Nvidia has unveiled its RTX Pro 5500 Blackwell workstation GPU, packing RTX 5090-class silicon with a whopping 84GB of GDDR7 memory.
The professional card slots between the RTX Pro 5000 and RTX Pro 6000, targeting AI, simulation and graphics workloads rather than gamers chasing another three frames. Nvidia lists the RTX Pro 5500 on its site as “Coming Soon”, with pricing and a firm release date still tucked safely out of sight.
The card uses Nvidia’s Blackwell architecture and carries 21,760 CUDA cores, the same number found in the GeForce RTX 5090. That comparison gets more interesting when you factor in memory. The RTX 5090 has 32GB of GDDR7, while the RTX Pro 5500 gets 84GB of ECC GDDR7.
That gives the workstation part more than 2.6 times the memory capacity of Nvidia’s gaming flagship, which should prove rather more useful for large local AI models. Nvidia rates memory bandwidth at 1,398GB/s and maximum power consumption at 600W.
The card connects through PCIe 5.0 x16 and can provide up to four DisplayPort 2.1b outputs. It carries three ninth-generation NVENC encoders and three sixth-generation NVDEC decoders for professional video workloads.
Nvidia is pitching the RTX Pro 5500 primarily at rack-mounted workstation deployments rather than conventional desktop towers. Air-cooled versions will be available, while Nvidia says a liquid-cooled RXM configuration is planned for systems where 600W of GPU needs slightly more civilised plumbing.
The card supports Multi-Instance GPU technology, allowing one physical GPU to be partitioned into two isolated 42GB instances. That could make the board particularly attractive for shared AI development, virtual workstations and other environments where dedicating an entire 84GB GPU to one user seems extravagant.
The RTX Pro 5500 is much closer to the RTX Pro 6000 than its name might suggest. Nvidia’s RTX Pro 6000 Workstation Edition carries 96GB of GDDR7 and has the same 600W maximum power rating.
Independent figures reported by Tweaktown put the RTX Pro 6000 at 24,064 CUDA cores, leaving the 5500 with roughly 9.6 per cent fewer. The RTX Pro 5000 sits further down with either 48GB or 72GB of ECC GDDR7 and a maximum power draw of 300W.
Nvidia says fifth-generation Tensor cores provide support for FP4 precision, while fourth-generation RT cores handle the ray-tracing end of the process. The company is pitching the card at agentic AI, physical simulation, scientific computing, data analytics, rendering and high-resolution video production.
Its official specifications remain preliminary, and Nvidia warns they may change before the RTX Pro 5500 finally escapes into paying customers’ racks.
