partners as ATI does, and you have a shortage, as ATI does, you don't screw your friends. ATI did. Really, ask any ATI graphics card manufacturer how good their supply of PCI Express parts are, I did. The responses are uniformly not polite.
So, what do you do if you are ATI, and you are looking at a lineup of parts that missed the boat? You deflect things. The last big FUD campaign was the native vs non-native PCI Express problem. ATI was giving out T-shirts proclaiming how Nvidia didn't have a 'native' PCI Express solution. They were using an ('eww' is intoned here, preferably with a sour look on your face) bridge chip.
So, what does this mean? Well, the benchmarks show that it means nothing at all. Common sense says that when you are not using all the bandwidth that AGP 8x offers doubling it won't add much performance. Adding a little latency obviously didn't hurt much either, all the benchmarks I have seen show no or minuscule differences between the AGP version and the PCI Express flavour of Nvidia products.
The problem for ATI is that it also shows the same thing for the X800 line. So much for the vast superiority of the 'native' way. It almost seems like Nvidia engineers were focused on doing things that mattered rather than marketing slogan engineering. Hard numbers can be a bitch when you are trying to spin.
Looking out, I see ATI with a minuscule advantage that it is clinging to, desperately hoping no one notices that the train left the station in early August with the release of Doom3. The future belongs to Nvidia right now, and the only hope ATI has is in the R500, but that won't be here for a long time. If it manages to catch up card for card, Nvidia still 'only' has a 2:1 advantage. I know where my money will be going, and an X800 seems like money badly spent if you don't plan on buying an new card every three months. µ
