Nvidia still sells 35% TNT2s

I've even seen a P4 computer with TNT2 M64. Great, isn't it?

 

No. 'Officially' Ti 4200 will come in at the $199 price point, Mx 460 is cheaper:

http://216.12.218.25/domain/www.beyond3d.com/articles/gf4launch/index7.php

 

Basically, unless the street price of the 460 MX comes down considerably from the $179 MSRP...that's the most ill conceived card in nVidia's line up....

 

Kristof is right.

Probably GF4MX has only 60% of the die area (silicon space) of GF3 Ti200 and much better yields, then it will be much less expensive to build and much, much more profit.

Today the GF3 Ti200 is as low as $132, probably Nvidia will not make much money with current GF3 under $100 then they need to replace it with some money maker chip.

Low cost DX8/DX9 will probably need .13 micron process (smaller die better yields).

<font size=-1>[ This Message was edited by: pascal on 2002-02-15 21:24 ]</font>

 

Better yield comes with time, definitely not with upgrades to lower micron processes. A smaller micron gives you potentially higher clockspeeds (mainly due to lower voltage needed and thus less heat) but most importantly it gives you more functionality on the same or smaller silicon space. In the end the cost is the silicon wafer, the more chips you can cram on one single waffer the more money you can make.

Now if ATI or NVIDIA or somebody else undercuts prices then it means either :

- They make less profit per chip
- They are cutting some functionality out (and reduce die space)
- They are on a more advanced process (lower micron than competitor "with" high yield)

 

Well, you could argue that GF3 Ti200 is mainstream/mid end from NVIDIA. They are cheaper than R8500 here in Norway. (And they are dx8 cards). Even when GF4 comes I am pretty sure people will continue to buy Ti200 cards. They are faster and got more features than GF4-MX. But then again GF4-MX is cheaper.

 

Unless of course Nvidia cancels production of the chips, which I would hate to see.

 

Y'all are making assumptions on a R8800 which I hope you know what happens when you assume something .

 

On the issue of silicon wafers, how much does a silicon wafer cost?

 

400-1000$ for 300mm prime wafers, depending on the quality (smaller processes need better quality). Reclaimed ("raw") 300mm wafers cost just 20-40$.

[Last number is wrong, its for old 200mm wafers. Reclaimed 300mm wafers are more expensive, from 50-100$]

<font size=-1>[ This Message was edited by: Nexus on 2002-02-16 03:40 ]</font>

 

You want to tell us something..?

 

Haven't read the GF4 launch articles? nVidia will phase out its existing product line, except Geforce2MX. No more GF3 Ti200.

 

As long as they are in stock.. else they have to wait for nVidia to release GF4 Ti4200. Lets hope its coming soon.

 

Nexus, I don't think it's appropiate, given the source of the R8800 rumors and what my source said , to comment really . I'll have to leave it at that.

 

What I was saying is that a .13 micron GF3 Ti200 potentially has better yields than a .15 micron GF3 Ti200. Same transistors -> smaller chip -> better yields

It will not be soon because they dont want to charge $200 for something they are selling for $130 now (GF3 Ti200). First they have to make GF3 Ti200 dissapear.

 

I'm wondering about this card too. We know that it'll come out soon, but will it just be a speed bumped 8500 or did ATI tweak it some (like a nVidia refresh part)?

Someone called "david" over at the rage3d forum claims to have access to ATI somehow. He wrote this recently:

He also claims that the card will be clocked at 350 MHz core/mem.

An other piece:

It could all very well be pure BS of course. But at least I'll hold off buying a new card this spring until I know how 8800 compares to GF4 ti4600.

Regards, LeStoffer

 

Depends on what "yield" means for you. For you it is:

Number of functional chips on the wafer.

But in the semiconductor industry it means:

Percentage of functional chips on the wafer compared to non-functional.

Lets say 0.15 process is already very reliable because its running for one year. Lets assume that with it you can put 150 chips on the wafer and 140 of it are functional.

On the other side, lets say 0.13 process is quite unreliable because its running just for 1 month. Lets assume that with it you can put 220 chips on the wafer but only 160 of it are functional.

Which process has better yields now?


(Note: I just put some numbers out of the air to make good examples, they mustn't be real life like and mathematically correct )

 

How would you factor clock speed into those definitions of yield?

 

Something that we tend to forget is the fact that it cost huge amounts of money to make a "plant" for a new process in the first place.

It really takes a lot of production on that new process before all the money invested is paid back. So it's not at all fair just to say: "Hey great, at smaller process, the chips will cost less now!". I don't know how TSMC gets paid, but I guess that companies like nVidida has to pay more for using the .13 process than .18 process. There's one more reason that the TNT-line is not made om .15 AFAIK.

Regards, LeStoffer

 

Well, if you define "functional" and "non-functional" with certain clock speed ranges you have factored clock speed into the definition. A Geforce4 chip from TSMC may work at 120MHz, but for Nvidia its non-functional because they have no need for it. Just like the number of non-functional chips decreases when a process matures, the number of higher clockable chips increases. In my example I used a black/white view of things, in real life the transistion is fluent. You can draw those nice curves of how many chips can clock how fast. Normally ~80% of all functional chips are within 10% clock speed range (IIRC). Maybe this percentages are wider with modern GPU transistor monsters.