During the Red Hat Summit celebration, Red Hat and AMD announced an expansion of their strategic collaboration aimed at delivering more efficient solutions in generative AI and modernizing virtualized infrastructures in hybrid environments. The combination of Red Hat’s open technologies with the power of AMD EPYC processors and AMD Instinct GPUs is intended to optimize both performance and scalability for increasingly diverse and demanding workloads.
AMD Instinct and Red Hat OpenShift AI: Optimized Inference in the Hybrid Cloud
One of the main announcements was the full integration of AMD Instinct GPUs into Red Hat OpenShift AI, allowing organizations to deploy artificial intelligence models more efficiently without requiring extreme resources. In tests conducted on Microsoft Azure, Red Hat and AMD demonstrated that it is possible to scale language models (SLM and LLM) within a single virtual machine with multiple GPUs, reducing performance costs by avoiding resource fragmentation across multiple VMs.
Thanks to the use of AMD Instinct MI300X GPUs and the Red Hat Enterprise Linux AI platform, the partnership has showcased a robust, efficient inference environment that is compatible with open architectures.
Collaboration on vLLM: More Efficient AI Inference with AMD GPUs
The two companies also announced their joint work in the vLLM (vectorized LLM) ecosystem, with direct contributions to the upstream project to accelerate the inference of language models. Notable improvements include:
- Performance enhancements on AMD GPUs through the incorporation of the AMD kernel library, optimization of the Triton kernel, and FP8 support for dense and quantized models.
- Improved support for multi-GPU systems, enabling more scalable and energy-efficient execution in distributed environments.
- Expansion of the vLLM ecosystem, with participation from other players like IBM, strengthening the ongoing development of the project and its compatibility with AMD hardware.
As a result of this collaboration, AMD Instinct GPUs will natively support the Red Hat AI Inference Server, the enterprise distribution of vLLM, facilitating the execution of open-source models on validated and optimized hardware.
EPYC and OpenShift Virtualization: Frictionless Modernization
Red Hat also highlighted the validation of Red Hat OpenShift Virtualization for AMD EPYC processors, enabling organizations to unify virtual machine and container workloads on a single cloud-native platform. This capability is key to upgrading traditional data centers, consolidating infrastructure, and reducing TCO (total cost of ownership) in terms of hardware, licenses, and energy.
AMD EPYC thus positions itself as the ideal platform for hosting GPU-enabled systems, improving return on investment even for highly demanding AI workloads. Compatible servers with this technology include solutions from Dell PowerEdge, HPE ProLiant, and Lenovo ThinkSystem.
Official Statements
Ashesh Badani, Senior Vice President and Head of Product at Red Hat, stated:
“Leveraging the real benefits of AI requires flexibility and scalability. This collaboration with AMD expands the options available for modernizing infrastructures and preparing production environments with next-generation accelerators and open technologies.”
Philip Guido, Commercial Director at AMD, also commented:
“The combination of Red Hat’s open platforms with our Instinct GPUs and EPYC CPUs delivers the performance and efficiency that our customers demand to accelerate innovation in AI, virtualization, and hybrid cloud.”
via: AMD