Most of the GT212 specifications have been leaked.

ahh... switching to GDDR5 with a 256 bit bus. We'll have to see how far they can clock it.

These look... impressive... very impressive...

I'm with -RK on this. These do look impressive. However, why take I away the 512 Bit memory bus? Is it to keep the cards at a "respectable" price? Does it not offer any real world performance gain? Whatever the reason, I still think they should keep it. I know GDDR5 will help make up for the loss of it, as said in the article, but those memory clock speed's seem insanely high. A 512 Bit interface will help keep the performance edge and allow for awesome memory bandwidth (~250GB/sec?) and overclocking, without a 4500MHz memory clock. I certainly hope they don't cut back on the GT300 based cards.

Anyway that's just my opinion. I'm just trying to say that, once it's improved, why take a step backwards? Most people who would be buying these flagship cards would be people with either a hell of a lot of money or who want the very best, if not both, and usually don't care about the price. Take the 8800 Ultra for example, I saw that card for $900AU upwards when first released, and still people parted with their hard earn cash. Now you can get that kind of performance for $240. I'll probably never buy one of these GT212 based cards, but if I ever did, and having seen cards before with 512 Bit memory bus, I would want it to implement it. After all isn't that what flagship models are all about - the latest technology thrown together at a usually ridiculous price.

Although it's lifespan as flagship model does look rather short. Maybe it'll curve out a nice little niche in the market offering decent price to performance ratio. Having said that, maybe it's just a dummy, to see how it performs and what will need revising on the new process before the GT300.

Achtung Achtung Ze germans have got zee informasionz DAH!?

yeah RK thats what im wondering, a solid shader count bump... but half the total memory bus width and they still managed to go up 400 mil?

Well presumably if they can manufacture at less than 300mm^2 I could see the GTS version of that core hitting $250 or $300 price point pretty quickly, Nvidia would still have a reaonably beefy profit margine at that price and would kill anything I know of from AMD (including this rumored RV775XT).

I wonder if they can really reach under 300mm2
@MrWizard6600- as far as I know , the RV775XT specifications are BS .
I can't help but wonder what will ATi respond with .

Well as Radiator said "can they reach 300Nm2" while still hold to their monolithic architecture? I'd say the issues driving the 256-bit/GDDR5 move... price, space and cost to manufacture chips with 512. Once ATI changed to the idea of a smaller architecture and went with GDDR5... "Gee it Forced' Nvidia to re-think their strategies direction. They can no longer base the strategy around the extremes and cut it back for mainstream offering. GDDR5 is so cheap that it make no sense to build in 512 bus lanes (cost) and then not really saturate them, better to use the physical space on the PCB for cheap memory chips than under utilize lanes for the traffic. While this ulitmatly this provides fo smaller silicone, less binned chips, and more of them per wafer... Which ultimately isn't this what it's about? I don't profess to be an expert by any means, but this is the understanding as I've read, so correct me if I'm wrong.

@Naito; This isn't the first time we've seen a drop from the 512bit bus to the 256bit. ATi did this when they moved from the 2900XTX to the 3870. Simply put, the bus width is not only expensive to manufacture it's also not that usefull. The analogy I would have for you is like going from a full sized SUV to a tractor-trailer to pickup your weekly groceries. Technically speaking the tractor-trailer should be more effective right? But the SUV has the ability to move much faster. The same principal applies. What interests me is to see that both companies are making a departure from a more or less adopted standard of processing graphics. With the bump in TMUs it seems that nVidia will keep using them for other operations like bump mapping and such. In contrast ATi is using shader processors to perform these operations (as well as a few more typically from the rasters). It all makes for a very interesting difference in the cards.

The "monolithic die" myth really means nothing at all. Its just marketing fluff. Both brands scale nearly the same when combined with mutliple gpus. One is just a larger monolith than the other. And yes, they can build a big chip and cut back. Everyone does it, no one doesn´t do it. A 4830 is a cut back 4850 isnt it? A 4650 is a further cut back 4850. The same goes for GT200 to G92, GT200 is just G92 with a couple tweaks.