2 ms·

Nvidia Unifies AI Compute with “Ampere” GPU

• mark_l_watson 6y ago
  "On single precision floating point (FP32) machine learning training and eight-bit integer (INT8) machine learning inference, the performance jump from Volta to Ampere is an astounding 20X." Incredible. I just have a 1070 in a Linux laptop which really helped my personal deep learning projects (before I retired, I had I the compute I needed). No way to justify the purchase, but I would like an Ampere based system at home. My workload has changed a lot since retiring, I train less and use pre-trained NLP models.
  • votepaunchy 6y ago
    > No way to justify the purchase, but I would like an Ampere based system at home. These will be available on every major cloud, on-demand. Is that not sufficient, to have a $20K GPU available in seconds for a dollar or so per hour?