Skip to content
KO EN
AI 인프라 Upcoming

ASRock Rack 4U16X-GNR2: NVIDIA B300 8-GPU Server Pushes Liquid Cooling and High-Bandwidth Networking

ASRock Rack has unveiled its 4U16X-GNR2 server, a high-performance AI system built on the NVIDIA HGX B300 platform. The server features two Intel Xeon 6 “G

ASRock Rack has unveiled its 4U16X-GNR2 server, a high-performance AI system built on the NVIDIA HGX B300 platform. The server features two Intel Xeon 6 “Granite Rapids” processors and eight NVIDIA Blackwell Ultra GPUs, connected via onboard NVLink interconnect. The 8-GPU topology is a well-established configuration for AI training and inference, and both versions of the server rely on liquid cooling—either traditional direct liquid cooling (DLC) or a two-phase cooling solution from ZutaCore for those who prefer to avoid water in their loops.

What Happened: Key Features and Design

The 4U16X-GNR2 is a dense 4U chassis that packs 12 front-facing 2.5-inch U.2 NVMe bays, with two connected to the CPU and ten to the PCIe switch complex. Front I/O includes four USB 3 Type-A ports, a VGA port, and a power button. Notably, ASRock Rack places the management port and two 1GbE ports (Intel i350-based) on the front panel, with internal cables allowing customers to relocate these ports to the rear if desired—a flexible cabling solution that avoids separate SKUs. Below these ports are eight OSFP 800Gbps cages, providing over 6.4Tbps of network bandwidth without any additional PCIe cards. The rear panel features redundant 3kW 80Plus Titanium power supplies, hot-swappable fans, and additional I/O ports.

Why It Matters: Evolution of AI Infrastructure

This server underscores the growing importance of liquid cooling and high-bandwidth networking in AI infrastructure. Eight Blackwell Ultra GPUs generate immense heat that air cooling cannot handle, making liquid cooling a necessity. The inclusion of ZutaCore’s two-phase cooling option also addresses concerns about water in data centers, offering a flexible path to adoption. Meanwhile, the front-panel 800Gbps OSFP ports aim to eliminate network bottlenecks in AI clusters, where GPU-to-GPU data movement is skyrocketing. This design could significantly boost overall system performance for large-scale AI workloads.

XPLAIN AI’s Interpretation and Analysis

We see this server as a signal that cooling technology and high-speed networking are becoming key differentiators in the AI hardware market. While NVIDIA (NVDA) provides the powerful Blackwell Ultra GPUs, their full potential depends on server-level engineering. ASRock Rack is differentiating itself with flexible cabling and liquid cooling options that appeal to large-scale data center operators. The ZutaCore partnership also highlights a trend toward diverse liquid cooling solutions, which could benefit private companies in the cooling ecosystem. However, we note that the success of this server hinges on the pace of liquid cooling adoption in data centers, which remains an uncertainty.

Beneficiaries and Risks

  • Beneficiaries: ASRock Rack’s parent company ASRock, NVIDIA (NVDA) as the GPU supplier, and Intel (INTC) whose Granite Rapids CPUs are used. Private liquid cooling firms like ZutaCore could also gain visibility.
  • Risks: Air-cooled server vendors such as Dell Technologies (DELL) and Hewlett Packard Enterprise (HPE) may lose competitiveness if they lag in liquid cooling. AMD (AMD) and its Instinct GPU-based servers face direct competition, potentially pressuring AMD’s ecosystem.

Counter-Scenarios and Uncertainties

The server’s market success depends on how quickly data centers adopt liquid cooling infrastructure. Many facilities are not ready for liquid cooling, and the initial cost and maintenance burden are higher than air cooling. ZutaCore’s two-phase technology has not yet been validated at large scale, adding risk. Competitors like Supermicro (SMCI) and Quanta Cloud Technology are likely developing similar liquid-cooled servers, so competition will be intense. We will need to see actual deployment cases before drawing firm conclusions.

Metrics to Watch Next

Investors should monitor adoption by major cloud providers (AWS, Microsoft Azure, Google Cloud), as their orders would be a strong validation. Also track NVIDIA Blackwell Ultra GPU shipments and pricing, as well as Intel Granite Rapids’ server market share. Indirect indicators include earnings of liquid cooling component suppliers like Boyd Corporation and CoolIT Systems.

#AIServer #LiquidCooling #NVIDIA #BlackwellUltra #IntelGraniteRapids #ASRockRack #DataCenter #GPUServer

Sources

Written by: XPLAIN AI Editorial Team · Reviewed by: XPLAIN AI Editorial Desk
This content was drafted with AI assistance based on publicly available sources and reviewed under XPLAIN AI's editorial standards.

Found an error? Request a correction →