Hardware accelerators
Hardware accelerators
Abstract
Chapter 1. About hardware accelerators
Specialized hardware accelerators play a key role in the emerging generative artificial intelligence and machine learning (AI/ML) industry. Specifically, hardware accelerators are essential to the training and serving of large language and other foundational models that power this new technology. Data scientists, data engineers, ML engineers, and developers can take advantage of the specialized hardware acceleration for data-intensive transformations and model development and serving. Much of that ecosystem is open source, with a number of contributing partners and open source foundations.
Red Hat OpenShift Container Platform provides support for cards and peripheral hardware that add processing units that comprise hardware accelerators:
- Graphical processing units (GPUs)
- Neural processing units (NPUs)
- Application-specific integrated circuits (ASICs)
- Data processing units (DPUs)
Specialized hardware accelerators provide a rich set of benefits for AI/ML development:
- One platform for all
- A collaborative environment for developers, data engineers, data scientists, and DevOps
- Extended capabilities with Operators
- Operators allow for bringing AI/ML capabilities to OpenShift Container Platform
- Hybrid-cloud support
- On-premise support for model development, delivery, and deployment
- Support for AI/ML workloads
- Model testing, iteration, integration, promotion, and serving into production as services
Red Hat provides an optimized platform to enable these specialized hardware accelerators in Red Hat Enterprise Linux (RHEL) and OpenShift Container Platform platforms at the Linux (kernel and userspace) and Kubernetes layers. To do this, Red Hat combines the proven capabilities of Red Hat OpenShift AI and Red Hat OpenShift Container Platform in a single enterprise-ready AI application platform.
Hardware Operators use the operating framework of a Kubernetes cluster to enable the required accelerator resources. You can also deploy the provided device plugin manually or as a daemon set. This plugin registers the GPU in the cluster.
Certain specialized hardware accelerators are designed to work within disconnected environments where a secure environment must be maintained for development and testing.
1.1. Hardware accelerators
Red Hat OpenShift Container Platform enables the following hardware accelerators:
- NVIDIA GPU
- AMD Instinct® GPU
- Intel® Gaudi®
Chapter 2. NVIDIA GPU architecture
NVIDIA supports the use of graphics processing unit (GPU) resources on OpenShift Container Platform. OpenShift Container Platform is a security-focused and hardened Kubernetes platform developed and supported by Red Hat for deploying and managing Kubernetes clusters at scale. OpenShift Container Platform includes enhancements to Kubernetes so that users can easily configure and use NVIDIA GPU resources to accelerate workloads.
The NVIDIA GPU Operator uses the Operator framework within OpenShift Container Platform to manage the full lifecycle of NVIDIA software components required to run GPU-accelerated workloads.
These components include the NVIDIA drivers (to enable CUDA), the Kubernetes device plugin for GPUs, the NVIDIA Container Toolkit, automatic node tagging using GPU feature discovery (GFD), DCGM-based monitoring, and others.
The NVIDIA GPU Operator is only supported by NVIDIA. For more information about obtaining support from NVIDIA, see Obtaining Support from NVIDIA.
2.1. NVIDIA GPU prerequisites
- A working OpenShift cluster with at least one GPU worker node.
-
Access to the OpenShift cluster as a
cluster-admin
to perform the required steps. -
OpenShift CLI (
oc
) is installed. -
The node feature discovery (NFD) Operator is installed and a
nodefeaturediscovery
instance is created.
2.2. NVIDIA GPU enablement
The following diagram shows how the GPU architecture is enabled for OpenShift:
Figure 2.1. NVIDIA GPU enablement
MIG is only supported with A30, A100, A100X, A800, AX800, H100, and H800.
2.2.1. GPUs and bare metal
You can deploy OpenShift Container Platform on an NVIDIA-certified bare metal server but with some limitations:
- Control plane nodes can be CPU nodes.
Worker nodes must be GPU nodes, provided that AI/ML workloads are executed on these worker nodes.
In addition, the worker nodes can host one or more GPUs, but they must be of the same type. For example, a node can have two NVIDIA A100 GPUs, but a node with one A100 GPU and one T4 GPU is not supported. The NVIDIA Device Plugin for Kubernetes does not support mixing different GPU models on the same node.
- When using OpenShift, note that one or three or more servers are required. Clusters with two servers are not supported. The single server deployment is called single node openShift (SNO) and using this configuration results in a non-high availability OpenShift environment.
You can choose one of the following methods to access the containerized GPUs:
- GPU passthrough
- Multi-Instance GPU (MIG)
Additional resources
2.2.2. GPUs and virtualization
Many developers and enterprises are moving to containerized applications and serverless infrastructures, but there is still a lot of interest in developing and maintaining applications that run on virtual machines (VMs). Red Hat OpenShift Virtualization provides this capability, enabling enterprises to incorporate VMs into containerized workflows within clusters.
You can choose one of the following methods to connect the worker nodes to the GPUs:
- GPU passthrough to access and use GPU hardware within a virtual machine (VM).
- GPU (vGPU) time-slicing, when GPU compute capacity is not saturated by workloads.
Additional resources
2.2.3. GPUs and vSphere
You can deploy OpenShift Container Platform on an NVIDIA-certified VMware vSphere server that can host different GPU types.
An NVIDIA GPU driver must be installed in the hypervisor in case vGPU instances are used by the VMs. For VMware vSphere, this host driver is provided in the form of a VIB file.
The maximum number of vGPUS that can be allocated to worker node VMs depends on the version of vSphere:
- vSphere 7.0: maximum 4 vGPU per VM
vSphere 8.0: maximum 8 vGPU per VM
NotevSphere 8.0 introduced support for multiple full or fractional heterogenous profiles associated with a VM.
You can choose one of the following methods to attach the worker nodes to the GPUs:
- GPU passthrough for accessing and using GPU hardware within a virtual machine (VM)
- GPU (vGPU) time-slicing, when not all of the GPU is needed
Similar to bare metal deployments, one or three or more servers are required. Clusters with two servers are not supported.
Additional resources
2.2.4. GPUs and Red Hat KVM
You can use OpenShift Container Platform on an NVIDIA-certified kernel-based virtual machine (KVM) server.
Similar to bare-metal deployments, one or three or more servers are required. Clusters with two servers are not supported.
However, unlike bare-metal deployments, you can use different types of GPUs in the server. This is because you can assign these GPUs to different VMs that act as Kubernetes nodes. The only limitation is that a Kubernetes node must have the same set of GPU types at its own level.
You can choose one of the following methods to access the containerized GPUs:
- GPU passthrough for accessing and using GPU hardware within a virtual machine (VM)
- GPU (vGPU) time-slicing when not all of the GPU is needed
To enable the vGPU capability, a special driver must be installed at the host level. This driver is delivered as a RPM package. This host driver is not required at all for GPU passthrough allocation.
Additional resources
2.2.5. GPUs and CSPs
You can deploy OpenShift Container Platform to one of the major cloud service providers (CSPs): Amazon Web Services (AWS), Google Cloud Platform (GCP), or Microsoft Azure.
Two modes of operation are available: a fully managed deployment and a self-managed deployment.
- In a fully managed deployment, everything is automated by Red Hat in collaboration with CSP. You can request an OpenShift instance through the CSP web console, and the cluster is automatically created and fully managed by Red Hat. You do not have to worry about node failures or errors in the environment. Red Hat is fully responsible for maintaining the uptime of the cluster. The fully managed services are available on AWS and Azure. For AWS, the OpenShift service is called ROSA (Red Hat OpenShift Service on AWS). For Azure, the service is called Azure Red Hat OpenShift.
- In a self-managed deployment, you are responsible for instantiating and maintaining the OpenShift cluster. Red Hat provides the OpenShift-install utility to support the deployment of the OpenShift cluster in this case. The self-managed services are available globally to all CSPs.
It is important that this compute instance is a GPU-accelerated compute instance and that the GPU type matches the list of supported GPUs from NVIDIA AI Enterprise. For example, T4, V100, and A100 are part of this list.
You can choose one of the following methods to access the containerized GPUs:
- GPU passthrough to access and use GPU hardware within a virtual machine (VM).
- GPU (vGPU) time slicing when the entire GPU is not required.
Additional resources
2.2.6. GPUs and Red Hat Device Edge
Red Hat Device Edge provides access to MicroShift. MicroShift provides the simplicity of a single-node deployment with the functionality and services you need for resource-constrained (edge) computing. Red Hat Device Edge meets the needs of bare-metal, virtual, containerized, or Kubernetes workloads deployed in resource-constrained environments.
You can enable NVIDIA GPUs on containers in a Red Hat Device Edge environment.
You use GPU passthrough to access the containerized GPUs.
Additional resources
2.3. GPU sharing methods
Red Hat and NVIDIA have developed GPU concurrency and sharing mechanisms to simplify GPU-accelerated computing on an enterprise-level OpenShift Container Platform cluster.
Applications typically have different compute requirements that can leave GPUs underutilized. Providing the right amount of compute resources for each workload is critical to reduce deployment cost and maximize GPU utilization.
Concurrency mechanisms for improving GPU utilization exist that range from programming model APIs to system software and hardware partitioning, including virtualization. The following list shows the GPU concurrency mechanisms:
- Compute Unified Device Architecture (CUDA) streams
- Time-slicing
- CUDA Multi-Process Service (MPS)
- Multi-instance GPU (MIG)
- Virtualization with vGPU
Consider the following GPU sharing suggestions when using the GPU concurrency mechanisms for different OpenShift Container Platform scenarios:
- Bare metal
- vGPU is not available. Consider using MIG-enabled cards.
- VMs
- vGPU is the best choice.
- Older NVIDIA cards with no MIG on bare metal
- Consider using time-slicing.
- VMs with multiple GPUs and you want passthrough and vGPU
- Consider using separate VMs.
- Bare metal with OpenShift Virtualization and multiple GPUs
- Consider using pass-through for hosted VMs and time-slicing for containers.
Additional resources
2.3.1. CUDA streams
Compute Unified Device Architecture (CUDA) is a parallel computing platform and programming model developed by NVIDIA for general computing on GPUs.
A stream is a sequence of operations that executes in issue-order on the GPU. CUDA commands are typically executed sequentially in a default stream and a task does not start until a preceding task has completed.
Asynchronous processing of operations across different streams allows for parallel execution of tasks. A task issued in one stream runs before, during, or after another task is issued into another stream. This allows the GPU to run multiple tasks simultaneously in no prescribed order, leading to improved performance.
Additional resources
2.3.2. Time-slicing
GPU time-slicing interleaves workloads scheduled on overloaded GPUs when you are running multiple CUDA applications.
You can enable time-slicing of GPUs on Kubernetes by defining a set of replicas for a GPU, each of which can be independently distributed to a pod to run workloads on. Unlike multi-instance GPU (MIG), there is no memory or fault isolation between replicas, but for some workloads this is better than not sharing at all. Internally, GPU time-slicing is used to multiplex workloads from replicas of the same underlying GPU.
You can apply a cluster-wide default configuration for time-slicing. You can also apply node-specific configurations. For example, you can apply a time-slicing configuration only to nodes with Tesla T4 GPUs and not modify nodes with other GPU models.
You can combine these two approaches by applying a cluster-wide default configuration and then labeling nodes to give those nodes a node-specific configuration.
2.3.3. CUDA Multi-Process Service
CUDA Multi-Process Service (MPS) allows a single GPU to use multiple CUDA processes. The processes run in parallel on the GPU, eliminating saturation of the GPU compute resources. MPS also enables concurrent execution, or overlapping, of kernel operations and memory copying from different processes to enhance utilization.
Additional resources
2.3.4. Multi-instance GPU
Using Multi-instance GPU (MIG), you can split GPU compute units and memory into multiple MIG instances. Each of these instances represents a standalone GPU device from a system perspective and can be connected to any application, container, or virtual machine running on the node. The software that uses the GPU treats each of these MIG instances as an individual GPU.
MIG is useful when you have an application that does not require the full power of an entire GPU. The MIG feature of the new NVIDIA Ampere architecture enables you to split your hardware resources into multiple GPU instances, each of which is available to the operating system as an independent CUDA-enabled GPU.
NVIDIA GPU Operator version 1.7.0 and higher provides MIG support for the A100 and A30 Ampere cards. These GPU instances are designed to support up to seven multiple independent CUDA applications so that they operate completely isolated with dedicated hardware resources.
Additional resources
2.3.5. Virtualization with vGPU
Virtual machines (VMs) can directly access a single physical GPU using NVIDIA vGPU. You can create virtual GPUs that can be shared by VMs across the enterprise and accessed by other devices.
This capability combines the power of GPU performance with the management and security benefits provided by vGPU. Additional benefits provided by vGPU includes proactive management and monitoring for your VM environment, workload balancing for mixed VDI and compute workloads, and resource sharing across multiple VMs.
Additional resources
2.4. NVIDIA GPU features for OpenShift Container Platform
- NVIDIA Container Toolkit
- NVIDIA Container Toolkit enables you to create and run GPU-accelerated containers. The toolkit includes a container runtime library and utilities to automatically configure containers to use NVIDIA GPUs.
- NVIDIA AI Enterprise
NVIDIA AI Enterprise is an end-to-end, cloud-native suite of AI and data analytics software optimized, certified, and supported with NVIDIA-Certified systems.
NVIDIA AI Enterprise includes support for Red Hat OpenShift Container Platform. The following installation methods are supported:
- OpenShift Container Platform on bare metal or VMware vSphere with GPU Passthrough.
- OpenShift Container Platform on VMware vSphere with NVIDIA vGPU.
- GPU Feature Discovery
NVIDIA GPU Feature Discovery for Kubernetes is a software component that enables you to automatically generate labels for the GPUs available on a node. GPU Feature Discovery uses node feature discovery (NFD) to perform this labeling.
The Node Feature Discovery Operator (NFD) manages the discovery of hardware features and configurations in an OpenShift Container Platform cluster by labeling nodes with hardware-specific information. NFD labels the host with node-specific attributes, such as PCI cards, kernel, OS version, and so on.
You can find the NFD Operator in the Operator Hub by searching for “Node Feature Discovery”.
- NVIDIA GPU Operator with OpenShift Virtualization
Up until this point, the GPU Operator only provisioned worker nodes to run GPU-accelerated containers. Now, the GPU Operator can also be used to provision worker nodes for running GPU-accelerated virtual machines (VMs).
You can configure the GPU Operator to deploy different software components to worker nodes depending on which GPU workload is configured to run on those nodes.
- GPU Monitoring dashboard
- You can install a monitoring dashboard to display GPU usage information on the cluster Observe page in the OpenShift Container Platform web console. GPU utilization information includes the number of available GPUs, power consumption (in watts), temperature (in degrees Celsius), utilization (in percent), and other metrics for each GPU.
Additional resources
Legal Notice
Copyright © 2024 Red Hat, Inc.
OpenShift documentation is licensed under the Apache License 2.0 (https://www.apache.org/licenses/LICENSE-2.0).
Modified versions must remove all Red Hat trademarks.
Portions adapted from https://github.com/kubernetes-incubator/service-catalog/ with modifications by Red Hat.
Red Hat, Red Hat Enterprise Linux, the Red Hat logo, the Shadowman logo, JBoss, OpenShift, Fedora, the Infinity logo, and RHCE are trademarks of Red Hat, Inc., registered in the United States and other countries.
Linux® is the registered trademark of Linus Torvalds in the United States and other countries.
Java® is a registered trademark of Oracle and/or its affiliates.
XFS® is a trademark of Silicon Graphics International Corp. or its subsidiaries in the United States and/or other countries.
MySQL® is a registered trademark of MySQL AB in the United States, the European Union and other countries.
Node.js® is an official trademark of Joyent. Red Hat Software Collections is not formally related to or endorsed by the official Joyent Node.js open source or commercial project.
The OpenStack® Word Mark and OpenStack logo are either registered trademarks/service marks or trademarks/service marks of the OpenStack Foundation, in the United States and other countries and are used with the OpenStack Foundation’s permission. We are not affiliated with, endorsed or sponsored by the OpenStack Foundation, or the OpenStack community.
All other trademarks are the property of their respective owners.