Non-business CPU credits

Question

I have a long-running CPU intensive Linux/C program I wrote for my personal recreation. I launched five t2.micro instances trying to parallelize the process, but the CPU credits maxed out, and the process could not complete.  I'm wondering 1) if the c5.large instance type would work for my use case and 2) would I qualify for reduced pricing or fixed pricing?  I need five instances with fixed CPU limits and 1GiB Memory.  I need to run them for about 600 hours.  When the process is finished I would delete the instances. As a retired computer programmer, my budget is $75 USD. My program is in a private github repository which I can share with you if you're interested.  (When this one is finished I have two more long processes to run that I estimate will take 600 hours each.)

re:Post Agent · Accepted Answer

Based on your description of a CPU-intensive workload that depleted T2 micro instance CPU credits, a C5 large instance would likely be more suitable for your use case. Unlike T2 instances which use a burstable performance model with CPU credits, C5 instances provide fixed CPU performance designed for compute-intensive workloads.

The C5.large instance offers 2 vCPUs and 4 GiB of memory, which exceeds your stated requirement of 1 GiB memory. This would give you consistent CPU performance without the throttling you experienced with T2 instances.

Regarding pricing options, AWS doesn't offer special reduced pricing based on personal or non-business use. However, there are several pricing models that might help you stay within your $75 budget:

1. Spot Instances: These can provide significant discounts (up to 90% off On-Demand prices) but come with the risk of interruption if AWS needs the capacity back.

2. Reserved Instances: While these offer discounts for longer commitments, they typically require upfront payment and at least a 1-year term, which may not align with your temporary 600-hour workload.

3. Savings Plans: Similar to Reserved Instances but with more flexibility, though still requiring longer-term commitments.

For your specific use case of running 5 instances for 600 hours each, I'd recommend exploring Spot Instances as they would likely provide the most cost-effective solution for your temporary computational needs. You could also consider using fewer, more powerful instances to reduce the overall cost while still completing your workload.

I don't need to review your GitHub repository as the decision would be based on your computational requirements rather than the specific code.  
**Sources**  
[Amazon SageMaker AI distributed data parallelism library FAQ - Amazon SageMaker AI](https://docs.aws.amazon.com/sagemaker/latest/dg/data-parallel-faq.html)

Non-business CPU credits

Relevant content