Yeah, but the Tesla parts have a trend of sharing their shader count, memory bus width and shader clockspeeds with their GeForce counterparts.
http://www.nvidia.com/object/product_tesla_c1060_us.html <- for example, shares those listed traits with the GTX 280.
http://www.nvidia.com/docs/IO/43395/C870-BoardSpec_BD-03399-001_v04.pdf <- and again with the G80 based C870 and the 8800 GTX.
Whilst it's not a 100% cert, it's probably more accurate in regards to the geforce part's specifications than certain rumours floating about the 'net.
What?
I don't know what he's smoking, memory has NOTHING to do with core size or disabling parts of the core. We know as a fact the cores being made are 512sp parts, if they are disabling bits of the core to hit power/yield requirements, thats what they are doing. They would make and sell full 512sp parts IF THEY COULD. They probably can get one or two parts, but they can't sell it in volume. The gpu market will take a 380gtx ultra mega uber 512sp part that has 10 available in retail ever. The Tesla market will not, if they can't produce what they release they'll lose all creditability, that means they can not make 512sp parts in any usable number for even a small scale Tesla product, let alone a high sales gaming version.
As I hinted at with one of the other rumours of insanely cut down Fermi's with only half the SP's working(unheard of before from gpu makers), it was an indication, aswell as being 6 months late and still having awful yield problems, that the top end part was always going to be a farce in terms of numbers.
They will either go 380gtx with less shaders for a semi decent numbers in availability, or make the 380gtx a full shader part, release the few working for reviews and make almost none available for sale.
As for the specs, 1.4Ghz isn't good, its early specs in terms of floppage and raw power, were based on estimates of 1.7Ghz shader speed, also were based on 750Mhz core clocks, considering shader clocks have barely advanced, the core clock which was only 500Mhz before could be well short of the 750Mhz which would also hamper it quite badly.
Its just crappy news all around, if we were all hoping for 448sp 360gtx's at or around the 5870 price, well, we're smegged. The high end 448sp part will beat the 5870 by much less than expected, and need to be priced high to make money back. A lower SP 360gtx will still be costing a lot and could well not beat a 5870 depending on where the SP count comes in, and it will probably still cost more.
IF Nvidia really do release a 350gtx with only 256shaders, it will cost a bomb compared to a mid range variant that only ever had 256shaders. It also means Nvidia's midrange if it ever comes will most likely have to be decent amount lower clocked or less shaders than its 256shader part, or it won't be able to sell those salvaged Fermi's. Meaning they might be forced to cut their midrange performance.
Every indication for over 6 months has been this ridiculous sized core can't be made effectively on a crap process at TSMC. It just can't compete on cost, ever and I really don't know how long Nvidia can keep selling everything they make at cost(or a loss even), before they get into real trouble.
