RTX PRO 5000 Blackwell 72GB

Specs, explained

Used for AI development, LLM inference and data science

1. 0172 GB of GDDR7 memory

   VRAM

   72 GB

   Memory type

   GDDR7

   Memory bandwidth

   1.344 TB/s

   GDDR is graphics memory mounted on the board around the GPU rather than stacked on the package. It is widely used on graphics, rendering and cost-efficient inference cards.

   VRAM is the working space on the card. Model weights, activations and datasets have to fit in it, or be split across several GPUs. Rule of thumb for the weights alone at 16-bit precision (about 2 GB per billion parameters): up to roughly 36 billion parameters, before activations and cache. This is arithmetic on the recorded VRAM, not a benchmark.

2. 02Blackwell architecture

   Architecture

   Blackwell

   Generation

   RTX PRO Blackwell

   Launch year

   2026

   Lifecycle status

   Current

   Blackwell added fifth-generation Tensor Cores and support for lower-precision formats down to FP4. (General context about the architecture, not a statement about this specific card.)

   Launched in 2026. Kovara's records list it as a current-generation part.

3. 03Form factor and power

   NVLink

   Not supported

   MIG

   Supported

   Form factor

   PCIe Gen5 dual-slot

   TDP

   300 W

   Kovara's records show no NVLink support, so multi-GPU jobs on this part rely on PCIe or the network between servers to move data.

   Multi-Instance GPU (MIG) can divide one physical GPU into isolated slices, each with its own memory and compute, so several smaller workloads can share a card.

   PCIe cards slot into standard server and workstation expansion slots, which suits a wide range of systems.

   TDP is the rated thermal design power: the planning figure for power delivery and cooling in a server.

4. 04What it is used for

   Primary use (recorded)

   AI development / LLM inference / data science / rendering

   These are the uses Kovara has recorded for this part. The notes explain the workload in general terms; they are not performance claims.

   • AI development
   • LLM inferenceRunning a trained model to answer requests. Memory capacity, latency and cost per request are the usual concerns.
   • data scienceSpeeding up analytics and classical machine learning on large datasets.
   • renderingProducing images and video from 3D scenes.

Figures are Kovara master data, last verified 03 Sept 2026. Explanations describe the technology in general; they are not benchmarks and say nothing about price or availability.