3 ms·
More like slightly slower but 10x cheaper. Honestly, using CPUs for inference just frees up more Nvidia GPUs to be used for training.
by wmf 3y ago
More like slightly slower but 10x cheaper. Honestly, using CPUs for inference just frees up more Nvidia GPUs to be used for training.
- yjftsjthsd-h 3y agoOops, read the numbers backwards; you're right, the GPU was still a bit faster. Still, yeah, if it's significantly cheaper that's still a huge win.