Thinking to use external GPU's in deep learning to train some CNN's. How far can such GPU's go with big dstasets?
I think of it as a quick solution for team members who may test their algorithms on their relatively-limited non-game machines.
Take a look at some benchmarks in the URL below:
https://www.notebookcheck.net/Razer-Core-Benchmarks-Analysis-and-Compatibility-Is-it-Worth-it.213526.0.html
It is not a surprise that such solutions have a sever performance degradation compared to desktop solutions, due to adding a new bottleneck (the connection to laptop). Also, the price is rather high ($500) just for the case. What I prefer personally is developing the algorithms on laptops (or other limited machines) and test the final code on a fully equipped machine.
Thank you Farshid for the feedback. I agree with you that the communication will be a big bottleneck. This is why I liked to hear the experience of people who have practiced it deeply.
- Badges
- Science topic