NVIDIA Launches RTX PRO 5500 Blackwell Workstation GPU
NVIDIA released the RTX PRO 5500 Blackwell Workstation Edition featuring 84 GB of GDDR7 memory for agentic AI and large-scale graphics workloads.
NVIDIA introduced the RTX PRO 5500 Blackwell Workstation Edition, a professional graphics card engineered for agentic AI, physical simulations, and large-scale graphics workloads. Based on GB202 silicon, the GPU features 21,760 CUDA cores, a specification that matches the consumer-grade GeForce RTX 5090.
The professional card distinguishes itself with 84 GB of GDDR7 memory, providing more than 2.5 times the capacity of the RTX 5090. It operates with a 600W TDP and supports both air- and liquid-cooled thermal solutions to accommodate rack-mounted deployments. Hardware specifications include a PCIe Gen5 x16 interface, four DisplayPort 2.1b outputs, and a 416-bit bus interface.
A key feature of the card is Multi-Instance GPU technology, which allows the hardware to be partitioned into two isolated 42 GB instances. NVIDIA stated that "every instance has dedicated memory, cache, and compute cores plus guaranteed quality of service," which enables more users to access reliable accelerated computing from the same hardware in rack deployments. While official pricing remains unannounced, industry estimates place the cost of the card above $10,000.