2 ms·
These GPUs are more than twice as fast as Amazon's GPUs. They also have 4x the memory and there is probably more GPU to GPU bandwidth as well. Doing deep learni
by modeless 11y ago
These GPUs are more than twice as fast as Amazon's GPUs. They also have 4x the memory and there is probably more GPU to GPU bandwidth as well. Doing deep learning on clusters is not practical yet due to bandwidth issues. If you want to train state-of-the-art models you need the biggest single machine you can get, and this is it.