How Close to 500MHz has NV got?

[RussSchultz]

Anybody have an idea how big the wafers are they're using?

If it holds 125 chips, and they ARE NV30's, and yield is in the normal range (i.e. 75-85%), and the wafers aren't the new 300mm wafers, then I'm guessing thier raw die cost is somewhere around $25-35. Add $5 for testing and packaging for a 30 or 40 dollar part. Double that for the cost to the ODM to give the requisite 50% margin. Double it again for how much it will cost the consumer in his board on the shelf.

I estimated a $3000 wafer cost, elevated it by 25% for yeild loss, then divided that by 125. I wildly guessed on the packaging costs, but .50$ is what a 100 pin tqfp costs so I don't think I'm too far off.

 

[emotionstation]

Now the Ultra version might be explainable in that it just might feature a 256 bit memory interface even though current reports currently blow any theory of such a feature being supported.

David Kirk on the record stated that the NV30 is using a 128bit interface. I believe this was on the are you ready page, as one of the Macro. Flash option buttons. He said they don't feel the need to move to 256bit just yet.

I doubt any 'high end' version of the NV30 would have such a huge architectural difference from one card to another model of the same card.

I'd imagine the NV40 would be it's 256bit brother

 

[kid_crisis]

Anybody have an idea how big the wafers are they're using?

Well, this is supposedly a picture of an NV30 wafer here:

http://www.hardocp.com/image.html?image=MTAzNzQ4MjEyNXlvbURJMmdIUFZfMV80X2wuanBn

Assuming its a pretty large die, say 0.5" by 0.5", then this looks like a 6" wafer. From this photo I would say 125 raw chips per wafer is pretty accurate.

 

[Simon Templar]

then this looks like a 6" wafer.

I'm pretty sure that the process is on a 300mm wafer and that would put the die at about ~1"x1".

Edit: I Found where I read it....

nVidia was unwilling to disclose GeForceFX's die size, but was willing to say that TSMC's 0.13-micron process is running with both 8" (200mm) and 12" (300mm) wafer sizes. They would not however identify which size of wafer the GeForceFX was being produced on.

http://www.extremetech.com/article2/0,3973,710351,00.asp

 

[Mark]

Here is about all I can contribute to this interesting discussion

looks like:
Chip # 739.1
Speedo ("o"?) 210
GL
Testmode

On the right it looks like
500/500
something I can't quite pick out
Fails
something else I can't quite pick out

 

[BoddoZerg]

500/500
Ultrafx
Fails
shoot?

What's up with "speedo"?

 

[LuKE!!!!]

What's up with "speedo"?

I think it says "speeds".

 

[RussSchultz]

It would have been polite for Dave to have remembered what was written on the tag.

 

[ram]

Errrm - I took that picture. It was one of the boards they had running.

Today, I asked specificly about the core clock speed of the demo sample (it was the same that was shown in GB yesterday). The answer was very clear that it is running at full speed of 500 MHz.

 

[Simon F]

Errrm - I took that picture. It was one of the boards they had running.

Today, I asked specificly about the core clock speed of the demo sample (it was the same that was shown in GB yesterday). The answer was very clear that it is running at full speed of 500 MHz.

The tags might well refer to the external memory interface - that might be more difficult to get "to time".

 

[Dave Baumann]

Today, I asked specificly about the core clock speed of the demo sample (it was the same that was shown in GB yesterday). The answer was very clear that it is running at full speed of 500 MHz.

Were the demo's stuttering a little?

 

[ram]

Were the demo's stuttering a little?

Yes. At least one demo (I think it was the dancing character) wasn't fluid at all.

 

[Dave Baumann]

I noticed the truck dome stuttering at out presentation.

Were you at the Munich one then?

 

[ram]

noticed the truck dome stuttering at out presentation

Yes, that one stuttered too. Nevertheless, it was quite impressive. STALKER was interesting too, moving grass is much cooler than the boring static grass in UT2K3. Though none of the demos had FSAA turned on, texture aliasing was horrible in that game.

Were you at the Munich one then?

Yes. Sitting right beside "Fuad".

 

[Dave Baumann]

Yes, that one stuttered too. Nevertheless, it was quite impressive.

Truck was least impressive IMO. The dancing Ogre's were very well animated. The detail on the fairy was very impressive as well (if not a little overdone on her skin), but it appeared to be just a flat background.

Yes. Sitting right beside "Fuad".

LOL. One of these days I'll get to meet him, I've 'just missed' him on a number of occasions now.

 

[ram]

The dancing Ogre's were very well animated. The detail on the fairy was very impressive as well (if not a little overdone on her skin), but it appeared to be just a flat background.

The "cloth simulation" in that demo was quite cool IMO, but the character itself lacked details like hairs over the original.