SuperMicro SYS-821GE-TNHR-LC 8-GPU liquid-cooled AI server. Supports 8× NVIDIA H200/H100/B200 SXM5 GPUs with direct-to-chip liquid cooling, dual 4th/5th Gen Intel Xeon Scalable, 32× DDR5 DIMM slots. 8U rack-mount AI training platform with NVIDIA NVLink Switch.
8U Rack-Mount
8× SXM5 (H200/H100/B200)
Dual 4th/5th Gen Intel Xeon
32× DDR5 DIMM (8TB max)
Direct-to-Chip Liquid Cooling
OCP 3.0 + 8× PCIe 5.0
12× 2.5" NVMe/SATA
6× 3000W Titanium PSU
SuperMicro's SYS-821GE-TNHR-LC is the industry-leading 8U liquid-cooled GPU server for hyperscale AI training clusters. The 8U chassis accommodates full-height SXM5 GPU trays with integrated cold plates and redundant pump modules for maximum serviceability. With 32 DIMM slots supporting up to 8TB of DDR5 memory, this platform is ideal for memory-intensive training of MoE (Mixture of Experts) models where CPU memory capacity is the bottleneck. SuperMicro's Building Block Solutions design allows flexible configuration — specify NVLink Switch for full GPU mesh, or dedicate PCIe lanes to 8× NVIDIA ConnectX-7 400GbE NICs for RoCEv2-based scale-out networking. QS Compute handles full rack integration including manifold plumbing, leak detection systems, and CDU commissioning.
Need SYS-821GE-TNHR-LC? Built-to-order liquid-cooled AI servers.
Configured, tested, deployed. Contact us for pricing and availability.
Request Quote