AEWIN detailed two new AMD EPYC 9006-based AI servers, the 1U BG18-A10710 and 5U BG58-A20810, both built around an integrated two-phase direct liquid cooling (2P DLC) design aimed at higher compute density and tighter thermal control in AI and HPC deployments.
The 1U BG18-A10710 is built for high-density heterogeneous compute in a compact chassis. It supports a single AMD EPYC “Venice” SP7 CPU with up to 600 W TDP, two PCIe Gen6 SW FHFL accelerators, 16 DDR5 RDIMM or MRDIMM slots with memory speeds up to 12,800 MT/s, and eight E1.S Gen6 NVMe SSD bays. For management, it includes OCP DC-SCM2 system management via an ASPEED AST2700 BMC, and it’s configured with redundant 2,700 W 80 PLUS Titanium power supplies.
The 5U BG58-A20810 targets larger-scale parallel compute. It supports dual AMD EPYC 9006 SP8 server CPUs rated at 400 W TDP and “next-gen DDR5/MRDIMM” memory subsystems. On the accelerator side, AEWIN’s top configuration supports up to 16 PCIe Gen6 NVIDIA RTX Pro 6000 GPUs integrated with 2P DLC. AEWIN also lists an alternative configuration supporting up to eight air-cooled PCIe Gen6 accelerators.
Two-phase DLC is a very different thermal design problem than single-phase cold plates: AEWIN describes using a low-boiling-point coolant and a phase-change mechanism (liquid-to-vapor transition) to improve heat transfer and cut energy consumption. If it performs as intended in real deployments, the practical upside for operators is more headroom before fans, heat exchangers, and facility cooling become the limiting factor on GPU and CPU density.
AEWIN said it is ready to provide samples of the BG18-A10710 and BG58-A20810 for technical evaluation.
Source: AEWIN
