Unveiling the NVIDIA RTX PRO 6000 Blackwell: A Workstation Powerhouse from GTC 2025!

This year’s NVIDIA GPU Technology Conference (GTC) 2025, held from March 17–20 in San Jose, was a showcase of NVIDIA’s relentless pursuit of innovation. Among the standout announcements was the NVIDIA RTX PRO 6000, a Blackwell-based workstation GPU that promises to redefine performance for professionals across industries like AI, 3D design and scientific computing. With its jaw-dropping specs—24,064 CUDA cores, 96GB of GDDR7 memory and a 600W power envelope—the RTX PRO 6000 isn’t just an incremental upgrade, it’s a bold leap forward. Let’s dive into what makes this card a game-changer and why it’s generating so much buzz.

Pic: NVIDIA RTX PRO 6000 Blackwell Workstation Edition

The RTX PRO 6000 is built on NVIDIA’s cutting-edge Blackwell architecture, a platform designed to push the boundaries of accelerated computing. At its heart lies the GB202 GPU, a near-fully enabled die boasting 188 Streaming Multiprocessors (SMs), translating to 24,064 CUDA cores. That’s 11% more than the consumer-facing RTX 5090, signalling NVIDIA’s intent to cater to professionals who demand uncompromising power. Add in 752 Tensor Cores for AI workloads and 188 RT Cores for ray tracing, and you’ve got a card that’s as versatile as it is powerful.

The real showstopper, though, is the memory: 96GB of GDDR7 VRAM with Error Correction Code (ECC) support, paired with a 512-bit memory bus. This delivers a staggering bandwidth of up to 1.79 TB/s—nearly double that of its predecessor, the RTX 6000 Ada Generation. For context, the RTX 5090 tops out at 32 GB, making the RTX 6000 Pro’s tripled capacity a clear nod to memory-hungry tasks like training large language models, rendering photorealistic 3D scenes or simulating complex datasets.

Pic: RTX PRO 6000 Blackwell Specification Table vs Previous Generation Flagships

NVIDIA didn't stop at one card model. The RTX PRO 6000 comes in three distinct variants, each targeting a specific professional niche:

Workstation Edition: This flagship model, with its 600W TGP, is crafted for desktop power users who take on the toughest workloads and offering the performance of 4000 AI TOPS (Tera Operations Per Second) and 125 TFLOPS of single precision floating point performance.  It employs a dual flow-through cooling system featuring two high-performance fans, closely mirroring the design NVIDIA implemented in their RTX 5090 Founders Edition GPUs. The card sports a double-width form factor and surpasses standard full-height specifications. It’s engineered for use in single-GPU, high-performance workstations.

Max-Q Workstation Edition: This model shares the same core specifications as its counterparts but is capped at 300W due to its form factor and cooling constraints which means that it offers slightly lower performance when comparing it to 600W parts of 3511 AI TOPS and 110 TFLOPS of FP32. It features a double-width, full-height design with a blower-style cooler, consistent with the previous-generation RTX 6000 Ada Generation and RTX A6000. Engineered for peak power efficiency, this card is optimised for mobile and multi-GPU workstation setups.

Server Edition: This passively cooled model leverages server chassis airflow and features a double-width, full-height form factor with a configurable TDP (400–600W), making it ideal for data centres running scalable AI inference or virtualised GPU workloads. It serves as a direct successor to the NVIDIA L40S, which was the server-grade counterpart to the RTX 6000 Ada in NVIDIA’s prior GPU generation and offering the performance of 3753 AI TOPS and 117.3 TFOPS of FP32.

Pic: RTX PRO 6000 Blackwell Edition trio at NVIDIA GTC 2025: Server (left), Max-Q (right) and Workstation (bottom)

NVIDIA claims the RTX PRO 6000 delivers 125 TFLOPS of single-precision (FP32) performance and a mind-blowing 4000 TOPS of AI performance at FP4 precision. That’s a massive leap over the RTX 6000 Ada’s roughly 2X inference performance boost, thanks to fifth-generation Tensor Cores and support for DLSS 4 Multi-Frame Generation. This translates to faster local LLM prototyping, accelerated neural rendering and real-time visualisation of complex 3D models.

The card also supports Multi-Instance GPU (MIG) technology, allowing a single GPU to be partitioned into up to 4 isolated instances. This is a boon for enterprises juggling diverse workloads—say, rendering a digital twin while training an AI model—without compromising performance or security.

Pic: NVIDIA DLSS4 on the RTX PRO 6000 Blackwell GPU uses transformer-based AI for advanced upscaling and ray reconstruction, boosting frame rates up to 8x for professional 3D and AI rendering workflows.

Who is this GPU for? The RTX PRO 6000 isn’t your average gaming GPU (though it could certainly handle that). It’s aimed squarely at professionals who push hardware to its limits:

• AI Researchers: With 96GB of VRAM and 4000 AI TOPS, it’s a dream for training and inferencing large-scale models locally.
• 3D Artists and Designers: Real-time ray tracing and neural rendering make it ideal for tools like NVIDIA Omniverse or Blender.
• Data Scientists: High memory bandwidth and MIG support streamline massive simulations and analytics.
• Enterprise IT: The Server Edition, paired with NVIDIA’s vGPU software, powers virtual workstations for remote teams.

Pic: RTX PRO 6000 Blackwell Workstation Edition performance improvements vs RTX 6000 Ada Generation

Pic: RTX PRO 6000 Server Edition performance vs NVIDIA L40S

The RTX PRO 6000 isn’t just a standalone release; it’s part of NVIDIA’s broader Blackwell strategy to dominate professional computing. Alongside announcements like the Blackwell Ultra GPU for data centres and AI factories, as well as the DGX Station, it underscores NVIDIA’s focus on AI-driven workflows. The “PRO” branding also hints at a sharper delineation between consumer (GeForce) and professional (RTX PRO) lines, echoing the spirit of the long-retired TITAN series but with a modern twist.

If the RTX PRO 6000 feels excessive for your needs, consider NVIDIA's lower-tier models unveiled at GTC this year. These cards directly replace the prior Ada Generation GPUs.

It is important to note that the RTX PRO GPU lineup does not support double-precision workloads. If your application requires FP64, consider datacentre GPUs like the NVIDIA H100NVL, H200NVL, B200 or B300.

Announced at GTC 2025, the RTX PRO 6000 is already rolling out. The Workstation and Max-Q Editions hit distributors in April 2025, with workstation manufacturers following in May. The Server Edition will soon land in configs from the server vendors like Supermicro, with cloud providers like AWS and Azure offering instances later this year. The workstation GPUs are no longer just about rendering pretty pictures—they’re about powering the future of AI, design and science. Its specs are overkill for most, but for the pros who need every ounce of performance, it’s a revelation. As it lands in workstations and servers over the coming months, expect to see it drive breakthroughs in everything from generative AI to virtual reality. Oh, and yes - it can run Crysis!

At Boston, we’re passionate about fuelling your business with cutting-edge innovation. Our collaboration with NVIDIA brings the powerful NVIDIA RTX 6000 Pro to our premium workstation and server solutions, ensuring top-tier performance and dependability. Exciting news: RTX 6000 Blackwell Edition units are arriving soon for exclusive customer test drives!  We understand the importance of the right fit, which is why we offer a no-obligation proof of concept to ensure the RTX 6000 Pro meets your unique needs. Reach out today to book a test drive or explore tailored solutions crafted by Boston. Our expert sales team is here to guide you every step of the way. Transform your business now - email [email protected] or call 01727 876100!

Author

Tom Michalski

FAE Manager

Boston Limited

Tags: nvidia, gtc, gtc 2025, rtx pro 6000, blackwell, gpu, nvidia rtx pro