Click to view our Accessibility Statement or contact us with accessibility-related questions
Showing 1 of 4 conversations about:
ihearttech
1
Jun 4, 2018
bookmark_border
This is confusing because some of these are "deep learning cards" and some are gaming cards. It would make sense to have a separate poll for gaming cards.
Citronnade
1
Nov 21, 2018
bookmark_border
ihearttechThis is from a long time ago, but for anybody interested: the "gaming" cards are in fact the best ones for deep learning. Nvidia workstation cards (K40, Quadros, etc.) focus on strong FP64 performance but deep learning primarily depends on FP32 and FP16 performance. Workstation cards will also have strong FP32 and FP16 performance, but incredibly inefficiently compared to a "gaming" card that doesn't need FP64.
The best single card here is probably the Titan V, but the 1080 ti is more cost efficient, particularly if you run multiple in parallel. Titan Xp is good as well. These cards are very popular in deep learning research labs; they used be used in large datacenters as well but nvidia banned the use of them in that use case because it was taking market share away from their more expensive workstation cards.
The newer generation of RTX cards changes things a bit with their tensor cores (the RTX 2080 ti is about 33% faster than the 1080 ti on resnet-50, for example), but the 1080 ti for now is still more cost efficient (~$700 USD vs. $1200 USD).
Drop Refurbished
Like-new products you can trustDrop Rewards
Get $5 for every 500 points you earn! Learn more
Drop Keyboard Club
Become a member and expand your keycap collectionCollaborate With Us
For Brands & DesignersFollow Drop