With the fall season fast approaching, NVIDIA appears to be getting ready to unleash their latest chips to the market. The NV-30 is now quite well known, and it will be the top offering when it is released in late Fall. This will replace the GeForce 4 Ti 4600 at the very top end, and it appears that the Ti 4600, 4400, and 4200 will step down a notch. In fact, the Ti 4400 and Ti 4200 may be discontinued after a fashion due to the price of the chips themselves as compared to what the cards will be selling at. The Ti 4600 looks to stay around the $200 to $250 mark, and will still be an enthusiast level card. Where then does this leave the bottom end?
I have a feeling that the NV-28 chip from NVIDIA will not be an advanced refresh from the current NV-25 chip. In fact, it is looking to be a cut down version of the NV-25, which will give the mainstream and low end full DX 8.1 compatibility for NVIDIA. Unlike the ATI Radeon 9000, which has 4 pipelines with a single texture unit per pipe, it appears that the NV-28 will only have two pipelines, with 2 texturing units per pipe. Instead of the dual Vertex Shader units on the NV-25, the NV-28 will feature one. All of the other parts look to be about the same (LMA II, AA engine, nView), but I am pretty sure the Video Processing Engine will be added to the NV-28 (such as the one in the NV-17). This chip will probably be as fast overall as the Ti 4200, but it will be much, much cheaper to produce. The NV-28 would weigh in around 36 million transistors, while the NV-25 core is around 64 million transistors. Die size would be approximately half of what the NV-25 is, assuming that NVIDIA will use the .15u process for this new chip. If NVIDIA goes ahead and uses the new .13u process, then the die sizes will be nearly 1/4th that of the current NV-25. This would make each chip substantially cheaper to produce, and they would be able to easily clock to 350 MHz or so. The NV-28 would of course also be a AGP 8X part. The die shrink and redesign would also allow the NV-28 to be used in high end notebooks, providing competition with ATI's recently announced Mobility Radeon 9000. It would then include all of the power saving tricks that NVIDIA has used in the NV-17M chip. The NV-28 appears to be the mainstream warrior for NVIDIA this coming season, and will provide excellent competition to ATI's lineup.
This leaves the NV-18 in a very interesting situation. Where exactly will it be placed in the lineup? It looks like the NV-18 is a slightly modified NV-17 core, with AGP 8X functionality as well as some other tweaks. It will still not be a DX 8.1 part, but its transistor count is reportedly around 17 million or so. In question is if it will be produced on the .13u process also. If so, then this will be an incredibly small die as compared to the rest of the products, and there will be hundreds of them per wafer (making each die very, very cheap). The new process would make it ideal for the extreme low end desktop (replacing the GeForce 2 MX series entirely) as well as thin and light notebooks. The clock speed would gain another boost, and notebook performance would also improve.
Remember, this is all speculation based on NVIDIA's past performance as well as rumors I have been hearing. This overall philosophy does seem to fit nicely into the big scheme of things, but I could easily be very, very wrong. Take a large bucket of salt with this one, but something about it just rings true to me.