{"id":104,"date":"2026-10-26T13:35:28","date_gmt":"2026-10-26T12:35:28","guid":{"rendered":"https:\/\/mspbusiness.com\/en\/it-news\/supermicro-launches-liquid-cooled-rack-with-256-nvidia-vera-cpus\/"},"modified":"2026-10-26T13:51:13","modified_gmt":"2026-10-26T12:51:13","slug":"supermicro-launches-liquid-cooled-rack-with-256-nvidia-vera-cpus","status":"publish","type":"post","link":"https:\/\/mspbusiness.com\/en\/technology-and-architecture\/supermicro-launches-liquid-cooled-rack-with-256-nvidia-vera-cpus\/","title":{"rendered":"Supermicro launches liquid-cooled rack with 256 NVIDIA Vera CPUs"},"content":{"rendered":"

Supermicro is introducing the NVIDIA Vera CPU rack, a fully liquid-cooled system with 256 Vera CPUs and 22,528 cores. The company is targeting the rack at agentic AI and reinforcement learning, with the CPUs running the sandbox environments of AI agents. The rack can be deployed on its own or combined with the Vera Rubin NVL72.<\/strong><\/p>\n

The rack is the latest addition to Supermicro’s DCBBS (Data Center Building Block Solutions) portfolio. Alongside agentic AI and reinforcement learning (RL), the company cites data processing as an application area. The system complements the Supermicro NVIDIA Vera Rubin NVL72, which the company is now shipping, and is available immediately.<\/p>\n

Specifications<\/h3>\n

The rack contains 256 NVIDIA Vera CPUs with a total of 22,528 cores and 45,056 threads. According to Supermicro, a single rack supports up to 22,500 concurrent sandbox environments for AI agents. NVIDIA BlueField-4 DPUs are built in for networking and security. Cooling is handled by Supermicro’s DLC-2 direct liquid cooling stack, making the entire rack liquid-cooled. Those who prefer air cooling can also opt for 1U and 2U systems within Supermicro’s Vera CPU portfolio.<\/p>\n

Background to CPU demand<\/h3>\n

Supermicro and NVIDIA attribute the arrival of the rack to a shift in AI workloads. With agentic AI and RL, processes no longer run linearly but branch out and are often cyclical, with many intermediate steps. GPUs generate tokens, code and queries, while CPU-based sandboxes execute actions, evaluate results, run tools and orchestrate workflows. According to both companies, progress is therefore increasingly dependent on available CPU capacity.<\/p>\n

Chris Marriott, vice president of enterprise platforms at NVIDIA, states that supporting thousands of concurrent agent workflows requires a new class of high-density CPU infrastructure. Charles Liang, president and CEO of Supermicro, says that by expanding DCBBS the company is responding to changing data centre requirements. Liang adds that Supermicro and NVIDIA aim to shorten deployment times through a common architecture.<\/p>\n

MGX design and 800 VDC power<\/h3>\n

The rack is based on the third generation of NVIDIA MGX, a modular reference architecture, and therefore shares its infrastructure architecture with the Vera Rubin NVL72. Supermicro says this supports standardised deployment. For power, customers can choose between traditional AC power to the rack or an 800 VDC sidecar power rack. With the latter option, battery backup units (BBUs) handle power outages instead of a conventional UPS. The company expects this to free up floor space, potentially improve power conversion efficiency and reduce both CapEx and OpEx.<\/p>\n

Control over AI agents<\/h3>\n

The rack supports the NVIDIA OpenShell Reference Architecture. OpenShell runs on the Vera CPUs and applies least-privilege policies and controls to AI agents at runtime. In addition, NVIDIA Sentry runs as an out-of-band monitor on the BlueField-4 DPUs. Supermicro says Sentry enforces policies, provides telemetry and can quarantine an agent within milliseconds if it oversteps the permitted boundaries. Because this monitoring takes place on the DPU, it consumes no CPU capacity on the host system, according to the company.<\/p>\n

Combined with NVL72 or standalone<\/h3>\n

For customers building on the DCBBS Blueprints for Vera Rubin NVL72, Supermicro offers an optional design that combines both rack types. This design includes a defined network topology based on NVIDIA Spectrum-X Ethernet, hot-aisle containment, redundant in-row coolant distribution units (CDUs) and rear door heat exchangers.<\/p>\n

As a standalone CPU platform, Supermicro cites applications such as agent orchestration and sandboxing, database operations, cluster management, job scheduling with Slurm and Kubernetes, data analytics and large-scale software build pipelines. In both scenarios, a single Supermicro team guides the process, from site survey and design to L11 and L12 integration, testing, on-site deployment and ongoing support.<\/p>\n","protected":false},"excerpt":{"rendered":"

Supermicro is making a fully liquid-cooled rack with 256 NVIDIA Vera CPUs available. The system delivers 22,528 cores and targets AI agent sandbox environments alongside GPU clusters.<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"mspb_bron_id":1169412,"mspb_url_nl":"https:\/\/mspbusiness.com\/technologie-en-architectuur\/supermicro-brengt-vloeistofgekoeld-rack-met-256-nvidia-vera-cpus-uit-3\/","mspb_url_de":"https:\/\/mspbusiness.com\/de\/technologie-und-architektur\/supermicro-bringt-fluessigkeitsgekuehltes-rack-mit-256-nvidia-vera-cpus-auf-den-markt\/","mspb_url_en":"","mspb_beeld_url":"https:\/\/mspbusiness.com\/wp-content\/uploads\/2026\/10\/hero-1793018117123.png","mspb_beeld_klein":"https:\/\/mspbusiness.com\/wp-content\/uploads\/2026\/10\/hero-1793018117123.png","mspb_auteur":"Redactie","mspb_beeld_alt":"","mspb_video":"","_yoast_wpseo_metadesc":"Supermicro launches a liquid-cooled CPU rack with 256 NVIDIA Vera CPUs and BlueField-4 DPUs for agentic AI and reinforcement learning.","footnotes":""},"categories":[5],"tags":[],"layout":[6],"class_list":["post-104","post","type-post","status-publish","format-standard","hentry","category-technology-and-architecture","layout-news"],"_links":{"self":[{"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/posts\/104","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/comments?post=104"}],"version-history":[{"count":1,"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/posts\/104\/revisions"}],"predecessor-version":[{"id":105,"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/posts\/104\/revisions\/105"}],"wp:attachment":[{"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/media?parent=104"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/categories?post=104"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/tags?post=104"},{"taxonomy":"layout","embeddable":true,"href":"https:\/\/mspbusiness.com\/en\/wp-json\/wp\/v2\/layout?post=104"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}