Supermicro has introduced a new 10U air-cooled server designed for high-performance artificial intelligence (AI) and inference workloads in data centers. This server integrates AMD Instinct MI355X graphics processing units (GPUs) and is aimed at organizations requiring advanced compute acceleration in air-cooled environments. Supermicro claims this configuration delivers up to four times the AI compute capability over the previous generation and up to a 35 times increase in inference performance compared to earlier models, as referenced from AMD.
The server is built on Supermicro’s Data Center Building Block architecture and uses the industry standard Open Compute Project Accelerator Module (OAM) form factor. Each AMD Instinct MI355X GPU features 288 GB of high-bandwidth memory (HBM3e), 8 TB/s bandwidth, and supports a thermal design power (TDP) boost from 1,000 W to 1,400 W per GPU. Supermicro states that this allows double-digit percentage performance improvement over comparable 8U air-cooled systems, resulting in higher rack-level efficiency and throughput.
According to Supermicro, this offering joins the company’s portfolio of both air-cooled and liquid-cooled GPU servers optimized for scalable AI and inference tasks in data center environments. The new server supports enterprise and cloud workloads where AI acceleration and inference at scale are critical.
The server is currently available for order and has been showcased at the 2025 Supercomputing Conference (SC25) in St. Louis, Missouri. For additional technical information, Supermicro directs engineers and operators to its AMD accelerator solutions page.
Source: Supermicro
