VAST Data Expands Partnership with AMD to Integrate AI Operating System with EPYC CPUs and Instinct GPUs

VAST Data announced an expanded collaboration with AMD to integrate its AI Operating System with 6th Gen AMD EPYC CPUs and Instinct GPUs, aiming to support high-performance AI infrastructure at scale. The collaboration includes a reference architecture with DriveNets, KV cache optimizations, and VAST's selection of AMD's upcoming 6th Gen EPYC processors, formerly codenamed Venice, for its CBox and EBox platforms. VAST said its early testing with an AMD Instinct MI355X GPU showed a 9X speedup in time-to-first-token; the performance results were not independently verified by AMD and may not be typical. VAST also cited deployments across AI cloud providers including Crusoe, Core42, and TensorWave.

Full-stack AI platforms

• ▪VAST Data expanded its collaboration with AMD to integrate the VAST AI Operating System with 6th Gen AMD EPYC CPUs and AMD Instinct GPUs
• ▪VAST Data, AMD, and DriveNets developed an AI Infrastructure Reference Architecture featuring AMD Helios rack-scale infrastructure, VAST AI OS, and DriveNets AI Fabric networking

Neocloud ecosystem development

• ▪AI cloud providers delivering AMD-powered services integrated with VAST Data include 5C, Core42, Crusoe, EmbeddedLLM, Phanos.AI, TensorWave, and Vultr
• ▪TensorWave vice president Ilya Tabakh said neoclouds and dense AI data centers require more GPUs connected more closely than general-purpose hyperscale environments

Open-source AI infrastructure

• ▪TensorWave introduced ScalarLM, an open-source AI platform for training, fine-tuning, and serving large language models on a GPU-agnostic architecture
• ▪Crusoe Cloud utilizes AMD's open architecture and upgradeable Supermicro systems to maintain technology flexibility and freedom of choice

AI storage bottlenecks

• ▪VAST Data said its early testing with an AMD Instinct MI355X GPU showed a 9X speedup in time-to-first-token and 9.7X more token throughput using VAST for KV Cache offloading; the performance results were not independently verified by AMD and may not be typical
• ▪The AMD Pensando Pollara 400 AI NIC connects AMD Instinct GPUs to the VAST AI OS using NFS over TCP and NFS over RDMA

CPU orchestration for GPUs

• ▪High-speed CPUs act as a bridge for PCIe lanes connecting SSDs and network interface cards to provide computational power for agentic AI workloads
• ▪AMD product marketing executive Penny Tseng said the company would launch what she described as the first x86 PCIe Gen 6 CPU around November 2026, adding that it would help GPUs access data and provide double the generational bandwidth

AMD EPYC Venice processor

• ▪AMD corporate vice president Derek Dicker said the next-generation EPYC Venice processor is expected to deliver a 70% generation-over-generation performance increase
• ▪AMD announced in May 2026 that its next-generation EPYC processor, codenamed Venice, was ramping up production in Taiwan
• ▪The 6th Gen AMD EPYC Venice processor features PCIe Gen 6 support, enabling twice the I/O bandwidth compared to PCIe Gen 5