AMD held an interesting presentation on the second day of Computex, in hot and humid Taipei [as always]. The company showed its Evergreen GPU in action, but the most interesting part for the financial community was the exchange of hands between Dr John Wei, Senior Director of Advanced Technology inside TSMC and Rick Bergman, AMD’s Senior VP.
Dr Wei gave a 40nm wafer containing Evergreen chip to Rick, giving an excellent opportunity for everyone to get a good look at tomorrows’ DirectX 11 processor. From what we can analyze, Evergreen’s die size is around 180 to 200mm2. This puts the GPU as the smallest of its DirectX 11 brethren, as the Intel’s Larrabee high-end part die is sized around 600mm2. Intel will produce smaller chips as the Larrabee line-up expands, but according to roadmaps that we saw – AMD will launch a new architecture at 28nm before Larrabee hits 32nm with its smaller parts.
nVidia’s GT300 comes at around 500mm2, meaning even if a single "GTX380" card is able to beat ATi’s Goblin [Dual GPU, the "5870X2"], it will still take at least 100mm2 more silicon. This puts AMD in quite an advantageous position.
The list is as follows:
- AMD Evergreen – 180mm2, 40nm TSMC
- AMD Cypress – 360-400mm2 [2x180mm2], 40nm TSMC
- nVIDIA GT300 – 500mm2, 40nm TSMC
- Intel Larrabee – 600mm2, 45nm Intel
Given the performance and pricing of Phenom II and the upcoming Radeon 5000 series, if AMD does not pull into black [records a profit] and achieves great sales success, we don’t know what needs to happen in order for AMD to actually earn some serious money.
All the parts of a cocktail called success are now in.