An 8-GPU Blackwell server from Aivres with 180 GB HBM3e per GPU. The HGX B200 has been succeeded by the HGX B300. Limited secondary market availability through Arc Compute.
Eight B200 GPUs on the HGX B200 baseboard with 180 GB HBM3e per GPU and 1,440 GB total GPU memory. Connected via NVLink and NVSwitch for high-bandwidth GPU-to-GPU communication.
Dual Intel Xeon 6 (6500P/6700P) or AMD EPYC 9005 processors. Up to 2 TB DDR5 system memory across 32 DIMM slots.
Supports air-cooling with 20 hot-swap fans, designed for sustained high-utilization workloads. Direct liquid-cooled alternatives also available.
Purpose-built for AI and HPC workloads in a 10U form factor by an OEM that specializes in GPU server infrastructure.
The HGX B200 was the first Blackwell-generation GPU platform, delivering a significant performance uplift over the Hopper H100 and H200 for AI training and inference workloads. It has since been succeeded by the HGX B300, which offers higher memory capacity (288 GB vs 180 GB per GPU), NVLink 5, and improved performance per watt.
For organizations that already own HGX B200 systems, the platform remains a strong production asset. For new deployments, the HGX B300 is the recommended platform. Arc Compute can help with both: sourcing secondary market B200 systems for expansion, or planning a migration path to B300.
Every cluster is designed around your workloads, your facility, and your operational model.
Systems are validated, burned in, and shipped with full deployment support.
Hands-on experience designing and deploying high-density GPU infrastructure.
From planning and procurement through deployment, operations, and scaling.
