Requirements

Requirement	Metric	Criteria	Constraint
The compute resource should be cheap or free to use.	Cost (CAD)	Lower is better	No more than ???
The compute resource should have enough RAM/VRAM to train the model.	Gigabytes (GB) of RAM/VRAM	More is better	At least ???
The compute resource should have enough cores to train the model.	Number of cores	More is better	At least ???
The compute resource should have enough storage to contain the dataset.	Gigabytes of Storage (GB)	More is better	At least ???
The compute resource should be easy to use
The compute resource should be easy for everyone to access

Options

UTAT Remote Desktop: GPU is old - probably not powerful enough
ECF:
- Inquire at <ecfhelp at ecf.utoronto.ca>
- Work station are: OptiPlex 7060 Desktop spec sheet | Dell USA
- Could contain a good chip - will inquire
CompSci compute?:
- @Rishit Dagli
GitHub:
- “You can request access to a GPU-powered codespace if you need a more powerful machine. This means a machine learning engineer can use an iPad or Chromebook to perform data-heavy deep learning computations via GitHub Codespaces.”
Getting started with GitHub Codespaces for machine learning - GitHub Docs
Colab:
- https://colab.research.google.com/signup?utm_source=notebook_settings&utm_medium=link&utm_campaign=premium_gpu_selector
Kaggle Code:
- Kaggle: Your Machine Learning and Data Science Community
- Efficient GPU Usage Tips and Tricks | Kaggle
- Free access to P100s
Azure ML
- Azure Machine Learning - ML as a Service | Microsoft Azure
- Migration Guide for GPU Compute Workloads in Azure - Azure Virtual Machines | Microsoft Learn
Amazon SageMaker
- Machine Learning Service - Amazon SageMaker - AWS
- [Deep Learning on GPU Instances (amazon.com)](https://aws.amazon.com/machine-learning/accelerate-machine-learning-P3/#:~:text=You can use Amazon SageMaker to easily train,by reducing training times from days to minutes.)

Assessment

Compute Resource	Cost	Chip(s)	RAM/VRAM	Cores	Storage	Ease of Use
ECF	Free	???				Access through VM, would need ECF login
CompSci	Free(?)	???
GitHub Codespaces	$2.88/hr		128 GB	32 cores		Already integrated with GitHub - easy
Google Colab	$13.99/month, buy additional compute at $13.99 for 100 Compute Units	A100 GPU or V100 GPU	80 GB or 32 GB	640 Tensor Cores, 5,120 CUDA Cores		Would need to move code to Colab, not sure if multiple people can use it
Kaggle Code	Free	P100 GPU	16 GB	3584 CUDA Cores		Would need to move code to Kaggle, but everyone can use it

Recommendation

Investigate GitHub Codespaces and verify compute is sufficient
Otherwise look into Colab Pro