I held off writing another blog post for a couple months just waiting for Fermi. I didn't have much else to write about, and as far as tech-related stuff goes, Fermi was the biggest thing on my mind. I've already said pretty much everything else I'd ever want to say about it, and there wouldn't be anything new to comment on until it finally released. (I did think about writing a rambling about Bioshock 2 but I didn't have anything to say about that that hadn't already been said elsewhere.) Since its release, it was just a matter of setting aside the time to do it.
The last six months have been agonizing. I remember the day the 5870 came out. I had WoW running in the background and a webpage sitting there in front of me with a NewEgg listing of all the 5870s available, all of them $380. I had to will myself not to pull the trigger on one, because NVIDIA might have something much better right around the corner, and it might be stupid not to at least wait and see. Usually in the tech world that's always the best policy, but this is one of the few times I'm kicking myself for not indulging in some impulse buying.
After waiting so long, to say I'm disappointed would be an understatement. I was expecting something with 512 SPs running at around 1800MHz. What I got instead was something crippled. Of course the price ended up exactly where I expected, $500, but at that point it isn't worth it. On top of that, 5870s have gone up in price, with no incentive to drop.
During the long wait I had to withstand a lot of ugly rumors surrounding the new GPU. It was only bearable since the bulk of said rumors spawned from Charlie Degenerate, but a few came from elsewhere making them a little harder to ignore. As it turns out, most of the ugly rumors were true (although Charlie never got any of the specifics right), and our worst fears were confirmed. Since very early on this chip has been compared to the FX series due to its problems, and after finally receiving the benefit of full disclosure, those comparisons haven't stopped. So instead of simply restating facts about the chip, let's do something a little more interesting, and draw comparisons with the past history that has supposedly been repeated, and see how much of it really holds up.
First let's get the obvious similarities out of the way. Like the FX 5800 Ultra, the GTX 480 launched six months late. Like the FX launch, the GTX 480 is hotter and louder than the competition. Like before, the new card only marginally outperforms the competition (say ~15%), if it outperforms it at all. Since power consumption wasn't an area of focus for hardware websites at the time of the FX series, we can't make any comparisons there.
Areas where the GTX 480 launch is worse than the FX 5800 Ultra launch: The FX card wasn't priced higher than its competition. The FX card didn't launch with the competition enjoying a full top-to-bottom complement of cards covering the next-generation API. The FX card enjoyed some time in the sun before being completely trounced by an even higher performing version of its competition (the 9800 Pro -- the GTX 480 must already contend with the 5970, whose performance advantage is much higher than ATI enjoyed before). By rough estimates, the FX card never ran as hot as the GTX 480 does.
Lastly, we'll look at how things are better this time. The GTX 480 doesn't have any glaring architectural weaknesses compared to the competition (128-bit bus says hello). The GTX 480 doesn't have issues supporting the next-gen API, and performance seems to be fine in that regard. The image quality doesn't take any noticeable hits compared the competition either, especially regarding AA/AF. The GTX 480 does not get as loud as the 5800 Ultra did. Availability for the GTX 480 remains to be seen, but we can probably assume it's going to be better than the 5800 Ultra.
Overall I'd say the GTX 480 isn't quite as bad as the 5800 Ultra was, but there is definitely cause to reflect back on it. One interesting similarity between the two is the questionable architectural choices touted by NVIDIA in regards to next-generation gaming. As I said before, it's already apparent that the GTX 480 won't have any problems with DX11 games like the FX series had with DX9, but the design is forward-looking in a way that may not mesh up with the direction developers choose to go.
The FX series was designed for what NVIDIA called "Cinematic gaming", in which games would take a dramatically photo-realistic turn primarily by way of high floating-point color, or FP32 specifically. NVIDIA felt that just having support for this was a marked advantage over their competition, and banked so heavily on it that they didn't even support the less demanding FP24 standard that DX9 called for. Unfortunately for NVIDIA, coming out six months late left the developers building on the 9700 Pro instead as their hardware standard, and Microsoft sided that way as well, not really exposing FP32 in the initial DX9 version but completely axing FP16 as being adequate enough. The problem was FP16 was NVIDIA's only fallback for the FX series, which NVIDIA had assumed would get DX9's support, and they didn't equip the architecture with enough resources to make FP32 performance acceptable.
The GTX 480 also has a major feature NVIDIA is counting on to (eventually) propel it ahead of the competition, by way of the parallel geometry processing architecture. Not so much a conflict of standards as just a huge strain on chip production, the very complex geometry design banks on the idea that developers will make huge investments in tessellation use, so much so that without such a design, performance will be greatly hampered. But just as before, because of their belated launch, developers have had to look to the competition as their hardware standard for implementing new DX features, leaving NVIDIA with a party nobody's showing up to. At least this time around, it's only a matter of too much performance rather than too little, which is rarely a bad thing in this field, but the payoff may be too long in coming to be of much relevance, and considering the delays it caused, it might have actually been more of a hindrance.
Architectural complexity aside, NVIDIA can't get all the blame for this botched launch. Too be sure, it is a failure of a launch, no two ways about it, and most of that rests on the chosen price point, but TSMC had more than a helping hand influencing the ultimate outcome here. I'd wager they shoulder an equal share of the failure, and in the grand scheme of things, potentially much more of a failing towards their overall business than anyone else. I'll write more about that in a later rambling.
The bottom line is, I have to buy a new graphics card. The cards are now on the table (ha), and reluctantly I must change sides for the first time in years, and go with the 5870 as so many other people ended up doing. Indeed, what was ironically supposed to be an event that provided a lot of publicity for NVIDIA may end up just being a huge sales boost for AMD, and more of one than AMD could have ever achieved with any marketing ploy of their own. Undoubtedly there will still be true-blue fanboys who flock to the new cards regardless of their disadvantages, but this self-proclaimed fanboy can't allow himself to be that stupid. Besides, I can still sell the thing down the road when NVIDIA finally gets its shit together.
