Originally Posted by iLeakStuff
GP100 is for Tesla and NVlink and have HBM2 because they are very expensive and because interlinks and NVlink is required here.
GP102 is for consumer cards like Geforce. Comes with GDDR5X to keep price down and to have enough supply for the massive interest.
Why do you think OP say the Titan card is long?
Because it have GDDR5X which requires a lot of PCB space
I suggest you stop leaking nonsense until a valid proof (like leaked picture) is out.
All this Tesla fail chip need to go somewhere, and has to go fast. The only chip that is available enough for the masses by SEPT release is the same chip that is used in Tesla P100 now.
G5X variant of GP100 (if it really does exist) is save for next year GTX 1100s. If not what happen when next year comes? Obviously this give Titan the edge since its on HBM and whatever the GTX 1180 (or even the Ti variant) churn out, it will never be as the same level as the HBM Titans. Even if its on 3840 cores. This is a good move as the 1180 can be priced reasonably well, like $700.
At the same time it also open up a whole new range of pricing for the Titans, which can be between $1000-$1500.
No matter what, I am gearing toward a chip that actually exists
, than some chip that is being imagined by some chiphell poster who has too much time on hand.Edited by guttheslayer - 7/8/16 at 5:23pm