Supermicro GPU SuperServer 420GP-TNAR+

 

Artificial Intelligence (AI), Deep Learning, and High-Performance Computing (HPC) demand hardware that can handle massive datasets, extreme parallelism, and rapid scaling. The Supermicro GPU SuperServer 420GP-TNAR+ is purpose-built to meet these challenges, combining GPU density, advanced cooling, and next-generation processing power into a single 4U chassis.

 

Compute and Memory Performance

At the core, the system supports dual 3rd Generation Intel Xeon Scalable processors (Socket P+ LGA-4189). Depending on the configuration, Platinum CPUs provide up to 40 physical cores and 80 threads, all with PCIe 4.0 support, which offers twice the throughput of PCIe 3.0. Each processor features 64 PCIe lanes and 8 memory channels, allowing for significant performance gains compared to previous generations.

The 420GP-TNAR+ features a total of 32 DIMM slots, supporting up to 12TB of memory. Maximum density is achieved through a hybrid approach: Intel Optane Persistent Memory (DCPMM) modules combined with RDIMMs or LRDIMMs. Standard configurations provide up to 8TB of DRAM, ensuring bandwidth and capacity for the most demanding AI and HPC workloads.

 

GPU and Expansion Capabilities

Designed for GPU acceleration, this server supports the NVIDIA HGX A100 8-GPU board (40GB or 80GB). Interconnected with NVLink and NVSwitch, GPUs communicate at up to 600 GB/s, nearly ten times faster than PCIe Gen4 alone. This architecture dramatically reduces time-to-solution for AI training, inference, and analytics.

Beyond GPUs, the chassis supports up to 8 high-speed I/O cards, including InfiniBand adapters with 100Gb/s or 200Gb/s bandwidth. This ensures a 1:1 ratio of GPUs to RDMA-capable NICs, enabling direct GPU-to-network communication for distributed AI workloads.

 

Storage and Networking

The system integrates 10 hot-swappable 2.5-inch drive bays — 6 in the front (SATA, SAS, or NVMe) and 4 in the rear dedicated to NVMe. For OS support, two M.2 slots provide additional flexibility. With tool-less trays and optional SAS RAID controllers, the storage subsystem ensures high throughput and simplified management.

Networking is equally robust, with dedicated OCP 3.0 AIOM slots, PCIe Gen4 NIC support, and a management interface via IPMI 2.0 and Redfish API.

 

Cooling and Power Efficiency

To maintain stable operation under extreme loads, the chassis includes four heavy-duty cooling fans at the front and redundant Titanium-level power supplies at the rear. The 420GP-TNAR+ features 4× 2+2 redundant 3000W PSUs, while the 420GP-TNAR variant offers 4× 3+1 redundant 2200W Platinum PSUs, providing flexibility based on deployment needs.

 

Conclusion

The Supermicro GPU SuperServer 420GP-TNAR+ delivers unmatched density and performance in a 4U form factor. With up to 8 NVIDIA A100 GPUs, dual Intel Xeon CPUs, 12TB memory support, and redundant Titanium-grade power, it is engineered for next-generation AI, HPC, and large-scale data workloads. Offering over 5 petaFLOPS of performance, this system is a future-ready solution for enterprises and research institutions requiring maximum compute efficiency, scalability, and reliability.