Reputation: 139
Has anyone run benchmarks of Deep Water and Driverless AI on various video cards? Will a TitanX offer real-world benefit on these tools?
Thanks,
Marc
Upvotes: 0
Views: 117
Reputation: 246
I don't know of any formal benchmarks. Deepwater uses TensorFlow and MXNet as deep learning engines, and there are benchmarks for those. In general you'll see significant speed-ups when convolution is involved.
The newer Nvidia cards are faster than old ones (no surprise there), and in particular they have more internal RAM, which determines how large mini-batches you can use for different problems, and how large problems you can work onsome . For example 1080 Ti has 11 GB of RAM while the older 1080 has 8 GB.
Titan X has 12 GB of RAM. It will give you significant speed-ups. However, it's more expensive than 1080 Ti, which is a much better value with almost as much memory.
Szilard Pafka has done benchmarks with different GPU cards. See https://github.com/szilard/benchm-dl
Upvotes: 1