Add AWS Graviton with Nvidia : G5g

[gmarchand]

• Implement the dockerfile
• Implement Compute Env, Queue, Jobs, API

https://towardsaws.com/cpu-vs-gpu-for-video-transcoding-challenging-the-cost-speed-myth-b5e0726ce3e8

[gmarchand]

FYI : https://docs.aws.amazon.com/batch/latest/userguide/gpu-jobs.html

Only instance types that support a NVIDIA GPU and use an x86_64 architecture are supported for GPU jobs in AWS Batch. For example, the G4ad and G5g instance families aren't supported.

Please contact AWS Support, if you need G5g with AWS Batch

[msencenb]

@gmarchand Does this mean it is possible to get G5G instances with AWS Batch by contacting support? Or does this mean it's not available and we should contact support to tell them we'd like that as a feature in the future?

[gmarchand]

Hello Please, contact support to indicate you would like this feature in AWS Batch

[msencenb]

Unfortunately I don't have a support plan that lets me contact AWS Support (I'm on the Basic tier). If it's helpful internally, feel free to reference this github issue though. I would like to use the G5g instances with AWS Batch for ffmpeg processing.

[aws-joelherd]

Hi, g5g can be used with Batch if the customer brings their own AMI. Batch does not vend a default ARM/Nvidia AMI today.

[aws-joelherd]

Will verify the expected behavior and follow up here.

---

Update 5/9: Yes, gpus can be accessed by code/config if the customer brings their own AMI or uses the Batch-default AMI for g5g. Will work with the docs team to get our public docs updated.

[aws-joelherd]

Question in re:Post: https://repost.aws/en/questions/QU_QGNAAUpREWWrZBdsHLNcw