NVIDIA’s 96GB RTX PRO 6000 Matches Four RTX 5090s on a 230B AI Model While Drawing a Quarter the Power

A benchmark showed a single NVIDIA RTX PRO 6000 Blackwell 96GB coming close to the token throughput of four RTX 5090s on a 230B-parameter AI model.

It delivered that performance with much lower power draw, highlighting the efficiency of a single workstation GPU.

The test also compared four RTX 4090s and DGX Spark, with the RTX PRO 6000 proving more practical than multi-GPU setups.

The results suggest that for large-model inference, one high-memory GPU may offer a simpler and better-value solution than several consumer cards.