Today I noticed a very interesting post at Chinese forums, which I would like to share with you. Please note that absolutely nothing has been confirmed and this post should be treated as editorial.
Many VC readers noticed that I didn’t post anything about recent GTX 880 revelations. That was because I already knew where it all came from, and it was pure speculation, which was even few months old. Of course I’m not saying that all of that was nonsense, but it was something every GPU enthusiast could came up with on his own.
Someome at Chiphell wrote a very brief post about the upcoming Maxwell GM204 GPU. It’s the same GPU that is believed to power soon-to-be-released GeForce GTX 880.
What really caught my attention was the part where he mentioned GM204-B1 stepping. According to his sources, the B1 stepping would power future GeForce 900 series, and it’s based on 20nm process, while the A1 and A2 steppings are reserved for GTX 800 series. I found this very interesting and it actually makes sense. The only part that still bothers me is where NVIDIA uses both GM1xx and GM2xx parts on the same node, instead of splitting the them to 28 and 20nm process respectively.
However it’s not that shocking as you might think. Let me remind you that this already happened before. The G73, G92 or G94, where all split into A and B,C stepping on different fabrication nodes.
The likelihood of launching Big Maxwell with GeForce 800 series is very low. In fact I believe we are going to see a very similar route NVIDIA took with 600 and 700 series. I made this simple chart just to visualize what we are talking about.
|GeForce 800 ==> GeForce 900|
|28nm GM204-A1:||GeForce GTX 880||==>||20nm GM204-B1:||GeForce GTX 970|
|probably TITAN||==>||20nm GM210-A2:||GeForce GTX 980 (Ti?)|
The ‘leaker’ also mentioned that GTX 970 is currently planned for the second half of 2015, which is pretty obvious. He also reported that GM204 (aka GTX 880) would be 10% faster than GTX 780 Ti and it would have more than twice of the 780 Ti’s frame buffer — 8GB. This would also mean it has 256b or 512b interface. Of course you all know that DX12 will be fully supported by this new card.
What’s even more intriguing, all these claims are also confirmed by Import/Export data from Zauba.
Whether all this data is true or not, I would still like to learn your point of view in the comments.