Add new mxnetlinux-gpu-g5 type nodes for testing Nvidia A10G GPUs.

[josephevans]

No description provided.

[DickJC123]

I support the idea of making Ampere GPUs available to the CI to continually validate MXNet's operation in the presence of Tensorfloat-32 (TF32) math. My approval is based on the intent of this PR, as I lack the jenkins experience to critique the added code lines.

I see that you've added mxnetlinux-gpu-g5, similar to the mxnetlinux-gpu-g4. There's also a restricted-mxnetlinux-gpu-g4, but you've not added a restricted-mxnetlinux-gpu-g5. How are restricted-* definitions used, i.e. what's the impact of not having a restricted-mxnetlinux-gpu-g5?

[josephevans]

I see that you've added mxnetlinux-gpu-g5, similar to the mxnetlinux-gpu-g4. There's also a restricted-mxnetlinux-gpu-g4, but you've not added a restricted-mxnetlinux-gpu-g5. How are restricted-* definitions used, i.e. what's the impact of not having a restricted-mxnetlinux-gpu-g5?

Thanks for the review. The restricted nodes are used for CD processes that require different permissions (such as publishing artifacts somewhere,) we shouldn't need a restricted node for this node type at this point. If it becomes needed down the road, we can add it.

[DickJC123]

Thanks for this additional capability. As I work to get MXNet CI using this new facility, is it also important to mention mxnetlinux-gpu-g5 in environment.yml, e.g. in lines like: https://github.com/apache/incubator-mxnet-ci/blob/master/services/jenkins-autoscaling/lambda_mxnet_ci/autoscaling/environment.yml#L15?