Looks like 28nm will be available by next year. http://www.xbitlabs.com/news/video/display/20100419141543_ATI_to_Build_28nm_Graphics_Chips_at_Globalfoundries.html Simon
Yes, but I'm pretty sure they don't want to give money to GloFo, seeing as AMD still has a stake in the venture.
From what I've heard the high-end parts of the next ATi generation called "Northern Isles" were supposed to be manufactured on a 32nm node at TSMC. That is until TSMC cancelled its plans for a 32nm node and replaced it with a 28nm node. After the 40nm debacle noone really believes TSMC will be able to pull off 28nm without problems. GlobalFoundries were always meant to handle the "Northern Isle" derivatives (lower-spec parts) at 28nm, but apparently because of the issues with TSMC "Northern Isles" has been more or less scrapped in favour of "Southern Isles", which is supposed to be a hybrid of "Northern Isles" and "Evergreen". "Southern Isles" will be manufactured on TSMC's 40nm node, which is "known". At some point ATi will use GlobalFoundries, probably with the "Southern Isles" derivatives and probably on a 28nm node, but I believe it'll be late 2011 rather than early 2011.
is ATI taking a tick-tock approach? Evergreen (Die shrink) --> Southern Isles (New architecture) --> a 28nm die shrink.
Yeah, ATI have been doing that for a while. They shrunk the die on the 4000 series halfway through its life to 40nm, they had yield problems with that initially but what they learned from that helped them with the 5000 series at the same process node. Nvidia have gone directly to a new achitecture and a new process with Fermi, which hasn't been ideal.
Just noticed that our illustrious friend Charlie over at Semiaccurate.com posted some news regarding Southern Islands a couple of days ago. Apparently Southern Islands has taped out already. Caveat emptor: The piece is written in Charlie's seminal colourful language.
Ha! As usual Charlie is predicting the end for Nvidia! I think if ATI can survive the 2xxx/3xxx generation then Nvidia aren't going to die just yet.
nVidia will be going very strong with 4xx series. it's all about marketing and now what nVidia has a product on shelves, the marketing machine should start rolling soon. remember Intel P4 vs AMD Athlon? marketing always wins for the general public.
I agree wuyanxu that marketing always wins for the general public. However, the general public won't be buying Fermi in its current manifestation. Marketing can't compensate for an absence of stock (there's none here in NZ) or a very high price (GTX470 is more expensive than an HD5870 in NZ, even if you could find a '470)(NZ is a small market but I can only go on my own experience). If priced appropriately then the Fermi derivatives could be competitive. At the moment ATi has a firm grasp of the mid- and low-ends of the graphics market. This is what the general public buy. Nvidia have a lot going wrong for them at the moment but it's far too early to predict their demise.
Actually ATi's midrange isn't that great as of right now. There's still a bit of breathing room for Nvidia. However when that's going to come is another issue.
But nVidia have NO new midrange. It's either Fermi on the high end, or 3 year old architecture recycled into a 250 GTWhatever-the-letter. Unless they can magically pull a non-room heater DX11 killer card for the price of a 5770...
I find it quite funny though, the GTS250 a 3 year old card is still holding the semi-low midrange decently. Of course you do have a good point, Nvidia has nothing in the mid-range right now, and as for me, I really want to see how well they scale when cut down.
