BRIGHT SIDE OF NEWS About | Advertise | Contact BSN USER Login
| Register 		SUBSCRIBE Newsletter | RSS Feeds
Wednesday, June 19, 2013
Email this to a friend.
Your friend's e-mail:
Your Name:
Your e-mail:
Message subject:
Comments
I agree with the General by: Gipsel on 5/14/2009
The die size isn't surprising.
(0.4/0.55) * 2.4/1.4 = 0.9
Factoring in the non ideal scalig and I would also expect a die size similar to the GT200b.

But you should consider something else. AMDs RV740 contains more than 800 million transistors (826 million if I remember right) and has a die size below 140 mm² That means they could produce a 2.4 billion transistor GPU with a 384 bit memory interface in roughly 400 mm².
Actually, I'm guessing RV870 to stay below 2 billion transistors and to have a die size of not much more than 300mm². Still, performance wise it could be right on the heels of GT300 (especially with aggressive pricing). The GT300 concept has to work better than the GT200 one, otherwise it won't help much financially.
Why's that a surprise? by: General Lee D. Mented on 5/12/2009
They've taken such a beating on cost I'd be surprised if they didn't aggressively come back with a squeezed die. Nvidia's strength was always when they listened to the market and fixed their biggest weakness (see Timeline: 3dfx Revisited, Aceshardware 2003).

As for transistor count figure 400-500k is carried over as the 512bit memory controller, its cache, and the ROPs and suddenly 1billion->2billion for 240->512 doesn't seem so improbable does it?

As for the die size... I dunno, check my math? 1.4 billion transistors in 576mm^2 = 411mm^2 per billion transistors. 2.4 billion in 490mm^2 = 204mm^2 per billion transistors. 411/204 = 2.015 so the area scaling factor is just over 2. 65^2 / 40^2 = 2.64x shrink. Theoretically 45nm should have been enough (2.09x), but obviously there's some margin there. If it were perfect then 2.4 billion transistors should be 373mm^2... maybe Nvidia's being a bit conservative? ;)

Anyway, die size does not say jack about power requirements, heat production, performance, or max clockspeed. So there's still a lot of variables in the air. Until working silicon goes toe to toe vs Larabee I'm unwilling to predict a winner in performance or even performance per dollar.
© 2009 - 2011 Bright Side Of News*, All rights reserved.
© 2009 - 2013 Bright Side Of News*, All rights reserved.