NV cuts the GF4MX transistor count, how much?

[Slashhead]

The GF3 has 57M and the GF4Ti 63M. How many transistors are used for the GF4MX?
I'm intersted, since it looks like the transistor cut is the main reason for producing cheaper processors.

 

[Sabastian]

The GF4MX is not a GF4 chip as far as I know.(runs while franticly looking for a place to hide before the impending flames.)

Sabastian.

 

[Slashhead]

I know it is not the same as the GF4Ti. (missing infinteFX, stripped LMA-2) That's why i'm interested in the transistor count of that chip.

 

[easyride]

The GF4MX isn't even a GF3... much worse both in features and fill rate.

 

[Slashhead]

Yes I know !!!!!
But that is not the question. :smile:

I only want to know the transistor count..... please...

 

[Thowllly]

Maybe nobody knows?

But I think that the transistor count increase in the MX series would be no more that the Gf3->Gf4 increase of 6 million. Gf2mx -> Gf4mx gained more transistors dedicated to AA than Gf3 -> Gf4 (since the Gf3 allready had HW AA), but on the other hand the Gf4 got something completely new, dual display suport, wich the mx allready had partial support for. And I don't think the MX got an extra T&L unit.

I can't quite remember how many transisters the gf2 mx had, but I think it was less than 20 million, so I would guess that the Gf4mx has about 20-22 million transistors.

Pure speculations, nothing you haven't allready thought of youself, I guess :smile:

 

[Humus]

26M. Not sure where I read it though, but it was in one of the reviews.

 

[Slashhead]

yes i figured out that has to be less then 30M to produce a relative cheap processor. This means the count is cut in half. Looking at the space the nFiniteFX engine takes, it has to be a lot less, also removing 2 pixel pipes and a stripped LMA-2 is saving transistors. Probably the decrease of transistors was a very important step in the GF4MX design. That's why i want numbers.
It is hard comparing with the GF2MX since there is logic for 128bits DDR memory, Accuview, Quincunx and nView present.

26M looks very realistic.

<font size=-1>[ This Message was edited by: Slashhead on 2002-02-11 17:38 ]</font>