Today, we’re introducing Nvidia’s new card. The newly launched RTX Pro 6000 Blackwell workstation graphics card is based on the Blackwell architecture and features the GB202 graphics processor, boasting 24,064 CUDA cores and 188 streaming multiprocessors, with a boost clock of up to 2,617 MHz. In comparison, the flagship gaming graphics card GeForce RTX 5090 also utilizes the GB202 chip but has 21,760 CUDA cores and a boost clock of 2,410 MHz. The RTX Pro 6000 is equipped with 96GB of GDDR7 memory with error-correcting code (ECC) support, achieving a bandwidth of 1,792 GB/s through a 512-bit memory bus, while the RTX 5090 has 32GB of memory with the same bandwidth.
This graphics card is offered in three models: standard workstation, Max-Q workstation, and server. The standard workstation version features a dual-fan, dual-slot, open-air cooling design with a total graphics power of 600W, powered by a single 16-pin PCIe 5.0 power connector. The Max-Q version has a power limit of 300W and employs a blower-style cooler, making it suitable for dense workstations. The server version relies on server fans for passive cooling and is designed for data centers. The card supports PCIe 5.0, offering double the bandwidth, and is equipped with four DisplayPort 2.1b ports, supporting multi-monitor configurations.
Geekbench 6 OpenCL benchmarks show the RTX Pro 6000 scoring 368,219 points, slightly lower than the RTX 5090’s 376,858 points, a difference of approximately 2.3%. In specific tests, the RTX Pro 6000 outperformed the RTX 5090 in Horizon Detection (21.3 GPixel/s), Edge Detection (32.1 GPixel/s), and Gaussian Blur (36.3 GPixel/s), but lagged slightly in Background Blur (263.9 FPS vs. 310.7 FPS) and Face Detection (196.7 FPS vs. 241.5 FPS). However, the drivers used for testing were not fully optimized, and OpenCL memory access was limited to 23.8GB, failing to fully utilize the 96GB of memory. Optimized drivers should yield different results.
GameTechBench full path tracing tests indicate that the RTX Pro 6000 performs about 5% better than the RTX 5090 at 4K resolution, with a 2% lead in offline path tracing rendering. Its 96GB memory and 752 fifth-generation Tensor Cores excel in handling large 3D models, AI inference, and real-time ray tracing. The ninth-generation NVENC engine supports 4:2:2 H.264 and HEVC encoding, increasing video encoding speed, and improves AV1 encoding quality by approximately 5%. The sixth-generation NVDEC engine provides double the H.264 decoding throughput and supports the 4:2:2 format, making it suitable for video editing workflows.
The Blackwell architecture introduces several technological innovations. Streaming multiprocessors integrate neural shaders, embedding AI capabilities into programmable shaders and supporting RTX Mega Geometry and DLSS 4.0 multi-frame generation technologies. Single-precision floating-point performance reaches 125 TFLOPS, and AI performance is up to 4,000 TOPS. The memory adopts a dual-sided layout, with 16 3GB GDDR7 modules on each side, totaling 96GB, making it suitable for virtual reality, scientific computing, and multi-application workflows.
The RTX Pro 6000 is already listed on US retailer Connection, priced between $8,435 and $8,565, about 26% higher than the RTX 6000 Ada. The standard workstation and Max-Q versions are similarly priced, while the server version is available through partners like Cisco and Dell. The workstation version is already available, the server version will ship in May, and the mobile version will be integrated into professional notebooks by manufacturers such as Dell and HP in June. The top-tier model, RTX Pro 6000 X Blackwell, may feature the full GB202 chip with 192 streaming multiprocessors.
At GDC 2025, the EmberGen real-time visual effects software ran 40% faster on the RTX Pro 6000 compared to the RTX 4090. Render Network showcased its efficient performance in neural rendering and generative AI creative workflows at GTC 2025. This graphics card targets fields such as game development, AI training, content creation, and industrial design, suitable for users who require large amounts of memory and computing power.
Although initial tests show little performance difference compared to the RTX 5090, driver limitations restricted memory utilization. Optimized drivers may further unlock its potential. Its high power consumption and price positioning make it more suitable for enterprise-level deployments. Nvidia differentiates its professional and consumer product lines through the RTX Pro brand, enhancing market targeting. More test data in the future will reveal its true performance.