ASRock Rack 1U8GPU-C621A/2T

The ASRock Rack 1U8GPU-C621A/2T represents ASRock's most aggressive GPU density product, fitting eight full-length GPU cards into a 1U chassis via PCIe 4.0 switching fabric. The system is built on Intel's C621A Xeon Scalable platform and uses a PLX/Broadcom PEX89000 series PCIe switch to fan out from CPU PCIe lanes to all eight GPU slots, enabling cloud-tier GPU density that was previously only achievable in custom hyperscaler hardware. This platform targets managed inference service providers and enterprises deploying large-scale AI inference fleets.

With eight PCIe 4.0 x16 GPU connections, the platform can host eight NVIDIA A30, A10, or L40 inference GPUs simultaneously, allowing a single 1U node to serve thousands of concurrent AI inference requests. The PCIe switch topology introduces a small hop latency but enables GPU densities that dramatically reduce per-inference TCO when amortized across fleet scale. Dual 25GbE SFP28 networking with RDMA support ensures sufficient network bandwidth for batched inference request handling.

Cooling is handled by a high-velocity fan array designed for N+1 redundancy, with individual GPU thermal sensors reported through the ASPEED AST2600 BMC. The system ships with 3+1 redundant 3000W 80 Plus Titanium PSUs to handle the combined power draw of eight high-TDP accelerators. ASRock's system management software provides GPU health dashboards integrated with IPMI/Redfish.

Ideal for

• Large-scale AI inference clusters requiring maximum GPU cards per rack unit
• LLM serving infrastructure with token throughput optimization
• Managed AI-as-a-service platforms requiring high GPU tenant density
• Real-time recommendation engine inference at e-commerce scale
• Multi-model concurrent serving for enterprise AI platforms
• Cloud provider GPU rental fleets focused on cost-per-GPU-hour optimization

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.