4 ms·
256, same as before. But they are Pascal, which, at 14nm, is 2x more efficient than 28nm Maxwell.
by mtgx 10y ago
256, same as before. But they are Pascal, which, at 14nm, is 2x more efficient than 28nm Maxwell.
- Symmetry 10y agoIt used to be the case that efficiency scaled reliably with feature size like that. Dennard scaling has broken down since the mid 2000s, however. https://en.wikipedia.org/wiki/Dennard_scaling https://en.wikipedia.org/wiki/Dennard_scaling