The intent of this article is to provide guidance in using Google CoLab Enterprise for students and researchers.
Colab Enterprise is a collaborative notebook environment that's managed and offers the security and compliance capabilities of Google Cloud. It's a Vertex AI tool that combines the collaborative features of Colaboratory with Google Cloud's security and compliance capabilities. Colab Enterprise is intended for data analysts, data scientists, and data developers who work with notebook environments.
The Google Cloud CoLab Enterprise environment is a UMD recommended fee based option instead of Google Workspace subscription solutions CoLab Pro and CoLab Pro+. UMD environment does not support credit card payments for Google Cloud Services. UMD CoLab Enterprise solution offers compatible features and instances to the subscription versions. Like the subscription option, CoLab Enterprise has costs associated with it. Contact DIT for more information.
Here’s a comparison of the specifications and offerings for Colab, Colab Pro, and UMD Colab Enterprise. For more information, see the Colab GPUs Features & Pricing article.
| Feature | Google Colab | Google Colab Pro | UMD Colab Enterprise |
|---|---|---|---|
| GPUs | T4 | K80, P100, T4, A100 | T4, V100,A100,* L4* |
| CPUs | 2xvCPU | 2xvCPU | 4xvCPU to more |
| RAM | 12GB | 32GB | 16GB - 60GB |
| Price | Free | 9.99/month for 100 compute units (~8-10 hrs total use per month) | Dependent on usage |
| Datasets/Notebooks Upload/Download locations | Local Drive, Google Drive, URL | Local Drive, Google Drive, URL | Local Drive, Google Cloud Storage Buckets, URL |
| Run/Code Execution Time | Up to 12 hours | Dependent on compute units usage | Dependent on user and data requirements |
| Access Control | Managed through Google Drive sharing credentials | Subscription managed | Managed through IAM UMD (user id and password) |
| Support | Blogs, Internet | Standard Support provided for each project | |
| Limitations | System time outs due to reaching max GPU use | Session time out when compute units needs have exceeded | Can’t mount google drive within enterprise |
| T4 is cheapest GPU | Not approved for use within UMD Domain | The file size of uploaded notebooks is limited to approximately 20 MB. |
* Denotes specialized machine type required for high performance GPU.
Google offers a number of virtual machines (VMs) that provide graphical processing units (GPUs), including the NVIDIA Tesla K80, P4, T4, P100, and V100.
You can use NVIDIA GPUs on GCP for large scale cloud deep learning projects, analytics, physical object simulation, video transcoding, and molecular modeling. GCP also provides virtual NVIDIA GRID workstations, which can let an organization’s employees run graphics-intensive workloads remotely.
Google Cloud’s Colab Enterprise utilizes 8 regions within the United States. The Colab enterprise service also limits the virtual machine types for GPU use to the N1, A2, G2 series.
| Regions | GPU | |||
|---|---|---|---|---|
| South Carolina - us-east1 | N1-Tesla T4 | N1-Tesla V100 | A2-Tesla V100 | G2- NVIDIA-L4 |
| Northern Va. - us-east4 | N1-Tesla T4 | G2- NVIDIA-L4 | A2-Tesla V100 | |
| Iowa - us-central1 | N1-Tesla T4 | N1-Tesla V100 | A2-Tesla V100 | G2- NVIDIA-L4 |
| Oregon - us-west1 | N1-Tesla T4 | N1-Tesla V100 | G2- NVIDIA-L4 | |
| Los Angeles - us-west2 | N1-Tesla T4 | N1-Tesla V100 | ||
| Los Angeles - us-west4 | N1-Tesla T4 | N1-Tesla V100 | A2-Tesla V100 | G2- NVIDIA-L4 |
Regions for GPU use. The listed GPUs T4, V100, L4 and A100 are found in all three regions.
The below Virtual Machine types coincide with the listed regions for supported GPUs.
During our assessment we were able to assess the cost to utilize a Colab enterprise project with a proposed configuration.
| GPU | Computer units/hr | Small models approx max runtime hours/month | Google GPU cost/hr |
|---|---|---|---|
| T4 | 1.92 | 52 | $.35 per GPU |
| L4 | 4.92 | 20 | $2.48 per GPU |
| V100 | 5 | 20 | |
| A100 | 15 | 7 |
| Google service | CoLab Ent (no assigned runtime) | CoLab Ent runtime (idle) (single runtime) | Idle CoLab Ent (3 or more runtimes) | CoLab Ent with T4 idle runtime | CoLab Ent runtime with T4 (single GPU) large dataset |
|---|---|---|---|---|---|
| Vertex AI | 0 | $1.47/day | $4.41/day | $5.15/day | $13.87/use |
| Compute Engine | 0 | $.011 | 0 | 0 | $.04 |
| Support | 0 | $.044 | $.13 | $.15 | $1.4 |
| Cloud Storage | 0 | 0 | 0 | 0 | $1.69 |
| Total per use | 0 | $1.5/day | $5/day | $17/use |
The costs assessment from the table below that at first look using the subscription version will be the most cost effective solution. However, for the subscription to be the alternative the use cases must fall specifically within the cost variables. For example the use case utilizing a single GPU will allow a maximum of 52 hours when using the T4. This is ideal for small models generally needed for teaching and learning. However as the use case gets more complex, the subscription scenario is not as cost effective and Colab Enterprise is a more efficient solution, flexible and comparable in cost with the subscription versions.