Information about available GPU resources
Google Cloud Platform
- Google Cloud Platform (GCP) is a service that provides you with specific machines for a specific amount of time to setup, run, and collect your experiments - https://cloud.google.com/
- The course has been granted coupons on GCP for $50 free credit in total for each student of this course. This should correspond to around 120 hours of using NVIDIA K80.
- Important note: GCP gives $300 free credit to all new customers! Make sure you sign up for it before you activate the coupon granted to you for this course - https://cloud.google.com/free/ (Links to an external site.)
- The coupons you get specifically for this course should be enough for course-related work, but if you activate them first you will no longer be a new customer and you will not be eligible for $300 free credit
- The coupons you get specifically for this course should be enough for course-related work, but if you activate them first you will no longer be a new customer and you will not be eligible for $300 free credit
- Please check out Sections 1-3 of : Google Cloud Setup and Tutorial - cs231n
Links to an external site. before you redeem your course GCP coupon.
- There is one coupon per student on the course. To redeem your coupon please follow the instructions here: How to redeem my GCP coupon
- To get GPU access on GCP please read Section 4 '"Request an Increase in GPU Quota"
Links to an external site. of the tutorial: Google Cloud Setup and Tutorial - cs231n
Links to an external site.
You also need to set up and access a Deep Learning Virtual machine. This is described in sections 5 and 6 of Google Cloud Setup and Tutorial - cs231n Links to an external site. where the solution is based on gcloud. We describe an alternate route without gcloud via our tutorial DD2424 Tutorial on Google Cloud Platform Download DD2424 Tutorial on Google Cloud Platform - Additionally, Google Colab provides free GPU (or TPU!) access for a limited time - https://colab.research.google.com
Additional tips
Keep jobs running after disconnecting from SSH
All processes started inside an SSH session will be terminated when the session is closed, either through disconnecting or losing internet access (SIGHUP (Links to an external site.)). To keep your processes running you can use:
- tmux (Links to an external site.)
- screen (Links to an external site.)
- nohup (Links to an external site.) (the least flexible)
Managing data (downloading and copying)
You can download data or datasets into your virtual machine using:
- wget url - when downloading from the web
- rsync or scp - to send from/to your local machine or between different VMs