Found this from a link in another forum, great read :The Path to the RV770
Tags:ATI Radeon, AMD GAME!, CrossFireX, Gaming, HD 4800, RV770
After we launched the RV770, I had spent some time on gaming and hardware enthusiast message boards reading what gamers really thought of our new baby. It became quite evident that it wasn’t just the competition that was totally surprised. It was as if some Alien 3D Demigod just dropped the RV770 in AMD’s quiver of products. How did AMD do it? Gamers wanted to know. Let us give you guys some insight in our development process and how the RV770 was created.
Fade back to 2005….. We were preparing the R520 for production and some of the estimated yield, cost and power data raised some eyebrows within the team. Red flags were being raised around the company as this was going to be the biggest and most power hungry ASIC that we had ever built.
Since the R580 and R600 at the time were essentially baked, we were already planning for the R600 successor in 2005.
We had a series of meetings at that time to discuss the future of our graphics strategy post-R600. As with any good company, a faction formed within the graphics team that wanted to embark in a new bold direction of designing more efficient and compact GPUs because they believed the big monolithic approach was coming to an end. This faction was pitted against a company that just had delivered killer products with the R300 and R350. The single monolithic GPU had served ATI well. This was going to be a fun series of meetings…but the discussion was robust and we discussed things like:
We were one generation away from the 300 watt limit for a graphics card
The need to transform the fixed function GPU into a programmable device. With the trend towards programmability, the old days of ripping out pipes and texture units from high-end parts to build low-end parts were disappearing and along with them, the ability to scale down big GPUs into smaller GPUs
Notebook gaming was becoming much more significant and clearly we had to enable gamers to play on their notebooks!
Even though large monolithic GPUs allowed us to pound our chest and declare victory as the builders of the single fastest piece of “sand” in the world, their high price meant that they only serviced a very small percentage of gamers
Through heated debate with lots of name calling, we made up our mind in the warm & stinky El Capitan conference room in the summer of 2005. We decided that we were no longer going to pursue big monolithic ASICs after the R600. After many beers, all factions within ATI rallied around the strategy as one single team.
Many thought that this was perhaps the biggest bet that we had ever made as a company. Our strategy changed from building the “biggest and baddest” GPUs to focusing on smaller yet powerful ASICs targeted at the sweet spot in the market, the $200 - $300 segments. We decided we would use our ATI CrossFireX™ technology to provide killer products for the rabid enthusiasts and we would develop architectures that would scale nicely for the mainstream and value oriented GPUs.
Now move ahead to the spring of 2006…. It was hot and sticky as my team and I stepped outside of our hotel in Taipei, we had a full week of meetings booked with power supply and chassis business partners. We were tasked to get support for the R600. We were in unchartered territory with the R600 in terms of size, weight and power for a graphics card. We knew the R600 was going to consume a lot of power, and with the current PSU technology, there was a risk that we could not have fed the R600 beast. Fortunately we had great support from these business partners and when the R600 came back at well over 200 watts and sounding like a jet, we knew we were on the right path for ASIC design.
The R600 is a good product but products are always judged against the competition. We also knew our Anti-Aliasing on the R600 could be improved. We were at another decision point:
Shrink the existing R600 to 55nm and become experts at 55nm and then develop a new architecture specifically for 55nm as planned
Redesign the R600 and improve the AA performance in 65nm technology
Develop a new architecture on the existing 65nm technology
Develop a new architecture on the new 55nm technology at the same time
After more meetings and name calling, we stuck to our guns on our strategy and plan because it was abundantly clear that we were on the right path. We also knew we were in for some short-term pain. If we were not vying to be the top dog in graphics, we knew our critics would accuse us of quitting or not investing in graphics anymore. We chose to shrink the R600 into 55nm with some minor fixes as planned. This was the birth of the RV670, the first product in our new strategy.
What a terrific product. We went into production on first silicon and it had a lot of the performance of the R600 but consumed a lot less power and yielded well. But this was not the magic bullet to capture the technology leadership and performance crown. Shrinking an existing architecture to a new process technology was not going to drastically improve performance. We knew that we were going to have to develop a new architecture in 55nm to challenge for outright supremacy.
Some people may ask, “why didn’t you develop a new architecture and use the new 55nm technology at the same time”? This is something that no semiconductor company does well consistently. The reason is fairly simple, you typically end up chasing both architecture and process issues at the same time resulting in significant product delays. This was too big of a risk.
Ironically, the cat calls from outsiders claiming AMD had given up on high performance ATI Radeon™ graphics came at a time when we were supremely confident in our strategy. We were emboldened by how well the RV670 turned out. We had to bite our tongues and focus on the new architecture.
In June 2008, we launched the fruits of the 3 year investment in the new ASIC strategy, the ATI Radeon™ HD 4850 and the ATI Radeon™ HD 4870. These both embody a new 3D architecture on 55nm technology with killer video features to boot! The 4870 delivers over 2 times the performance of the R600 at approximately 60% of the power and features a whopping 800 shader processors in a small die, that’s an amazing 2.5X more shader processing power over the ATI Radeon HD 3870! That is progress. Of course it does very well against competitive products, so much so they had to react aggressively with pricing to keep their product line even remotely interesting.
What people don’t realize is that the RV770 is just the first salvo of our new strategy. The R700 is almost ready to launch and also we plan to have a little somethin somethin for our customers in the mainstream and value spaces very shortly. And we believe that this 3 year investment in our “small but powerful” ASIC strategy was a necessary step to help set AMD up for success with the 7 series of products and ultimately for the 8 and 9 series of products.
AMD giving up on graphics!?! Ha! Let me tell you, the mission from Rick Bergman and other senior management has never been clearer… Become both the technical and financial leaders in the graphics industry. As a 10 year veteran of ATI / AMD, I have never seen a team more focused on this goal!
Godfrey Cheng is Director, Product Marketing at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.