Nvidia GT200b rumours and speculation thread

[fellix]

From a newer XFX XXX:
... so it seems there is a better version of GT200A out, since usual GTX 280 an older XXX used G200-300-A2?

Here is, what's hidden under the hood of the latest G200-302 batch:

Still no 55nm in there.

 

[BRiT]

They likely have excess inventory of older GPUs to burn through...

 

[DegustatoR]

Still no 55nm in there.

Well, actually it looks exactly like that 'mystery chip' i've posted a month ago and as far as i can tell from this photo it's not 578mm^2 either...

 

[fellix]

Well, the straight visual difference between 578 and 484 sq.mm shrink is not that perceivable at first glance.

-Edit-

Side-by-side comparison tells enough (perspective corrected):

 

[DegustatoR]

Yeah, that's better =)

 

[Blazkowicz]

I think I read here that Tesla cards would be first to use the 55nm GPU (perf/watt is more vital in supercomputing than in the high-end gamer market).
That Quadro FX5800 has Tesla specs (the speed and amount of ram) and is probably intended for an amount of GPGPU too.

 

[Psycho]

From what I can see on the board shot from http://www.xtremesystems.org/forums/showthread.php?t=207011 it is 24.8*23.8 = 590 mm2, using the same metric as we used for getting 275mm2 for RV770, using the 89mm pcie connector as reference. The resolution is limited so it's not as accurate, probably within 3%.

 

[fellix]

In fact, the die-count on the 300mm GT200 wafer suggests an area, larger than the officially announced 576 sq.mm, anyway.

 

[Jawed]

In fact, the die-count on the 300mm GT200 wafer suggests an area, larger than the officially announced 576 sq.mm, anyway.

But some of that is inter-die waste isn't it?

Jawed

 

[fellix]

Here is a really big shot -- not really well packed ASIC.

 

[Jawed]

Ooh, nice picture

Though it doesn't answer the question as to the amount of wastage there is between dies on the wafer. How much area does the sawing of a wafer into individual dies consume?

Jawed

 

[fellix]

Relative margins are very small, I presume, especially for big chips.

 

[HAL]

http://www.dvhardware.net/article31192.html
good. very good.

 

[Jawed]

http://www.dvhardware.net/article31192.html
good. very good.

If you're going to link something, at least link the original article that's sourced:

http://theovalich.wordpress.com/2008/11/11/55nm-gt206-gpu-powers-both-gtx290-and-quadro-fx-5800/

where we see the following addendum:

UPDATE: Picture that accompanied the story did not feature GT206 chip, thus I removed it. The rest of the info is pretty valid

Jawed

 

[CarstenS]

Me, I need to see it for myself before i believe it. Only TDP indicates a lower power chip and in turn for GT200b.


If I was Nvidia, I'd strip GT200b of DP-Units for improving margins on consumer-Level products and keep the enourmously high-margin Quadro and Tesla at 65nm (lower-) volume production until the next chip.

 

[Blazkowicz]

thus designing a entirely new chip about to be replaced by GT206 and GT212. There's a reason half-node re-releases are also called an optical shrink, and it's the same reason they make save money.
Power usage is also critical for the market where quadro and tesla sells, whereas most high end gamers don't care.

the cheaper variant will be here, the rumored GT206, probably that one won't have the DP units.

 

[HAL]

If you're going to link something, at least link the original article that's sourced:

http://theovalich.wordpress.com/2008/11/11/55nm-gt206-gpu-powers-both-gtx290-and-quadro-fx-5800/

where we see the following addendum:




Jawed

OK. I hawen't seen it.

 

[silent_guy]

How much area does the sawing of a wafer into individual dies consume?

The space between dies ("scribe") is between 100 and 200 um wide.

 

[Jawed]

The space between dies ("scribe") is between 100 and 200 um wide.

So about 1% of the area of GT200's die.

Jawed

 

[CarstenS]

thus designing a entirely new chip about to be replaced by GT206 and GT212. There's a reason half-node re-releases are also called an optical shrink, and it's the same reason they make save money.
Power usage is also critical for the market where quadro and tesla sells, whereas most high end gamers don't care.

I am aware of that, but given Nvidias current situation, I think, a reassessment of old wisdom might be in order, no?