The RV770 Story: Documenting ATi's Road to Success
Citiraj:
When RV770 launched in the summer we took for granted that it was a great part, it upset nVidia’s pricing structure and gave us value at $200 and $300. We went through the architecture of the Radeon HD 4800 series and looked at performance, but I spent only a page or so talking about AMD’s small-die strategy that ultimately resulted in the RV770 GPU. AMD had spent much of the past 8 years building bigger and bigger GPUs yet with the RV770 AMD reversed the trend, and I didn’t even catch it. I casually mentioned it, talked about how it was a different approach than the one nVidia took, but I didn’t dig deeper.
It all started back in 2001 when ATi, independent at the time, was working on the R300 GPU (Radeon 9700 Pro). If you were following the industry at all back then, you’d never forget the R300. nVidia was steadily gaining steam and nothing ATi could do was enough to dethrone the king. The original Radeon was a nice attempt but poor drivers and no real performance advantage kept nVidia customers loyal. If ATi could produce the fastest GPU, it would get the brand recognition and loyalty necessary to not only sell those high end GPUs but also lower end models at cheaper price points. The GPU would hit the high end first, but within the next 6 - 12 months we’d see derivatives for lower market segments. One important takeaway is that at this point, the high end of the market was $399 - keep that in mind. R300 was going to be built on the same process, but with 110M transistors - nearly twice that of the 8500 without a die shrink. Its competition, the GeForce4 was still only a 63M transistor chip and even nVidia didn’t dare to build something so big on the 150nm node, the GF4 successor would wait for 130nm. We all know how the story unfolded from here. The R300 was eventually branded the ATi Radeon 9700 Pro and mopped the floor with the GeForce4.
What made R300 successful was ATi re-evaluating the way it made GPUs and deciding on something that made sense. At the time, it made sense for ATi to work towards building the biggest chip possible, win at the high end and translate that into better, more competitive products at all other price points. It is worth mentioning that part of R300’s continued success was due to the fact that nVidia slipped up royally with the GeForce FX, it was the perfect storm and ATi capitalized. ATi built the biggest chip and nVidia had no real answer for some time to come. ATi didn’t want to build a GPU that would rely on excessive repair and harvesting to keep yields high. And then came the killer argument: building such a GPU was no longer in the best interests of its customers. ATi’s RV770 design took three years of work, that means ATi started in 2005. It takes about a year for manufacturability from tapeout to final product ship, another 1 - 1.5 years for design and another year for architecture. The team that I met with in Santa Clara was talking about designs that would be out in 2012, we’re talking 4 TFLOPS of performance here, 4x the speed of RV770.
http://pce.hostili.com/images/diecomparisonuuy.png It was time to refocus. Instead of tailoring to the needs of the high end, ATi wanted to make a product that would be the best in the $200 - $300 range. To do so would mean that it would have to reverse the strategy that made it successful to begin with, and hope that somehow nVidia wouldn’t follow suit. In the Spring of 2005 ATi had R480 on the market (Radeon X850 series), a 130nm chip that was a mild improvement over R420 another 130nm chip (Radeon X800 series). ATi was still trying to work through execution on the R520, which was the Radeon X1800, but as you may remember that part was delayed. ATi was having a problem with the chip at the time, with a particular piece of IP. The R520 delay ended up causing a ripple that affected everything in the pipeline, including the R600 which itself was delayed for other reasons as well. Not only was G80 good, but R600 was late, very late. Still impacted by the R520 delay, R600 had a serious problem with its AA resolve hardware that took a while to work through and ended up being a part that wasn’t very competitive. ATi had lost the halo, ATI’s biggest chip ever couldn’t compete with nVidia’s big chip and for the next year ATI’s revenues and marketshare would suffer.
With its only competitor hell bent on making bigger and bigger GPUs, ATi took care of half of the problem - it would be free to do whatever it’d like, without any real competition. The question then became - could it work? But you have to understand that this was 2005 and the first specifications of RV770 were being drafted. Imagine sitting at a table full of people whose jobs were supported by building the biggest GPUs in the world and suggesting that perhaps we sit this round out. In the Spring of 2005, ATi decided to shoot for the Performance segment, and not Enthusiast. The entire RV770 design took around three years, which means that while we were beating ATi up over the failure that was R600, those very engineers had to go into work and be positive about RV770. The R600 GPU had an incredibly wide 512-bit memory interface, the problem with such a large interface is that it artificially makes your die bigger as you’ve got to route those interface pads to the memory devices on the board. For RV770 to have the die size ATi wanted, it needed to have a 256-bit memory interface, but using (at the time) current memory technology that wouldn’t give the GPU enough memory bandwidth to hit the performance targets ATi wanted.
Again, put yourself in ATi’s shoes, the time was 2005 and ATi had just decided to completely throw away the past few years of how-to-win-the-GPU-race and on top of that, even if the strategy were to succeed it would depend on a memory technology that hadn't even been prototyped yet. The spec wasn’t finalized for GDDR5 at the time, there were no test devices, no interface design, nothing. Just an idea that at some point, there would be memory that could offer twice the bandwidth per pin of GDDR3, which would give ATi the bandwidth of a 512-bit bus, but with a physical 256-bit bus. It’s exactly what ATi needed, so it’s exactly what ATI decided to go with. ATi did much of the heavy lifting with the move to GDDR5, and it was risky because even if RV770 worked out perfectly but the memory wasn’t ready in time the GPU would get delayed. RV770 was married to GDDR5 memory, there was no other option, if in three years GDDR5 didn’t ship or had problems, then ATi would not only have no high end GPU, but it would have no performance GPU to sell into the market.
Initially, RV770 was targeted at 1.5x the performance of R600, which looking back would not have been enough. During the next 1.5 years that 1.5x turned into 2x R600 and finally settled at 2.5x the speed of R600, at a price in the $200 - $300 range. The Radeon HD 4850 was originally a 256MB card with a 500MHz core clock and 900MHz memory clock. The RV770 products were finished in May of 2008, production started by June. Even up until the day that the embargo lifted there were some within ATi who felt they had made a mistake with the smaller-die strategy, but they were going to find out how right the strategy was, even sooner than expected. Within 30 hours we had our first preview up and made it already clear that ATi was on to something. The GeForce 9800 GTX got an abrupt price drop to remain competitive and even then it wasn’t enough, the Radeon HD 4850 was the card to get at $199.
The next question is how will nVidia respond to ATi’s strategy? Jen Hsun runs a very tight ship over there and does not take kindly to losing, especially not like this. nVidia continues to have very strong engineering talent and over the next couple of years we’ll see how RV770 has impacted nVidia’s development. Much like the R300 days, the success of the RV770 was partially ensured by nVidia’s failure. Unlike NV30 however, GT200 wasn’t delayed nor was it terribly underperforming - it was simply overpriced. ATi got very lucky with RV770, nVidia was tied up making a huge chip and avoided two major risks: 55nm and GDDR5, both of which ATi capitalized on. If you ended up buying a Radeon HD 4800 or derivative, you already know why you’re thankful. If you ended up buying something green, you most likely paid a much lower price than you would have.
|
Izvor: AnandTech
Sve u svemu zanimljiva priča i pregled događaja koji su doveli do današnjih RV770 čipova i stanja na tržištu.:) |
