NVIDIA's Beastly RTX Pro 5500 Blackwell Packs 84GB GDDR7 & RTX 5090 Core Count
NVIDIA has quietly expanded its professional graphics lineup with the RTX Pro 5500 Blackwell Workstation Edition, a Blackwell-based beast designed for heavy AI workloads, simulations, and other graphics chores that extend well beyond the realm of gaming. There is no press release or formal announcement (that we're aware of), though an official product page highlights some key specs.
Incidentally, the RTX Pro 5500 Blackwell Workstation Edition's naming scheme seemingly positions NVIDIA's newest GPU in the middle of its workstation stack. However, the hardware specs paint a much higher-end story, starting with the number of CUDA cores—it packs 21,760 CUDA cores, which is the same as NVIDIA's consumer flagship GeForce RTX 5090.
This is not just a retooled GeForce RTX 5090 for the workstation, though. NVIDIA also stuffed the RTX Pro 5500 Blackwell Workstation Edition with a generous 84GB of ECC GDDR7 memory. That is over 2.6 times the 32GB memory capacity of its consumer sibling.
In addition to the huge pool of GPU memory, NVIDIA lists 1,398GB/s of memory bandwidth. There is no mention of the bus width, though one of NVIDIA's hardware partners, PNY, lists it as being 416-bit. If true, it indicates NVIDIA is mixing 3GB and 4GB die densities. Alternatively, it could be a typo on PNY's part and the actual bus width might be 448-bit, which would be the cleaner fit. As a point of reference, the RTX Pro 6000D (China) sports 28 x 3GB for 84GB of total VRAM on a 448-bit bus.
Memory configuration mystery aside, the card also features PCIe Gen 5.0 x16 connectivity, four DisplayPort 2.1b outputs, and three ninth-gen NVENC and sixth-gen NVDEC media engines.
NVIDIA rates the GPU for a full 600W maximum board power, matching the thermal envelope of its top-tier RTX Pro 6000 (not the China model). In fact, the Pro 5500 sits much closer to the flagship Pro 6000 (24,064 cores, 96GB VRAM) than it does to the 300W RTX Pro 5000 (14,080 cores, 48GB/72GB VRAM), effectively filling what was previously a notable gap in its enterprise lineup.
Beyond raw horsepower, another key feature of interest to AI developers is support for NVIDIA's Multi-Instance GPU (MIG) technology. The card can operate as a unified 84GB pool for serving massive local LLMs and complex agentic AI workflows, or it can be partitioned into two completely isolated 42GB hardware instances.
"Every instance has dedicated memory, cache, and compute cores plus guaranteed quality of service. In rack-mounted deployments, that means more users get reliable accelerated computing from the same hardware," NVIDIA explains.
There is no mention of official pricing for the RTX Pro 5500 Blackwell Workstation Edition, either on NVIDIA's listing or PNY's product page. Considering the 48GB RTX Pro 5000 commands $8,599 and up on the street, this 84GB, 600W monster will likely land somewhere north of $10,000.
