|
|
Post by Fidite Nemini on Dec 1, 2015 12:30:15 GMT -6
I wouldn't put down that much money with the new GPU generation just around the corner. 16/14nm vs 28nm, HBM2 vs GDDR5 (or GDDR5X vs GDDR5 in the lower tier GPUs) plus the usual architectural improvements.
It's going to be a genuine generation switch, not like just changing the architecture and some refinements on the same manufacturing process.
Even without thinking optimistic, having the manufacturing node cut down nearly to half of what it is now, Nvidia and AMD could put as much as twice the streaming processors/CUDAs, etc on a similarily sized GPU die, which isn't unlikely since with space saving HBM2 replacing GDDR5, putting big dies on the PCB is easy as they no longer have to share the space with loads of VRAM modules (although lack of manufacturing optimization means we won't see as big and complex chips right at the start). Couple that with the usual architectural performance improvements, a new x80 GPU could be around 50% faster than a 980 for a comparable price (plus more energy efficient).
The previous generation switch as example:
GTX 580 was 40nm manufacturing process, Fermi architecture, 512 CUDAs (forsaking TMUs and ROPs for simplicity) on a 529mm GPU die. It was around 15% faster than the previous GTX 480 (40nm/Fermi/520mm/480 CUDAs).
The GTX 680 was 28nm manufacturing node, Kepler architecture, 1536 CUDAs (thrice as much!) on a 294mm GPU die (almost half the die size!). Performance gains range from 20-30% during its introduction and 30-50% after driver optimizations.
Both retailed at the same price.
Of course no such huge jump in performance is guaranteed, but if it happened and you just put down 500 bucks you'd gnash your teeth until your jugular breaks into pieces.
|
|
