On Wed, 14 Apr 2004 15:59:38 -0500, duralisis
wrote:




Keep in mind that the reviewer samples are only an A1 revision
engineering reference board, straight out of the bin. It's likely that
come production time, each manufacturer will choose things such as a
cooler, the mosfets, caps, vregs, etc.


The 110 watts is nVidia's max. spec.....according to the reviews.

So by the time you get your hands on one (a GF6-6800U), most card
manufacturers will likely go with a _strict_ reference design (including
the HSF and ramsinks), but you might find some that just barely fit in a
single slot with a different cooler.

Also, the voltage requirements right now are a "clean 12v", so depending
on what components are certified by NVidia, some manufacturers /JUST
MIGHT/ be able to get away with a single plug from a good PSU (like a
high rated 350w). NVidia's Q.C. program they've inacted since the FX
series launch is likely going to be the limiting factor in this.


I agree. But I will go with the 2-plug version if it is available. The
lower the PS impedance at the board, the less noisy the power
on the board.

What I'm really excited about is the next card revision. In the same way
the NV35 completed the promise of the NV30, The NV45 (or whatever core
revision gets stuck on the PCB in 6 mo.'s), will likely be a much more
efficient and slimmed fighter.

Not likely for quite a while. Manipulation in the current 0.13u
process will have very little power-savings without cutting
functionality. And the masking cost is horrendous. Probably
at least $1 million for a chip this size, assuming first-pass
no errors...

I would expect the next real iteration to be either a .09u
..065u shrink. IBM is working with AMD on .065u. And that
sure won't happen in six months.

The NV30 to NV35 interation was a significant DESIGN
improvement on the SAME process. The design of the NV40
seems near-perfect for the current graphics state-of-the-art.
Only benefit would be a mask-shrink, which would
potentially improve yield (assuming a stable process)
and significantly raise the number of die per wafer -
thus doubly reducing production costs.

John Lewis


Most likely even better memory
bandwidth, some tweaked RGMSAA modes, probably a few Doom III tricks ,
and hopefully a much more efficient power usage.

As for now, it looks like brute force is doing pretty well.