vCPU Quota Requests

Question

Greetings from Vancouver Island.
I'm trying to work on building a custom llm for a hobby project. I have an ubuntu instance (g4dn.xlarge) that barely does training within a reasonable amount of time and ever suggestion I get for make this usable and not cost excessive is to bump this up to a 2xlarge and run it when needing to do the work and keep it off when not for cost control. Seems smart to me but the request to do this is unclear to me given it is a specific request to be allow to spin up the next size up, if you will, in the g4dn series.  Am I missing something on how to request access?  Any thoughts, suggestions, alternative machines, etc. would be helpful.

rick

re:Post Agent · Accepted Answer

Hello Rick from Vancouver Island! I understand your desire to improve the performance of your custom LLM training for your hobby project. You're on the right track with considering an upgrade from g4dn.xlarge to g4dn.2xlarge for better performance.

To request access to a larger instance type like g4dn.2xlarge, you need to submit a vCPU quota increase request. Here's what you should do:

1. Go to the AWS Service Quotas console.
2. Look for the EC2 service and find the quota for "Running On-Demand G and VT instances" in your desired region.
3. Request an increase to accommodate the vCPU count of the g4dn.2xlarge instance (which has 8 vCPUs).

Your approach of using a larger instance when needed and turning it off when not in use is a good strategy for cost control.

As for alternatives, you might also consider:

1. Using g5 instances, which offer newer GPU architecture and could potentially provide better performance for your LLM training.
2. Exploring the m7g.4xlarge instance type, which has shown good performance for some AI workloads, though it doesn't have a GPU.

Remember to benchmark your specific workload on different instance types to find the best balance of performance and cost for your project.

Lastly, make sure to use AWS Cost Explorer to monitor your spending and optimize your usage patterns as you experiment with different instance types.  
**Sources**  
[Run protein folding on AWS with Quantori | AWS HPC Blog](https://aws.amazon.com/blogs/hpc/run-protein-folding-on-aws-with-quantori/)  
[Specifications for Amazon EC2 accelerated computing instances - Amazon EC2](https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html)  
[Explore your AWS Costs and Usage with AWS Cost Explorer](https://aws.amazon.com/getting-started/hands-on/explore-your-costs-and-usage-with-aws-cost-explorer/)

vCPU Quota Requests

Add your answer

Relevant content