Supermicro AS -4125GS-TNMR2-LCC barebone server, 4U liquid-cooled, AMD EPYC 9004, 8× MI300X GPU

Purpose-built AI compute platform

The Supermicro AS -4125GS-TNMR2-LCC is a 4U liquid-cooled barebones server designed for large-scale AI training and HPC workloads. It integrates eight AMD Instinct MI300X GPUs with dual AMD EPYC 9004 series processors in a single rackmount chassis. The platform targets research labs and cloud builders who need maximum GPU density with direct liquid cooling for sustained performance.

Replaces multi-node GPU clusters

This system consolidates what previously required several air-cooled 4U or 8U servers linked by external fabric into one 33.2-inch deep rack unit. The jump is worth it when workload demand exceeds the memory or compute capacity of four-GPU platforms and when facility power and coolant loops can support 5250W redundant Titanium supplies. It is not for buyers who need fewer than eight accelerators or who lack liquid-cooling infrastructure.

Fits liquid-cooled rack rows

The chassis provides 24 DDR5-4800 DIMM slots for up to 6TB memory and eight front hot-swap 2.5-inch NVMe bays for local datasets. Eight PCIe 5.0 x16 low-profile slots and two full-height half-length slots remain open for networking or storage expansion. Builders must supply SP5 processors, DDR5 modules, NVMe drives and a facility coolant distribution unit; the integrated cold plates and manifold remove the need for per-GPU air coolers and free rear-door space for dense row deployment.