Author: Tony Huynh, VCF Product Management

Selecting the right hardware for enterprise AI deployment can be a complex and daunting task. With VMware AI ReadyNodes, customers have a clear, reliable path to running Artificial Intelligence (AI) model fine-tuning and inference workloads on VMware Cloud Foundation (VCF). AI ReadyNodes are pre-validated, fully tested server systems designed to take the complexity out of hardware selection, helping guarantee seamless interoperability and effortless lifecycle management.

Why Choose VMware AI ReadyNodes? Key Customer Benefits

Deploying AI infrastructure requires high reliability, predictable performance, and smooth operational integration. The AI ReadyNode program delivers crucial advantages designed to accelerate your AI journey:

• Rigorously Tested and Validated: OEM and ODM partners perform comprehensive validation across both servers and GPUs, helping ensure compatibility with VMware Cloud Foundation.
• Purpose-Built for AI Workloads: Systems are explicitly certified with pre-qualified GPU hardware setups optimized specifically for intensive AI model fine-tuning and real-time inference tasks.
• Zero Guesswork and Faster Time-to-Value: By choosing end-to-end validated configurations, organizations eliminate hardware trial-and-error, reducing setup risks and speeding up enterprise AI adoption.

Finding Certified Systems on the Broadcom Compatibility Guide

Broadcom simplified the process of identifying servers and GPUs that have been certified to run as VMware AI ReadyNodes systems. Use the Broadcom Compatibility Guide (BCG) to quickly explore valid configurations by clicking on the “AI Ready Node (Server + GPU)” section on the main page.

Figure 1. AI Ready Node (Server + GPU) option in Broadcom Compatibility Guide

Clicking on a specific server model opens a page that provides more details on the specific GPUs supported.

Figure 2. Details on AI ReadyNode Servers and GPUs supported

Figure 2 shows that this specific server has been certified with NVIDIA H100 NVL, NVIDIA H200 NVL, and NVIDIA RTX PRO 6000 series GPUs in both Direct Passthrough and vGPU mode.

Expanding Ecosystem and Future Outlook

The current list of AI Ready Nodes continues to grow as we work with OEMs and ODMs to certify new AI Servers and GPUs.

Figure 3. Certified AI Ready Nodes

Please refer to the AI Ready Node (Server + GPU) page on the BCG for the latest list of AI Ready Nodes.

The Path to AI ReadyNode Certification: Four Essential Steps

To earn the distinction of an AI ReadyNode, server vendors must successfully clear four stringent certification and functional testing phases to guarantee enterprise-grade performance and stability:

1. vSphere Server Certification: Validates overall host server compatibility with vSphere through intensive stress benchmarks under simulated enterprise workloads.
2. VM DirectPath I/O for General GPU (Passthrough): Helps ensure graphics processors can pass through directly to virtual machines (VMs) bypassing the hypervisor layer.
3. DirectPath GPUs and Accelerators Certification: Certifies specialized vendor host drivers (such as NVIDIA, AMD, and Intel) to allow flexible GPU partitioning and multi-VM sharing within VCF.
4. GPU Functional Testing on vSphere: Executes real-world functional tests recommended by NVIDIA and AMD for their respective GPUs.  

OEMs and ODMs complete the certification steps for vSphere Server Certification, VM DirectPath I/O for General GPU, and GPU functional testing on vSphere. Broadcom collaborates with GPU vendors to complete DirectPath GPU and Acceleration certification. 

Simply passing a subset of the certification/functional test above does not qualify a server as an AI ReadyNode; all certifications and functional tests need to be completed before a server is listed as an AI Ready Node. If a specific server and GPU are listed on the VM DirectPath I/O for General GPU page and not on the AI ReadyNode page, the GPU can be used for Virtual Desktop Infrastructure (VDI) and Financial (ML) Application workloads.

vSphere Server Certification

The vSphere Server certification program is a well-established initiative designed to verify complete hardware compatibility between vSphere and equipment from OEMs and ODMs. Certification involves a series of functional tests to confirm key capabilities, including:

• ESX processor support
• Successful ESX system boot functionality
• Operation of specific operating systems (OS) on virtual machines (VMs)
• Interoperability with supported virtual devices and their corresponding physical devices

Upon successfully passing these tests, qualified servers are published under the “System / Servers” section of the Broadcom Compatibility Guide (BCG).

VM DirectPath I/O for General GPU

All GPU-enabled servers are certified for VM DirectPath I/O for General GPU. Certification testing may be performed on Windows or Linux running a Graphics or Compute workload and validates that a GPU can be directly passed to a VM, bypassing the hypervisor. Customers can refer to the VM DirectPath I/O for General GPU page on the BCG, which will show a list of Server, GPU and ESX configurations certified to run ESX in DirectPath mode. The listing provides the supported Guest OS and GPU workload for a given server + GPU + ESX configuration. 

DirectPath GPU and Accelerators

Broadcom works with GPU vendors such as NVIDIA and AMD to validate and certify host drivers, which allow a GPU to be assigned to one or more virtual machines. With host drivers, customers get additional visibility into the device and operational flexibility with VCF with capabilities like GPU sharing with multiple VMs, and providing deeper operational insights into the GPU through management tools such as vCenter and VCF Operations. An entry in this does not mean that the GPU is available for consumption with a server. Based on the use case, the user needs to refer to either the AI ReadyNode listing or VMDirectPath I/O for General GPUs listing to figure out which server this GPU is certified with. The list of GPUs and certified host drivers can be found on the DirectPath GPUs and Accelerators page. Please note that this page also includes hardware accelerators that are not GPUs, which are outside the scope of this blog.

GPU Functional Tests on vSphere

After completing the certifications listed above, OEM and ODM vendors execute GPU functional tests on vSphere, including:

• Real-world functional tests recommended by NVIDIA and AMD for their respective GPUs.
• Testing NVIDIA GPUs in DirectPath and vGPU modes, and AMD GPUs in DirectPath and Enhanced DirectPath modes.

Before executing GPU functional tests on vSphere, vendors must run bare-metal server and GPU performance tests from the GPU vendors. For example, the NVIDIA-Certified Systems program ensures systems are optimized for enterprise-grade performance, manageability, scalability, and security for AI and accelerated computing applications.