$190,000 AI Server with 8x RTX PRO 6000 Blackwell GPUs

Описание: 768GB of RAM. 768GB of VRAM. One server.

This is a 4U AI server built around eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, 96GB each, for 768GB of GPU memory in a single machine. Enough to keep massive models and datasets resident on one system instead of splitting the work across many.

Feeding the GPUs is a dual-socket setup: two AMD EPYC 9755 processors, 128 cores each, 256 cores total, with the PCIe lanes and memory channels to drive all eight cards at full bandwidth at once. Alongside them, 768GB of DDR5 ECC across 24 modules, populated to keep every channel on both CPUs full.

Storage is tiered for how the work actually flows. Fast NVMe M.2 for the OS, and 30TB of enterprise U.2 NVMe for active datasets. Networking is dual 200GbE for moving data in and out at speed, and the whole system runs on redundant power for continuous operation.

Every card was inspected and seated by hand, then the full system was validated under load before it shipped.

Build Specs
Chassis: ASUS ESC8000A-E13P, 4U, 8-bay
CPU: 2x AMD EPYC 9755 (128C each, 256 cores total)
Memory: 768GB DDR5-5600 ECC RDIMM (24x 32GB)
GPUs: 8x NVIDIA RTX PRO 6000 Blackwell Server Edition (96GB each, 768GB total)
Boot: 2x 1.92TB NVMe PCIe4 M.2
Storage: 4x 7.68TB PCIe 4.0 NVMe U.2
Networking: Dual 200GbE QSFP112
Power: 3+1x 3200W redundant (240V)
OS: Ubuntu 24.04 LTS

#aiworkflow #pc #nvidia #gpuserver #multigpu #gpu #highperformancecomputing #aiserver
#workstation