Josh Walrath van Penstar Systems heeft een leuk verhaal geschreven over het verleden, het heden en de toekomst van AMD in zijn visie. Een leuk en interessant artikel waar je wel een poosje mee zoet kunt zijn.
The initial Athlons were made with off-die L2 cache that was located on the PCB of the processor, so the Athlon was severely limited in the speeds it could attain due to the availability of high speed SRAMs used in the cache. AMD solved this problem by integrating the L2 cache onto the processor die. AMD was one of the first to do this with their K6-III, a truly groundbreaking (but expensive) processor for them. There are currently two versions of this new Athlon core out, one with 256 KB of L2 (Thunderbird), and one with 64 KB of L2 (Duron). AMD is very smart in its allocation of Fab space with these processors. The high end T-Birds (900 MHz and above) are for the most part made in Dresden, while the rest are made in Austin. Since the Duron is a value processor, it makes no sense in producing it on the copper process in Dresden, so all Durons are fabbed in Austion (Fab 25).
Integrating the L2 cache also helps to keep performance increases linear with successive speed increases. The faster the processor goes, the faster the cache transfers are. This gives AMD more head room since they no longer have to rely on expensive high speed SRAMs from other manufacturers. Also with the inclusion of the L2 cache AMD is able to revert to a socket architecture instead of the bulky and more expensive slot processors. The classic Athlon cost about $110 for AMD to produce from start to finish, by going to the socketed form AMD is saving over 40% in production costs.
In December of this year we will see the next incarnation of the Athlon in the form of the Mustang core. This will feature a reworked Athlon core that takes up less die space, as well as new cache configurations with L2 size increasing up to 2 MB. The consumer level Mustang will feature 512 KB of L2, and other products based on this core will feature 256 KB and 64 KB. There have also been rumors that the connection from the core to the L2 cache will be increased from 64 bits up to 128 bits in width. AMD is tackling a huge range of products with the Mustang, including the mobile space. This will be a lower power part than the current generation of Athlons, and will also include the PowerNow! implementation for mobile devices.
AMD has designed the Athlon core to be used in many different configurations and products. This scalability as well as high performance lends itself nicely to AMD’s business plan of taking one step at a time, and doing it very well. By March of 2001 AMD will have covered the entire spectrum of processor products, from those aimed at the servers and multiprocessor platforms, to the slowest mobile part based on the Athlon core. AMD has set very achievable goals, and has met nearly every one in a timely manner.