2 GPU 1U inference server
- GPUs
- 2 x NVIDIA L40S (48 GB)
- CPU
- 1 or 2 x AMD EPYC 9005, or 2 x Intel Xeon 6
- Memory
- Up to 768 GB DDR5 RDIMM
- Chassis
- 1U, about 1.2 kW, air cooled
A 200 W Blackwell workstation GPU with 32 GB GDDR7 ECC and 896 GB/s, for compact developer workstations, 1U inference servers and visualization systems.
RTX PRO 4500 Blackwell delivers the Blackwell feature set, including fifth generation Tensor cores with FP4, fourth generation RT cores, two NVENC and two NVDEC engines, in a dual slot card that draws only 200 W. Its 32 GB of GDDR7 with ECC at 896 GB/s is enough to run 7B to 14B parameter models at full precision or larger quantized models, and it drives four DisplayPort 2.1 displays.
The low power envelope is the point. It fits small form factor workstations with a modest power supply, it can be installed in pairs in 1U servers for edge and departmental inference, and it keeps noise and heat down on a desk. We use it in the compact developer workstation and the 1U and 2U inference server classes, and we supply it as an individual card.
We quote every class that carries it, from a single node to a pod. Tell us the workload and we propose the chassis.