Ok, I thought I would chime in on this one since I wrote that little blurb on PenStar.
I remember getting excited about Mr. Huang's statements about the transistor count on the NV-18 and NV-28 being in the 80 million mark. But that really didn't make a whole lot of sense, especially for the NV-18. You do not want to produce a budget chip that has that many transistors! It is just too expensive! Also, the NV-18 will be featured in mobile markets as well as the chipset markets in the future, so having 80+ million trannies just doesn't make a whole lot of sense, and NVIDIA has in fact told me that NV-18 won't be significanlty larger than the NV-17! So I am thinking that Jen-Hsun and gang suffered a TIA (transient ischemic attack- aka STROKE) and those numbers are way, way off base.
NVIDIA is in a position that they have to produce a smaller and more affordable processor, while still addressing DX 8.1 compliance at the low end. Taking the current NV-25 and making that a low end chip will cut the margins to nothing on that particular chip, and NVIDIA is all about margins. To address that area, they have to make a smaller chip with the same functionality, and have it be competitive with the Ti 4200 and Radeon 9000 Pro. Making a 2x2 processor makes quite a bit of sense vs a 4x1 (such as the Radeon 9000 pro). My justification in this is that we are honestly no longer fillrate bound. More pipes means more raw fillrate, but with DX 8.1 apps there are more texturing and shading operations going on, so that is the bottleneck. Also, if using the current 1 pixel 2 texture pipeline, it doesn't need to be modified to provide 2x the amount of loopback (like the Radeon 9000 Pro has to do over the Radeon 8500). So with shading operations being the overall bottleneck, keeping with a 2x2 pipe makes a lot of sense vs the way ATI is doing it. And knowing NVIDIA, they would probably optimize the pipelines even more for both the pixel and vertex shaders, so performance most likely would not be much less than a Ti4200, but at over half the cost per chip.
Now, this was entirely speculation from me with little or no confirmation. So do not take this as the truth, but looking at the way the market is moving, this makes quite a bit of sense. Why would NVIDIA make the NV-28 a super GF4 Ti chip, when the NV-30 is their high end product? It would make much more sense for them to be working on a value DX8.1 part, so they can address the mainstream and counter ATI's budget minded moves. As for the NV-18, I highly doubt it will have any DX8.1 functions, but it could be the extreme low end and mobile chip of choice.
As for the .13u speculation, if I remember correctly, Jen-Hsun basically just said that the NV-30 is going to be made on it, and didn't mention that it was going to be the only chip made on that process. NVIDIA has had a long tradition of utilizing advanced process technology accross the board for their value products (GeForce 2 MX, GeForce 4 MX), so I honestly can see NVIDIA going ahead full speed on .13u across the board. From all indications TSMC has finally got their act together on the .13u process and are making significant strides in improving the yields. I would bet the process would be readily available this month, with mass production starting on that line shortly.
As for NV-30.... I have very little faith that it will show up in mass production before January 30th. We may see some running silicon at Comdex, but it will be very close to the first silicon stuff. I have again heard rumors that the NV-30 still hasn't taped out yet (eg. data for the masks sent to TSMC). If this is true, NVIDIA is going to be very hard pressed to have it out in a timely manner. To give an indication where that might be, last year at this time NVIDIA had GeForce 4 Ti samples in their labs for testing and qualification, and I believe it was A1 silicon. The GeForce 4 Ti was released about 7 months after that point. There is going to be a lot of the midnight oil being burnt at NVIDIA for the next few months for anything to be shown in November.
