3 ms·
There are enough companies out there with deep pockets that want to do some ML. They'll pay pay those prices, no questions asked.
by dx034 9y ago
There are enough companies out there with deep pockets that want to do some ML. They'll pay pay those prices, no questions asked.
- IanCal 9y agoPer the marketing material it's up to a PFLOP of mixed precision (is that the same as just saying "half precision"? or is it 8 bit?) for $25/hour. I can easily see people paying full price for that. Still, spot price is currently $2.40.
- Beltiras 9y agoIt must be a misplaced comma. 15.7 single precision can never translate to 125 mixed.
- dharma1 9y agoThe reason Nvidia quote 120 TLFOPS mixed precision on V100 is because of the new tensor core. https://devblogs.nvidia.com/parallelforall/inside-volta/ https://devblogs.nvidia.com/parallelforall/inside-volta/
- julien_c 9y agoAlso all startups with (basically) unlimited aws credits