ConstReadonlyGPU_1X_A10: "GPU_1xA10"Single A10 GPU per node. Good for development and small workloads.
ReadonlyGPU_1X_H100: "GPU_1xH100"Single H100 GPU per node.
ReadonlyGPU_8X_H100: "GPU_8xH100"Eight H100 GPUs per node. Typical for distributed training.
Hardware accelerator type for the AiRuntime workload. Per-node accelerator count is encoded in the value name (e.g.
GPU_8xH100means 8 H100s per node).