PCIe GPU servers for RTX PRO 6000 Blackwell, H200 NVL and L40S — enterprise AI from data center to edge.
Supermicro offers more than 20 systems qualified for the NVIDIA RTX PRO 6000 Blackwell Server Edition, from 2U edge-friendly servers to 5U systems with up to eight GPUs. They form the basis of NVIDIA RTX PRO Servers for enterprise AI factories.
The same platforms also accept H200 NVL, L40S and other PCIe GPUs, so one standard air-cooled server design can cover inference, fine-tuning, rendering and virtual workstations.
2x the memory of L40S, GDDR7, PCIe 5.0 and MIG with up to 4 instances per GPU.
Modular designs with up to 4 GPUs in 2U or 8 GPUs in 4U.
Thermally optimised chassis for up to 10 GPUs, or 8 RTX PRO 6000, air-cooled.
Single- or dual-socket Intel Xeon 6 and AMD EPYC configurations.
Options from full-depth racks to short-depth edge servers.
| 2U MGX | Up to 4 double-width GPUs |
|---|---|
| 4U MGX | Up to 8 double-width GPUs |
| 5U PCIe GPU systems | Up to 10 GPUs (up to 8× RTX PRO 6000 Blackwell), air-cooled |
| CPUs | Intel Xeon 6 or AMD EPYC, single or dual socket |
| NVIDIA RTX PRO 6000 Blackwell Server Edition | 96 GB GDDR7, MIG up to 4 instances |
|---|---|
| NVIDIA H200 NVL | 141 GB HBM3e, NVLink bridge |
| NVIDIA L40S | 48 GB GDDR6 |
Front-I/O air-cooled server for 8× 1100 W Blackwell Ultra GPUs with integrated 800 Gb/s networking.
DLC-2 liquid-cooled 8-GPU node capturing up to 98% of heat for dense, efficient AI factories.
The most compact hyperscale HGX B300 platform — up to 144 liquid-cooled GPUs in one 21-inch rack.
Tell us your workload and timeline for the Supermicro NVIDIA RTX PRO Servers. We reply within one business day with a configuration, pricing and lead time.