Chip Development Process
X-bit labs: Generally speaking, both DeltaChrome and GammaChrome graphics chips utilized the same graphics architecture. Given that you have not declared any tangible feature-set improvements in the Chrome20, does it mean that this is still the architecture originally unveiled in 2002?
Nadeem Mohammad: We have made a conscious decision not make too many disclosures about the internal operation and architecture of our processors. Our DirectX 9 architecture has been improved substantially from DeltaChrome and even GammaChrome, once you have a chance to run the new Chrome20 series in your labs you will find better performance per clock than our previous products.
X-bit labs: How much time does it take to design a GPU from scratch?
Nadeem Mohammad: Starting from the initiation of a new architecture to first silicon can take technology investment of between 2 to 3 years. Fast cycle – time-to-market products can be developed as derivatives in a much shorter time.
X-bit labs: How much time does it take to shrink a graphics core to a new manufacturing process and add minor improvements?
Nadeem Mohammad: It is highly dependent on the actual process transition, standard library model availability and changes. Some process changes can be considered optical shrinks, this usually implies that the same manufacturing equipment is being used in the fabrication line and the process steps are the same – just more precise and tuned for the new geometry. I would say that could be about 9 months, including time for testing and qualification.
X-bit labs: Feature-set wise S3 Graphics was late with some of its products, in our perspective. Did you do anything to improve the situation?
Nadeem Mohammad: We have been improving our time-to-market with each new product – I would be first to acknowledge we still have a lot of room for improvement. We have made many changes to the way we work and operate, particular to reduce the time lag between production silicon to high quality production desktop boards – a critical item which previously was out of our control.
X-bit labs: Tell us a little bit about your Destination Generation products that should sport Shader Model 4.0 capability. When did you start the development? Is it more challenging to design a chip with unified pixel and vertex processors? What manufacturing technology will you use in late 2006 when the Destination products are supposed to arrive?
Nadeem Mohammad: This architecture for this new generation has been in development quite a while, originally as part of our advance research group. This was well before any engagement in Microsoft on the DirectX 10 specification. I can not comment on any of the features or specification of this product family. I will say that I am personally very excited about this design – it will have features and capabilities which I have dreamed about since my first work on graphics processors back in 1987! 2006 year is going to be a milestone year for the entire graphics industry.