The eternal fight between the former ATI Technologies, now the graphics product group of AMD, and Nvidia Corporation is not unlike the famous space race between the USA and the Soviet Union as each side is trying to outpace the other by any, even irrational, means. Nvidia’s G200 adventure may be compared to the Space Shuttle program, for example. Originally designed as the ultimate weapon that would have unrivalled performance and functionality, the chip came out too complex and expensive for the tech process Nvidia had at hand and ATI, having introduced the less advanced but cheaper RV770, easily got the upper hand. The only thing Nvidia could do was cut the price of its first generation of G200-based products and quickly transfer the graphics core to thinner 55nm tech process to make its production commercially profitable.
The renewed hostilities between the AMD and Nvidia camps appeared to have grown into a positional war when September 22, 2009, the former ATI Technologies announced the world’s first DirectX 11 compatible GPU codenamed RV870 Cypress. That was a devastating blow as AMD’s new single-GPU flagship was overall as fast as the older dual-processor Radeon HD 4870 X2. The dual-GPU Radeon HD 5970, announced on November 18, confirmed AMD’s technical superiority over Nvidia.
One might think that Nvidia should have learned the lesson that developing a rather inexpensive and well-balanced mainstream GPU is strategically more profitable than issuing a wonder chip that would deliver unsurpassed performance but would also be too expensive and complex to be a commercial success. This is not the case, however. Nvidia once again took to developing yet another ultrafast GPU codenamed GF100 Fermi. Striving to reach an unprecedented performance, Nvidia broke its earlier and dubious record of maximum transistor count. The G200 incorporated 1.4 billion transistors manufactured on 65nm tech process whereas the GF100 consists of 3 billion transistors on 40nm tech process. As a result, Nvidia found itself in the same trap as with the G200. The chip came out too complex and too expensive to make. The chip yield, on the contrary, was too low to make it commercially profitable. And while AMD was rolling out one Radeon HD 5000 series product after another, Nvidia had to postpone the release of Fermi-based graphics cards till 2010 and lose some more ground on the market of top-end discrete graphics cards.
The GF100 Fermi made it to the market eventually, having lost 1 multiprocessor block on the way. Nvidia had mentioned 512 shader processors but had to cut the number of active ALUs to 480 in order to achieve an acceptable chip yield. On March 26, 2010, half a year after the official announcement of AMD’s RV870, Nvidia’s new-generation architecture saw the light of day, being embodied in a series of graphics cards. We dedicated our review Nvidia GeForce GTX 480 – Our First Date to the senior model of the series and now we’ve got the junior model GeForce GTX 470 which is going to be interesting for a larger category of users. We will test it and see how appealing it is in comparison with AMD’s solutions.