NVIDIA RTX PRO 5000 Blackwell with 72 GB, targeted VRAM correction instead of a real surprise

NVIDIA has officially launched the RTX PRO 5000 Blackwell with 72 GB GDDR7 onto the market, closing a gap that it had previously created for itself. Technically, this is not a new GPU, but strategically it is a necessary step. The PRO series previously suffered from an unfortunate staggering: the RTX PRO 6000 with 96 GB at the top and an abrupt drop to 48 GB at the bottom. For many professional workloads, this gap was problematic, both technically and in terms of price.

With the new variant, NVIDIA has increased the memory capacity of the RTX PRO 5000 by 50 percent without changing the rest of the configuration. The core remains the GB202 chip, equipped with 14,080 CUDA cores, 2,142 AI TOPS, a 384-bit memory interface, 1.34 TB/s bandwidth and a TDP of 300 watts. The design also remains unchanged, air-cooled, dual-slot, workstation-compatible. The only real difference is where it counts for AI and professional visualization: in the memory. The path to the 72 GB is technically simple, but deliberately chosen. NVIDIA uses 24 GDDR7 memory chips instead of the previous 16, still connected at 28 Gbps. No higher clock rate, no risky thermal experiments, but more capacity with an identical platform. This underlines the fact that this SKU is not intended as a prestige product, but as a pragmatic response to real demand.

Especially in the field of agentic AI, LLM inference, simulation and neural rendering, VRAM has long been the limiting factor. Models are getting bigger, textures heavier, scenes more complex. 48 GB is theoretically enough in many cases, but only with compromises in practice. 72 GB pushes this limit noticeably. According to NVIDIA, this allows larger language models to be run locally, more parameters to be held simultaneously and more complex pipelines to be run without constant swapping. The nominal 2,142 AI TOPS ensure that the computing power does not become a bottleneck. According to the manufacturer, the RTX PRO 5000 Blackwell 72 GB is significantly more powerful than the previous generation. Performance is said to be 3.5 times higher for image generation, 2 times higher for text generation and 2.1 times higher for LLM inference. Classic Pro software such as Arnold, Chaos V-Ray, Blender, D5 Render or Redshift should also benefit from greatly reduced render times, in some cases up to 4.7 times faster. As expected, these figures come from NVIDIA’s own benchmarks, but clearly show where the card is positioned: not as a gamer product, but as a tool.

The larger context is interesting. NVIDIA is demonstrating a flexibility that has long been missing in the consumer segment. Memory placement is modular, no longer rigid. This is fueling speculation that similar steps could also follow with upcoming RTX 50 SUPER models, at least in theory. For the professional market, the message is clear: those who were stuck between 48 GB and 96 GB now have a realistic option. NVIDIA is still keeping its prices under wraps. The card is available immediately via partners such as Ingram Micro and Leadtek, but there are no concrete figures. In view of its positioning, it is likely to be significantly lower than the RTX PRO 6000, but noticeably higher than the 48 GB version. Exactly where many customers have not had a clean choice so far.

The bottom line is that the RTX PRO 5000 Blackwell 72 GB is not a technological sensation, but an overdue correction. NVIDIA is responding to a hunger for memory, which it is fueling with ever larger models. This is good news for professional users, but for NVIDIA it is above all damage limitation with an announcement.