I’m currently performing computations that could benefit from GPU acceleration, specifically using CUDA for tensor multiplication. I’m looking for a solution in AWS that allows me to invoke instances with GPUs for faster computations.
Can anyone suggest the best way to approach this on AWS?
Hey @neha.jlly! Use AWS EC2 GPU Instances (p3, p4).
AWS provides several types of EC2 instances equipped with GPUs specifically designed for machine learning and computational tasks. The p3 and p4 instance families are ideal for leveraging CUDA acceleration, as they feature NVIDIA Tesla V100 and A100 GPUs, respectively.
In my opinion, you can invoke these instances via AWS Lambda in combination with EC2, or directly configure your EC2 instances based on your computational requirements.
These instances are optimized for high-performance computing (HPC), including deep learning tasks like tensor multiplication and large-scale matrix operations.
They also provide good integration with popular frameworks like TensorFlow and PyTorch. If you’re testing or running experiments in the cloud, using EC2 instances with GPUs can significantly boost the performance of your computations.
I hope I was able to help you 
I see @vindhya.rddy has already provided an answer, but I will let you know about my approach to the same. Simply, we leverage AWS Lambda with GPU Instances.
AWS Lambda typically runs in CPU-based environments, but you can trigger GPU-powered EC2 instances (such as p3 or p4) indirectly using AWS SDK or Lambda functions integrated with Step Functions.
When you need to run computations requiring CUDA acceleration, you can set up an EC2 instance to spin up automatically and execute your task.
This setup offers scalability, flexibility, and on-demand GPU usage, so you only pay for compute power when necessary.
If you’re using testing tools like LambdaTest for automated cross-browser testing, this approach is also great for integrating computational tests that require GPU acceleration while managing your resources effectively.
To learn more about how GPU can accelerate your web performance, check out this blog:
Now you’re good to go with the right ammunitions for your endeavours. Bye bye! 
Hey @neha.jlly! You can use any of the methods mentioned by @vindhya.rddy and @richaaroy, but I will also suggest you the one I use.
AWS SageMaker for Machine Learning Tasks
If your computations are related to machine learning, AWS SageMaker might be a suitable option. SageMaker offers managed Jupyter notebooks and training environments that support GPU instances (like p3 and p4) for heavy computation.
It simplifies the process by abstracting much of the complexity of setting up environments for deep learning or data-intensive tasks. SageMaker also integrates with other AWS services, making it easier to scale your operations.
For example, you can run training jobs on GPU-powered instances and use built-in frameworks like TensorFlow or MXNet.
SageMaker provides a more streamlined, managed approach compared to manually configuring EC2 instances, making it an excellent option for those who want to focus on development rather than managing infrastructure.
That’s it! Happy to help 