Business
Supermicro Among First to Unveil NVIDIA BlueField-4 STX Storage Server to Improve AI Inference Performance
Supermicro illustrates leadership with one of the first Context Memory (CMX) storage servers, built on the NVIDIA STX reference architecture for AI
About this update from Super Micro Computer, Inc.
[{"type":"text","content":"Supermicro illustrates leadership with one of the first Context Memory (CMX) storage servers, built on the NVIDIA STX reference architecture for AI storage.The BlueField-4 STX storage server combines NVIDIA Vera CPU and NVIDIA ConnectX-9 SuperNIC.Supermicro's storage server builds upon last year's introduction of the Petascale JBOF all-flash array powered by NVIDIA BlueField-3.SAN JOSE, Calif., March 17, 2026 /PRNewswire/ -- Supermicro, Inc. (NASDAQ: SMCI), a Total IT Solution Provider for AI, Cloud, Storage, and 5G/Edge, today unveiled one of the industry's first context memory (CMX) storage server as part of NVIDIA STX reference architecture announced at NVIDIA GTC 2026. STX is a new modular reference architecture from NVIDIA which is designed to accelerate the full lifecycle of AI.\n \n \n \n \n \n \n \n\"Supermicro continues to be first to market with new rack scale architectures designed to exceed the needs of a rapidly evolving AI Factory customer base,\" said Charles Liang, president and CEO of Supermicro. \"Building upon last year's introduction of the Petascale JBOF (Just a Bunch of Flash), where we proved the feasibility of a JBOF powered by NVIDIA BlueField-3 DPUs, we have developed the CMX storage server. Our prototype of the latest storage architecture demonstrates the level of our collaboration with NVIDIA, and our commitment to be first-to-market with game changing technologies.\"For more information about the new Supermicro storage server built on the NVIDIA STX reference architecture please visit: www.supermicro.com/en/solutions/ai-storageLeveraging the STX architecture, the CMX server is designed to address the challenge of long-lived AI queries and multi-stage chain-of-thought agentic workloads, which require the prior and intermediate tokens associated with the user's query to be accessed. This solution both accelerates the results and reduces the power which would otherwise be required to recompute the results when the local storage required for the tokens is exceeded. This storage of tokens, called Key Value (KV) cache, is managed by NVIDIA Dynamo, NVIDIA's inference orchestration layer.As the STX solution comes to market, Supermicro will be working with these software partners and others on porting and validation. Additionally, Supermicro long-standing relationships with leading SSD providers such as Micro...