NVIDIA has officially rolled out a new version of its RTX PRO 5000 Blackwell GPU, and the headline upgrade is hard to miss: a huge 72 GB of GDDR7 memory. For professionals working with AI, neural rendering, simulation, and heavyweight 3D production pipelines, this is the kind of VRAM jump that can immediately change what fits in memory—and what no longer needs to be split into smaller chunks.
The RTX PRO Blackwell stack first arrived with the flagship RTX PRO 6000, featuring an enormous 96 GB memory pool. But for many creators and AI developers, stepping down one tier to the RTX PRO 5000 meant a much sharper drop in VRAM. NVIDIA is now closing that gap by making a 72 GB RTX PRO 5000 available alongside the existing 48 GB model, giving workstation buyers a more flexible middle option between “good” and “maxed out.”
Importantly, the new 72 GB edition isn’t a different performance tier—it keeps the same core specifications as the RTX PRO 5000 Blackwell already on the market. It’s powered by NVIDIA’s GB202 GPU core and includes 14,080 CUDA cores, up to 2,142 AI TOPS, and a 384-bit memory interface delivering 1.34 TB/s of bandwidth. Total board power is rated at 300W, and the card comes in an air-cooled dual-slot design.
So how does NVIDIA get from 48 GB to 72 GB without changing the rest of the card? It comes down to memory module density. The 72 GB version uses 24 GDDR7 memory “sites” instead of the 16 used by the 48 GB card, while maintaining the same 28 Gbps memory speed across the same 384-bit bus. The result is a straightforward but meaningful upgrade: 50% more VRAM while keeping the same overall class of GPU.
That extra memory is especially relevant right now because professional GPU workloads are increasingly memory-limited. Agentic AI tools, larger LLMs, bigger datasets, and more complex scenes all benefit when more of the workload can sit entirely in VRAM. NVIDIA positions the RTX PRO 5000 Blackwell 72 GB as a solution for these growing demands, with the capacity increase aimed at enabling larger models, bigger AI agents, and fewer compromises when juggling multiple high-memory tasks.
NVIDIA also highlights significant generational gains over the prior RTX PRO 5000 generation in key AI and creator workloads. The company cites up to a 3.5x performance gain for image generation and a 2x gain for text generation versus the previous generation. For LLM inference, NVIDIA claims a 2.1x improvement. In professional rendering and content creation applications—such as Arnold, Chaos V-Ray, Blender, and GPU renderers like D5 Render and Redshift—the new card is said to reduce render times by as much as 4.7x compared with the previous generation, depending on the workload.
Availability is already underway, with the RTX PRO 5000 Blackwell offered in both 48 GB and 72 GB variants through NVIDIA’s partner ecosystem. Pricing hasn’t been confirmed yet, but given the large memory uplift and the professional target market, the 72 GB model is expected to slot in as a premium option for studios, engineering teams, and AI builders who need more headroom without stepping up to the RTX PRO 6000.
For quick context, the broader RTX PRO Blackwell lineup spans multiple tiers, from the 96 GB RTX PRO 6000 flagship down to compact and lower-power models like the RTX PRO 4000 SFF and RTX PRO 2000. Within that range, the RTX PRO 5000 remains a key “sweet spot” class—and with a 72 GB configuration now on the table, it becomes a much more compelling choice for VRAM-hungry professional workflows.
