company news about Instinctive and EPYC; VAST Data sets up widespread AMD CPU/GPU collaboration

John Mao

Key Expanded Collaboration Pillars

• -VAST adopts 6th Gen AMD EPYC (codenamed Venice) to power its 6th-gen CBox and 3rd-gen EBox platforms underlying VAST AI OS. Gen 6 EPYC features PCIe Gen 6, delivering double the I/O bandwidth and lower latency than PCIe Gen 5 for VAST DataBase and DataEngine-backed AI data services including databases, data warehouses and event streaming.
• -A tripartite AI Infrastructure Reference Architecture co-developed by VAST, AMD and DriveNets, incorporating AMD Helios rack-scale AI hardware, VAST AI OS and DriveNets AI Fabric networking. The blueprint details infrastructure support for training, inference, reinforcement learning and KV cache workloads, plus sizing guidance for teams building AI factories.
• -Wider ecosystem partnerships with TensorMesh, EmbeddedLLM and other vendors to accelerate production-grade inference stacks optimized for agentic AI.
• -New KV cache and inference optimizations merging AMD Instinct GPUs, Infinity Context, ROCm software and VAST AI OS.
• -Automated KV Cache Lifecycle Management: Clearing cache holding sensitive personal data poses major enterprise compliance hurdles. The integrated stack leverages VAST’s native data lifecycle rules to auto-expire and erase KV cache data, delivering robust security, privacy and regulatory compliance without manual admin work.
• -AMD Pensando Pollara 400 AI NIC creates high-speed data pathways linking Instinct GPUs to VAST AI OS. Supporting NFS over TCP and RDMA, it seamlessly shuttles data between GPU memory and VAST’s NVMe SSD storage clusters, powering KV cache and storage access critical to large-scale inference and agentic AI.
• -Verified deployments across leading AI cloud providers offering AMD-based AI services globally, including 5C, Core42, Crusoe, EmbeddedLLM, Phanos.AI, TensorWave and Vultr.