At GTC 2025, Nvidia officially unveiled the RTX Pro 6000 GPU, built on the Blackwell architecture and specifically engineered for the professional workstation and server markets. This GPU uses the same GB202 chip as the consumer-focused RTX 5090 but introduces significant performance and configuration enhancements aimed at professionals such as designers, data scientists, and AI developers. The RTX Pro 6000 is available in three versions: Workstation, Max-Q Workstation, and Server, each tailored to meet the demanding needs of different high-performance scenarios.

The RTX Pro 6000 boasts impressive core specifications, featuring 188 stream processors (SMs), with only a 2% disablement compared to the GB202 chip's maximum of 192 SMs. This represents a 10.6% increase in SMs over the 170 in the RTX 5090 and includes an impressive array of 24,064 CUDA cores, 752 tensor cores, and 188 ray tracing cores. Nvidia claims single-precision floating-point performance (FP32) of up to 125 TFLOPS and a staggering 4,000 TOPS of AI compute capability (based on FP4 precision), extrapolating a boost clock around 2.6 GHz. Additionally, the card is equipped with a full 128MB L2 cache, improves upon the RTX 5090's 96MB, and integrates four NVENC encoders and four NVDEC decoders to enhance video processing capabilities.

The graphics memory configuration is another standout feature of the RTX Pro 6000. It utilizes a 24Gb (3GB) GDDR7 memory chip instead of the standard 2GB chip seen in the RTX 50 series. With 96GB of memory capacity via a 512-bit wide memory bus and a memory speed of 28Gbps, this design is ideal for handling large datasets and complex AI model training tasks. For comparison, the RTX 5090 offers only 32GB of memory with the same bandwidth but with significantly less capacity.

The Blackwell Workstation Edition mirrors the RTX 5090 in appearance with its glossy black finish and features four DisplayPort 2.1b outputs. Its power consumption (TGP) is rated at 600W, slightly exceeding the RTX 5090's 575W. The Max-Q Workstation Edition is designed for high efficiency and lower power consumption, with a reduced TGP of 300W, dual blower fans, and maintains a full-height, full-length dual-slot form factor with four DisplayPort 2.1b ports. The Server Edition, similar in form factor to the Max-Q Edition, is fanless and relies on server chassis airflow for cooling, with flexible configurations allowing it to consume up to 600W. All models utilize a 16-pin 12V-2x6 power connector, with the server and workstation versions benefitting from enhanced connector reliability due to superior manufacturing standards.

Alongside the RTX Pro 6000, Nvidia has introduced other models in the RTX Pro series, including the GB202-based RTX Pro 5000 (with 14,080 CUDA cores, 48GB of video memory, and a 300W TGP) and the GB203-based RTX Pro 4500 (10,496 CUDA cores, 32GB of video memory, and a 200W TGP), as well as the RTX Pro 4000 (8,960 CUDA cores, 24GB of video memory, and a 140W TGP). These models cater to desktop workstations and mobile devices, addressing professional requirements from high-end to entry-level. They all support the PCIe 5.0 x16 interface, offering double the bandwidth of the previous generation, and are compatible with DisplayPort 2.1b, supporting display outputs up to 4K 480Hz or 8K 165Hz.

The RTX Pro 6000 series is particularly notable for its advanced features. For example, the Workstation and Server Editions support Multi-Instance GPU (MIG) technology, which allows a single GPU to be split into four independent instances (each with 24GB of video memory) to improve multitasking and parallel processing. Furthermore, the ninth-generation NVENC encoder adds 4:2:2 encoding support, and the sixth-generation NVDEC decoder doubles the H.264 decoding throughput, making it ideal for video editing and real-time streaming applications.

Performance-wise, the RTX Pro 6000 surpasses its predecessor, the L40S GPU with Ada Lovelace architecture. Its large language model throughput is enhanced by up to 5x, genome sequencing is accelerated by nearly 7x, and text-to-video generation is sped up by 3.3x. Recommender system inference and rendering performance have approximately doubled, underscoring the RTX Pro 6000’s advantage in AI, scientific computing, and visual creation tasks.

The RTX Pro 6000 Workstation Edition is expected to become available through distributors in April, and from manufacturers starting in May. Server editions will be provided by manufacturers such as Cisco, Dell, HP, Lenovo, and Supermicro. Cloud providers like AWS, Google Cloud, and Microsoft Azure are also expected to launch instances later in the year. While pricing details remain unannounced, based on previous professional-level GPU pricing, the RTX Pro 6000 is anticipated to exceed $10,000, positioning it significantly above the approximately $3,000 retail price of the RTX 5090. This premium reflects its targeting of the professional market and the advanced specifications it offers.

The launch of the RTX Pro 6000 Blackwell series marks a significant advancement for Nvidia in the realm of professional computing. With its extensive 96GB video memory, robust AI computation capabilities, and adaptable design, it promises a substantial upgrade for technical enthusiasts and professionals demanding exceptional performance. Further details and performance data are eagerly awaited in the coming days.