Originally posted by: Cookie Monster
Highend was always a toss up between ATi and NVIDIA. But here are the important facts that cleared the paths to nVIDIA success:
-The 6800GT was the key for nVIDIA striking back after the FX debacle. It was cheaper than the flagship cards, OCed up to Ultra speeds with ease, and most importantly beat the X800pro across the board. ATi was too late with the X800XL.
-The 6600GT beat everything on the market at its price range. X700? this card totally failed because of the 6600GT. This card definitely gave nVIDIA the edge. (ATis response was the usual, with a low margin card, in this case the X800GTO)
-Introduction of SLi when NV45 hit the scene. No competition from ATi in the multi GPU scene til later later when RV670 came out with native crossfire which infact is pretty damn competitive with SLi if not better (but this is after 2~3 or so years). Notice how ATi completely abandoned the hyped "super tiling".
-The R520 was way too late, because everyone just went out and bought the 7800GT, GTX cards or two.
-By the time it was released, the X1800XT was barely faster than the 7800GTX 256mb at the time of its release. (Only after a couple of months with optimized drivers did it performed better by alot but by then it was too late anyway because of the refreshes).
-All 7800GTX 512mb were sold and people who
could buy it bought one, or two. For the price premium, the amount of profit earned per card must've been pretty damn high.
-7600GT again dominated the mid range since the X1600 series was too slow. X1800GTO was a poor attempt in ATI's part since it had to use a 256bit memory interface AND a crippled R520 core (no margins at all) just to fight off this mid range card based on the G73.
-X1900 vs 7900 series. Id say it favored ATi, but this was a tossup. G71 cores were tiny compared to the R580. It packed less transistors also compared to the G70. More G71s can be produced per wafer. So in terms of production cost, well overall cost id think that G71 cards had a bigger margin compared to its ATi counterpart. Many people bought 7900GTs, 7900GTXs OCed damn mad. Many people also bought X1900 series cards. When these cards were introduced the 7900GTX was roughly on par with the X1900XT with a slight edge to the ATi card.
-7950GX2 was definitely the fastest single PCI-e slot solution compared to the X1950XTX. But price difference, so it was a tossup here as well. Once again ATi late with X1650XT, and X1950pro but these products did give ATi quite the boost needed.
-G80/G9x, well we all know how this turned out.
You can see that ATi was always late, and seemed to do badly in the mid range (something that contrasts to their high end GPUs). Not to mention the tendency of releasing cards that sometimes had low to no margins at all. They recovered by releasing more competitive products (products such as X700, X800pro, X1600, X1800GTO and so on) but by then its already too late with the next generation of cards close to release.
There are other things that can be mentioned. Innovation is great, but nVIDIA and ATi are business companies also. The keyword here is business. Like for instance the R5x0 architecture excled in great dynamic branching because this part of the architecture was greatly improved. Well was this ever used? Anwser is no. And by the time it does, there will be faster cards out that excel in those situations. These design decisions can later come back to haunt you, like in this case the R580 could have been a cheaper GPU to produce if all these unneeded features were taken away. This is what nVIDIA excels at. Look how theyve successfully turned the NV47 to a 7 series product (potentially saving loads of R&D and other related costs) and still be competitive. The way they introduce features at a timely fashion, not being too late or too early (just look at ATI's tessellation feature, its pretty much wasted transistor budget/die space). An example is S.M 3.0 with NV40. Sure most of its features were too slow to run, but this gave the devs an opportunity to do something new. The kind of support nVIDIA gives to devs is light years ahead of ATi. Just look at the "GPU gems" series and the amount of effort has gone to support the devs in utilizing these features.
Although sometimes this can be harsh on consumers, it gives these companies the edge needed to beat the other. That is why ATI is acquired by AMD and nVIDIA is the last one out of many (yea and i mean many
) that started many years ago in the graphics race.
Ok enough OT, RV770 will potentially bring some competition in the highend. Judging by the specs, im still not sure what its specs going to be. 32 TMU is a given. I highly suspect this RV770 is based on the RV670 with its drawbacks greatly looked at (poor texturing performance, AA performance etc). Im also certain that R700 is dual RV770 like what the R680 is (dual RV670). For reference a 8800Ultra is 30~50% faster than the HD3870.
edit - think i went a little overboard
Facebook
Twitter