Staring at the early morning GTC conference users may be slightly disappointed, the new gaming graphics really did not open in the AI, deep learning graphics summit. Of course, NVIDIA CEO has launched a new nuclear bomb, Based on GV100 core Tesla V100 new products and Quadro GV100.
Tesla V100 was actually released at the 2017 GTC Conference, but this time , upgraded to 32GB HBM2 memory.
The picture shows the 16GB Tesla V100
To say that, thanks to the suppliers Samsung and SK Hynix for upgrading, the capacity of a single HBM2 has been increased from 4GB to 8GB, which brings the capacity to double when the area does not change, in other words, simply replace it.
Of course, NV's skill is also there, that is, the frequency / thermal design power consumption, etc. have not increased, remain consistent, want to upgrade, then you can fool to replace (with or without TM?).
Starting from the second quarter , NV will withdraw 16GB shipments, the main push 32GB models. It is worth mentioning that, 3 years ago, AMD will ship 32GB GDDR5 memory FirePro S9170, this count NV and AMD tie? (self-brained).
The second thing to mention is the Quadro professional card, The price of GV100 is 9000 US dollars, 3 times TITAN V (State Bank 22999 yuan) .
Quadro is still aimed at workers, professional graphics workers, the new GV100 has 5120 CUDA cores, 32GB HBM2 memory, can provide up to 7.4 trillion double-precision floating-point performance per second and 14.8 trillion per second The performance of single-precision floating-point operations and deep learning floating point performance of 118.5 teraflops per second.
The Quadro GV100 supports NVlink 2.0, but also does not support RTX ray-tracing technology. It claims to bring movie quality to the game.
