Fudo of Fudzilla, one of the few reporters calling for a November/December 2009 release (a rumour which has now been squashed), suggests Nvidia Fermi is only going to ship with an A3 revision.

Currently, we are at A2, which was rumoured to have been taped out in WW42, which was end of October 2009. What is baffling is how Fudzilla is speculating a January 2010 release date when we are still one silicon behind, and they themselves claimed in a previous article GPU retail availability takes 3-6 months from tape out.

More next page.


Fudo of Fudzilla, one of the few reporters calling for a
November/December 2009 release (a rumour which has now been squashed),
suggests Nvidia Fermi is only going to ship with an A3 revision.

Currently,
we are at A2, which was rumoured to have been taped out in WW42, which
was end of October 2009. What is baffling is how Fudzilla is
speculating a January 2010 release date when we are still one silicon
behind, and they themselves claimed in a previous article GPU retail availability takes 3-6 months from tape out.

While we are starved for facts – let us take an Occam’s Razor approach, and logically time out things with the best known evidence.

Back in July, Charlie of SemiAccurate revealed that Fermi (then referred to as GT300) had taped out in WW28. This was the first silicon, which Nvidia calls A1. (Unlike CPU makers, who generally refer to it as A0) He further predicted that there would be no Nvidia next-gen cards for the entire 2009. At the time, this idea was ridiculed, and most admitted that while GT300 will come after Evergreen, it will definitely release by December 2009. Many months on, Charlie’s sources have proved to be spot on.

Fourteen weeks later, WW42, end of October 2009, Nvidia taped out A2. We are looking at a straight 6-8 weeks before Nvidia receive the A2 silicon. Now, it appears that an A3 silicon is required, something which has been rumoured, and something Fudzilla is now suggesting. Let us assume Nvidia can speed up and tape out the A3 silicon within 10 weeks (as opposed to the 14 weeks between A1 and A2). That takes us into WW52 and end December 2009. Assuming this is the final retail silicon, and everything has gone according to plan, we are looking at 3 months, at best, till retail availability from tape-out. This takes us dangerously close to end of March, or dangerously close to Q2 2010. So far, all evidence suggests that Fermi has had a bumpy ride, so it is unlikely from here on everything will be perfect. Even minor slip ups and Fermi’s release is well into Q2 2010; April, maybe May. Let us hope there are no major problems, anything beyond Q2 will be disastrous as ATI will be preparing shrinks, if not the next N. Islands generation.

A January 2010 release makes very little sense – even if A2 was the final silicon. Perhaps January 2010 is the date Nvidia receives their A3 samples? Even for that, a February date seems more likely. Take mass production of the chips, assembly of graphics boards, printing of boxes, etc. and we are well into March, best case scenario.

Regardless, all logic seems to point at Nvidia’s monster chip being in trouble at a troublesome 40nm TSMC process. At this point, rumours mean little – it has just got to the point where we cannot be sure of anything till we see the final product. If you holding your breath for Fermi, it is best you exhale, for now.

Reference: Fudzilla, SemiAccurate