ASUS RS700A-E12-RS4U 1U AMD EPYC GPU Inference Server

The ASUS RS700A-E12-RS4U is a 1U rack-mounted GPU inference server engineered around the AMD EPYC 9004 series (Genoa) processor platform, delivering exceptional compute density within a single rack unit. Supporting dual AMD EPYC 9004 series processors with up to 192 cores combined, the system is built on the ASUS KRPA-U16 server board architecture and provides substantial PCIe 5.0 bandwidth to accommodate high-throughput GPU and accelerator workloads. The platform supports up to 6TB of DDR5 ECC Registered memory across 24 DIMM slots, enabling large in-memory datasets critical to AI inference pipelines.

Designed specifically for AI inference and GPU-accelerated compute, the RS700A-E12-RS4U accommodates up to four dual-slot FHFL (Full Height Full Length) GPUs through a rear-loading, tool-less mechanism, making it well suited for deployment of NVIDIA data center GPUs such as the L40S or H100 PCIe variants within constrained 1U rack space. The system provides up to 3200W of redundant power supply capacity through 80 PLUS Titanium certified PSUs, ensuring efficiency under sustained GPU inference loads. Dedicated OCP 3.0 network mezzanine support and multiple PCIe 5.0 expansion slots facilitate high-bandwidth networking and storage connectivity.

The RS700A-E12-RS4U integrates ASUS ASMB11-iKVM for out-of-band management, supporting IPMI 2.0, Redfish, and ASUS Control Center Enterprise (ACCE) for centralized fleet management across datacenter deployments in enterprise, cloud, and colocation environments. Front-accessible NVMe storage bays combined with rear GPU access enable operational efficiency in dense rack configurations. The system is validated for demanding AI inference, HPC, and deep learning serving workloads where per-rack-unit GPU density and total memory bandwidth are primary design constraints.

Ideal for

• Large-scale AI model inference serving for natural language processing and computer vision workloads requiring high GPU density per rack unit
• Enterprise deep learning inference pipelines where low-latency response times and high concurrent request throughput are mission-critical requirements
• High-performance computing (HPC) workloads leveraging AMD EPYC core counts and PCIe 5.0 bandwidth for parallel scientific simulations
• Colocation and cloud datacenter GPU node deployments where 1U form factor maximizes GPU capacity within constrained rack space allocations
• Generative AI and large language model (LLM) inference at the edge of enterprise datacenters requiring redundant power and remote management
• Data analytics acceleration and GPU-enabled database query processing for real-time business intelligence platforms

Technical specifications

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