GPU Worker Node AMD

High-efficiency Kubernetes compute with AMD GPU

VPNaaS – Client to Site
Request Callback

Overview

Boost your Kubernetes workloads with AMD GPU Worker Node that seamlessly extends your existing K8s clusters. These nodes deliver up to 81.7 TFLOPS (FP64), 163.4 TFLOPS (FP32), 2,614 TFLOPS (FP16), and 5,229 TFLOPS (FP8) with 192 GB HBM3 memory.

Engineered for compute-intensive HPC simulations and AI training at scale, AMD GPU worker nodes integrate with Kubernetes for resource scheduling and ROCm-optimised pod deployment. Run compute-intensive simulations, train deep learning models, and render complex visual workloads at speed – while ensuring operational consistency across your cloud-native environment.

Pricing

To know more about the SKUs and pricing click below.

Core Features at a Glance 

Native kubernetes integration
Seamlessly add AMD GPU compute to clusters with automatic node registration, GPU resource advertisement through AMD device plugin, and ROCm runtime support.
Pod-level GPU scheduling
Allocate specific GPU resources via standard Kubernetes specifications with ROCm-optimized containers.
Extreme model capacity
Train and run large LLM/ML models with 192 GB HBM3 memory per GPU and bandwidth up to 5.3 TB/s, supporting extended context lengths and memory-intensive AI workloads.
Optimized precision modes
Support FP8, BF16, FP16, INT8, plus full FP32 and FP64 for efficient AI training, inference and HPC workloads, reduced memory usage, and maximum flexibility.
Container-native deployment
Deploy GPU workloads with automatic ROCm drivers, HIP libraries, and AMD GPU Operator integration.
Multi-GPU pod support
Scale pods across 1, 2, 4, or 8 GPUs with high-speed interconnects for distributed training.
Cluster autoscaling
Scale AMD GPU worker nodes automatically based on GPU resource demand, ensuring optimal resource utilization and cost efficiency with intelligent workload placement.
Framework-ready environment
Run AMD ROCm, along with PyTorch, TensorFlow, and other AI/ML frameworks optimized for AMD architecture, ready for immediate container deployment.

What You Get

Kubernetes-native GPU nodes
Cutting-edge AMD MI300X worker nodes with automatic registration, tainting, and resource advertisement.
Pre-configured GPU runtime
AMD device plugin, ROCm drivers, and HIP libraries automatically configured on each worker node.
Enterprise-grade orchestration
Native Kubernetes HA, pod security, network isolation, and persistent storage integration to run production GPU workloads
Seamless cluster integration
Support for mixed CPU/GPU clusters with intelligent workload placement.
Monitoring and observability
GPU utilization metrics through Kubernetes monitoring stack, resource quotas, and performance dashboards, with API access for automation.

Still have questions?

AMD GPU worker nodes automatically join existing clusters via kubeadm/kubelet processes, with the AMD device plugin exposing GPU resources through the Kubernetes API for scheduling.
MI300X nodes excel at LLM inference pods, distributed training jobs, HPC simulation workloads, and ROCm-based applications that require high memory capacity and bandwidth for memory-intensive AI tasks.
MI300X worker nodes offers 192 GB HBM3 memory per GPU with ~5.3 TB/s bandwidth, enabling large language models and memory-bound AI tasks accessible through standard Kubernetes resource limits.
Yes, AMD GPU worker nodes are automatically tainted with amd.com/gpu:NoSchedule to prevent non-GPU workloads from scheduling, and support standard Kubernetes scheduling
Pods request AMD GPUs using amd.com/gpu resource limits, with the Kubernetes scheduler placing them on available nodes.
Individual pods can request 1, 2, 4, or 8 AMD GPUs depending on workload requirements, with automatic scheduling to worker nodes with sufficient GPU resources.
All ROCm-compatible frameworks work through containerized deployments with standard Kubernetes manifests, supporting PyTorch, TensorFlow, and HIP-based applications with ROCm driver access.
Yes, we provide ready-to-deploy container images with pre-installed ROCm drivers, HIP libraries, and ML frameworks for seamless pod deployment on AMD GPU worker nodes.

Ready to Build Smarter Experiences?

Please provide the necessary information to receive additional assistance.
image
Captcha
By selecting ‘Submit', you authorise Jio Platforms Limited to store your contact details for further communication.
Submit
Cancel