News report
NVIDIA RTX PRO 5500 Blackwell Packs 84GB of GDDR7 Into a 600W Workstation GPU
NVIDIA has added the RTX PRO 5500 Blackwell Workstation Edition with 84GB of ECC GDDR7, 21,760 CUDA cores and a 600W maximum power envelope.
On this page
At a glance
Key facts
- Architecture
- NVIDIA Blackwell
- CUDA cores
- 21,760
- GPU memory
- 84GB GDDR7 with ECC
- Memory bandwidth
- Up to 1,398 GB/s
- Maximum power
- 600W
- Target
- Professional workstation, AI, simulation and visualization workloads
- Availability
- Listed as coming soon; NVIDIA has not published a specific ship date or price
The RTX PRO 5500 fills the gap below NVIDIA’s 96GB flagship
NVIDIA now lists the RTX PRO 5500 Blackwell Workstation Edition in its professional desktop GPU family. The card is positioned below the RTX PRO 6000 Workstation Edition, but it remains a very large Blackwell configuration: 21,760 CUDA cores, fifth-generation Tensor Cores, fourth-generation RT Cores and 84GB of GDDR7 memory with ECC support.
The memory configuration is the headline specification. NVIDIA lists up to 1,398 GB/s of memory bandwidth, while the 84GB capacity is intended for workloads where fitting a larger model, dataset or scene into local GPU memory matters more than the economics of a gaming card. NVIDIA specifically describes uses spanning agentic and generative AI, simulation, rendering and professional visualization.
| GPU | CUDA cores | Memory | Max power |
|---|---|---|---|
| RTX PRO 5000 Blackwell | 14,080 | 48GB GDDR7 ECC | 300W |
| RTX PRO 5500 Blackwell | 21,760 | 84GB GDDR7 ECC | 600W |
| RTX PRO 6000 Blackwell Workstation Edition | 24,064 | 96GB GDDR7 ECC | 600W |
This is workstation hardware, not an 84GB GeForce card
The RTX PRO 5500 is relevant to PC hardware enthusiasts because its CUDA-core count matches the 21,760 cores listed for the GeForce RTX 5090, but the products target different markets and should not be treated as interchangeable. The PRO card emphasizes substantially larger ECC memory, professional software and deployment features rather than gaming value.
NVIDIA also lists Multi-Instance GPU support. The RTX PRO 5500 can expose its resources as a single 84GB instance or two 42GB instances, allowing isolated workloads to share one physical GPU. That capability is aimed at professional and enterprise workflows rather than normal desktop gaming.
Practical context
What this means for workstation PC builders
The RTX PRO 5500 gives workstation buyers another Blackwell option when 48GB is too restrictive but the full 96GB RTX PRO 6000 is unnecessary. Its unusually large local memory pool is particularly relevant to AI and content workloads whose working sets can exceed consumer-GPU VRAM capacity.
- 84GB of ECC GDDR7 creates substantially more headroom for memory-heavy local AI, simulation and rendering workloads than mainstream desktop GPUs.
- The 600W maximum power envelope makes PSU capacity, chassis airflow and workstation integration first-order considerations.
- MIG support can divide the GPU into two isolated 42GB instances for shared professional deployments.
- The lack of announced pricing means value comparisons against the RTX PRO 5000, RTX PRO 6000 or GeForce cards would currently be speculative.
Evidence status
Confirmed versus not established
Confirmed
- NVIDIA lists the RTX PRO 5500 Blackwell Workstation Edition with 21,760 CUDA cores, 84GB of ECC GDDR7 and up to 1,398 GB/s of memory bandwidth.
- NVIDIA lists a 600W maximum power consumption and Multi-Instance GPU support with one 84GB or two 42GB instances.
- The product is listed as coming soon in NVIDIA’s workstation lineup.
Unconfirmed
- NVIDIA has not published a specific retail price or exact shipment date on the cited product page.
- This article does not infer gaming performance from CUDA-core count or compare benchmark performance without independent testing.
Sources
Primary and technical sources
These sources support the reporting and analysis above. Current stories are updated when later evidence materially changes the facts.