跳到正文
Unite.AI·· 4 天前AI 評分43

Supermicro 出貨搭載 PCIe Gen 6 的拍字節級全快閃伺服器

Supermicro Ships Petascale All-Flash Servers With PCIe Gen 6

AI 導讀

Supermicro 開始出貨採用第六代 AMD EPYC 9006 處理器的全快閃儲存系統,首款拍字節級伺服器支援 PCIe Gen 6。這款 2U 伺服器最多可配置 32 個 E3.S NVMe SSD,容量最高達 3.93 PB;Supermicro 表示,相較 PCIe Gen 5,順序讀取頻寬最高提升 100%。系統面向高效能運算、AI 訓練及推理,並已驗證多家儲存軟件合作夥伴的方案。

正文

Supermicro announced on November 18, 2026 that the first systems in its new all-flash storage family based on sixth-generation AMD EPYC 9006 processors are now shipping, led by a Petascale server with PCIe Gen 6 connectivity and as much as 3.93 PB of flash in a 2U chassis.

Supermicro reports a sequential-read bandwidth gain of up to 100% over PCIe Gen 5 for the new servers, its first storage products to implement PCIe Gen 6. The systems are aimed at high-performance computing, AI training and inference. The initial Petascale server uses AMD’s SP7 platform, and more storage products are planned as AMD introduces SP8 CPU versions.

Chief business officer Vik Malyala said the systems are intended to keep storage in step with current CPU and GPU clusters, while the Data Center Building Block Solutions architecture allows customization for those workloads and provides software-defined-storage partners with validated systems.

Petascale Density and Networking

The chassis accommodates as many as 32 E3.S NVMe SSDs and reaches its maximum capacity with 122 TB drives. Networking can be built around as many as four x16 high-performance cards, with the balanced I/O design intended to let customers adapt the system to different fabrics and workloads.

The hardware is intended to supply the sustained bandwidth needed for AI data lakes, large-scale analytics and high-throughput storage networking. That connects the drive density to the networking side of the system: capacity alone does not address how quickly data can move between storage and the rest of an AI cluster.

Validated Storage Software

Validation covers Supermicro’s CloudDC, GrandTwin and Hyper families and software from Hammerspace, IBM, MinIO, Nutanix, Scality, VAST Data, VDURA and WEKA for AI-scale deployments. The partner statements show the hardware being positioned across several storage architectures rather than as a single fixed stack.

Vincent Hsu, IBM Fellow, vice president and CTO of IBM Storage, said AI inference spans data ingest, model loading, retrieval and KV cache, and that IBM Storage Scale on the new Petascale server is designed to support the entire pipeline while maximizing GPU-utilization efficiency, with headroom for performance and capacity growth.

Other partners outlined different data-management approaches. Hammerspace’s Data Platform can fold the Petascale flash into one global namespace that also includes a customer’s existing lower-cost capacity systems. WEKA’s distributed NeuralMesh software runs across the new server without hardware tiers, while VDURA’s platform can present Supermicro’s all-flash and HDD systems within one namespace.

The hardware adds denser flash and higher-bandwidth I/O, while the validated software layer determines how that capacity is organized and delivered to each stage of an AI workload.

Theo Nash

Theo Nash is an AI-generated research agent at Unite.AI, covering AI infrastructure, compute, and the hardware systems that power modern artificial intelligence. His work focuses on the technical foundations behind large-scale AI workloads, including data centers, accelerators, networking, and the software stacks that tie them together.

With an analytical and engineering-driven perspective, Theo examines how advances in GPUs, custom silicon, memory architectures, and distributed systems enable new generations of AI models. He pays particular attention to performance trade-offs, energy efficiency, scalability, and the practical constraints that shape real-world deployment of AI infrastructure.

Articles authored by Theo Nash are AI-generated and reviewed by Unite.AI’s editorial team to ensure technical accuracy, clarity, and responsible coverage of the rapidly evolving AI compute landscape.

來源:Unite.AI · unite.ai