While Charlie Demerjian is not the most impartial source, he's broken various significant stories about Nvidia shortcomings in the past - most notably, Bumpgate and the Fermi wood screw fiasco. This article gives clear and specific reasons to think that Pascal isn't nearly as far along as some Nvidia boosters seem to believe.
Let's keep in mind that Nvidia has
never beaten ATI/AMD to a new node in the past. That's not to say it can't happen, just that it's not the way to bet. ATI's first 40nm card, the HD 4770, was
released in late April 2009, while Nvidia didn't release 40nm Fermi products
until late March 2010. That's a full
11-month lead for ATI, despite the fact that both companies were using the exact same TSMC process. Aside from Fermi architectural challenges, ATI's lead was also due to the fact that they started near the bottom, with the 137mm^2 RV740 chip, while Nvidia insisted on debuting their big-die GF100 (529mm^2) first.
Nvidia learned their lessons from that one; Kepler was a refinement to Fermi rather than a whole new architecture, and they started off with the smaller GK104 chip (294 mm^2), saving the big die for later. That enabled them to be only 3 months behind AMD on 28nm. But could they be repeating their past mistakes this generation? We know that AMD's Polaris 10 chip will be a small die (probably 100-140 mm^2), and this has already been demonstrated in public, so it could come to market very soon - perhaps even in the March refresh of the MacBook Pro. Meanwhile, most of the rumors around Nvidia are focused on "Big Pascal" (presumably GP100), with virtually nothing known about GP104 or GP107 even though these would be easier to manufacture and higher in volume. They could still pull out a surprise, but it's not looking likely at this point. I think they want to coast on Maxwell in user space for most of 2016, and are obsessed with bringing out Big Pascal to fight off Intel's advances into the HTPC market.
I now firmly believe that AMD will hold the
technical lead in GPUs for the majority of this year. Whether they will be able to translate that into an increase in profits and market share is not a question I'm equipped to answer. AMD
could come out with a clearly superior product and still get crushed by Nvidia's advertising might and inertia. But I think the smart gamers will be buying AMD Polaris cards in 2016.
Facebook
Twitter