NVIDIA has officially launched its RTX PRO 6000 "Blackwell" series of GPUs aiming at the presumer and server segment with loads of power.
NVIDIA Rolls Out RTX PRO 6000 "Blackwell" GPUs With Tons of Compute Capabilities & Massive 96 GB VRAM
Today, NVIDIA introduces its next-generation Prosumer, Data Center, and AI-specialized graphics card, the RTX PRO 6000 "Blackwell". This monster of a graphics solution comes with even more compute power than the gaming RTX 5090. The card will be available in three variants, starting with the standardized design along with a Max-Q and server flavor, depending on the demands of the consumers.
- NVIDIA Streaming Multiprocessor: Offers up to 1.5x faster throughput and new neural shaders that integrate AI inside of programmable shaders to drive the next decade of AI-augmented graphics innovations.
- Fourth-Generation RT Cores: Delivers up to 2x the performance of the previous generation to create photoreal, physically accurate scenes and complex 3D designs with optimizations for NVIDIA RTX™ Mega Geometry.
- Fifth-Generation Tensor Cores: Delivers up to 4,000 AI trillion operations per second and adds support for FP4 precision and NVIDIA DLSS 4 Multi Frame Generation, enabling a new era of AI-powered graphics and the ability to run and prototype larger AI models faster.
- Larger, Faster GDDR7 Memory: Boosts bandwidth and capacity — up to 96GB for workstations and servers and up to 24GB on laptops. This enables applications to run faster and work with larger, more complex datasets for everything from tackling massive 3D and AI projects to exploring large-scale virtual reality environments.
- Ninth-Generation NVIDIA NVENC: Accelerates video encoding speed and improves quality for professional video applications with added support for 4:2:2 encoding.
- Sixth-Generation NVIDIA NVDEC: Provides up to double the H.264 decoding throughput and offers support for 4:2:2 H.264 and HEVC decode. Professionals can benefit from high-quality video playback, accelerate video data ingestion and use advanced AI-powered video editing features.
- Fifth-Generation PCIe: Support for fifth-generation PCI Express provides double the bandwidth over the previous generation, improving data transfer speeds from CPU memory and unlocking faster performance for data-intensive tasks.
- DisplayPort 2.1: Drives high-resolution displays at up to 4K at 480Hz and 8K at 165Hz. Increased bandwidth enables seamless multi-monitor setups, while high dynamic range and higher color depth support deliver more precise color accuracy for tasks like video editing, 3D design and live broadcasting.
- Multi-Instance GPU (MIG): The RTX PRO 6000 data center and desktop GPUs and 5000 series desktop GPUs feature MIG technology, enabling secure partitioning of a single GPU into up to four instances (6000 series) or two instances (5000 series). Fault isolation is designed to prevent workload interference for secure, efficient resource allocation for diverse workloads, maximizing performance and flexibility.
The NVIDIA RTX PRO 6000 "Blackwell" will have 24,064 cores, 10.5% more than the RTX 5090's 21,760 cores. In addition to the core count, the chip will also pack 752 tensor cores and 188 RT cores. The card will offer up to 125 TFLOPs of FP32 and 4000 AI TOPS worth of performance. But the biggest upgrade over the RTX 5090 will be its insane memory capacity.
Unlike the NVIDIA GeForce RTX 5090, which features 32 GB GDDR7 memory across a 512-bit bus interface, the RTX PRO 6000 "Blackwell" will get 96 GB of GDDR7 (ECC) memory across the same 512-bit bus. The graphics cards will be configured at 28 Gbps speeds, delivering up to 1.8 TB/s of total bandwidth.
As for the TBP, the NVIDIA RTX PRO 6000 will be rated at 600W, which is the full capacity allowed by a single 12V-2x6 16-pin power interface. Cooling such a beastly card would require loads of work and NVIDIA's thermal engineering team has repurposed the dual-fan and dual-slot cooler to meet the needs of this card.
But this is the cooling for the standard variant. The other two variants will serve different needs. The RTX PRO 6000 "Blackwell" Max-Q will come with a blower-style design and is aimed at the work-station segment, with its dual-slot and blower-fan design. The other variant is designed for server environments and will use a passive cooling solution. The Max-Q variant will operate at a TDP of 300W.
Lastly, in terms of pricing, expect this workstation behemoth to cost anywhere around $10-$15K or even higher. Canadian retailer Direct Dial is currently listing the card for CAD 11,933 or around $8300 US. This is over four times more than the RTX 5090's MSRP, though in reality, the RTX 5090 costs around $3000 US, which still makes the RTX PRO 6000 "Blackwell" over two times more expensive.
NVIDIA has confirmed that the RTX PRO series "Server" GPUs will be available in May while the RTX PRO Workstation variants will be available in April from leading providers. There will also be RTX PRO 5000, 4500, 4000 GPUs arriving in summer.
NVIDIA Workstation Graphics Card Lineup:
| Graphics Card | RTX PRO 6000 | RTX 6000 Ada | RTX A6000 | Quadro RTX 8000 | Quadro RTX 6000 | Quadro GV100 |
|---|---|---|---|---|---|---|
| GPU | Blackwell GPU | Ada Lovelace GPU | Ampere GPU | Turing GPU | Turing GPU | Volta GPU |
| GPU SKU | GB202 | AD102 | GA102 | TU102 | TU102 | GV100 |
| GPU Process | 5nm | 5nm | 8nm | 12nm | 12nm | 12nm |
| Die Size | 750mm2 | 608mm2 | 628mm² | 754mm² | 754mm² | 815mm² |
| GPU Cores | 24064 Cores | 18176 Cores | 10752 Cores | 4608 Cores | 4608 Cores | 5120 Cores |
| Tensor Cores | 752 Cores | 568 Cores | 656 Cores | 576 Cores | 576 Cores | 640 Cores |
| Boost Clock | TBD | 2.50 GHz | 1.80 GHz | 1.77 GHz | 1.77 GHz | 1.62 GHz |
| Single Precision | 125.0 TFLOPs | 91.1 TFLOPs | 38.7 TFLOPs | 16.31 TFLOPs | 16.31 TFLOPs | 16.66 TFLOPs |
| Ray Tracing Spec | TBD | 210.6 TFLOPs | 75.4 TFLOPs | 10 GigaRays/Sec | 10 GigaRays/Sec | N/A |
| VRAM | 96 GB GDDR7 | 48 GB GDDR6 | 48 GB GDDR6 | 48 GB GDDR6 | 24 GB GDDR6 | 32 GB HBM2 |
| NVLINK VRAM | N/A | N/A | 96 GB With NVLINK | 96 GB With NVLINK | 48 GB With NVLINK | N/A |
| Memory Bus | 512-bit | 384-bit | 384-bit | 384-bit | 384-bit | 4096-bit |
| Memory Bandwidth | 1.8 TB/s | 960 GB/s | 768 GB/s | 672 GB/s | 672 GB/s | 870 GB/s |
| TDP | 600W 300W (Max-Q) | 300W | 300W | ~225W | ~200W | 250W |
| Launch Price | TBD | $6800 US | $4650 US | $10000 US | $6300 US | $9000 US |
| Launch Date | 2025 | Q1 2023 | Q4 2020 | Q4 2018 | Q4 2018 | 2018 |
About the author: A Software Engineer by training and a PC enthusiast by passion, Hassan Mujtaba serves as Wccftech's Senior Editor for hardware section. With years of experience in the industry, he specializes in deep-dive technical analysis of next-generation CPU and GPU architectures, motherboards, and cooling solutions. His work involves not only breaking news on upcoming technologies but also extensive hands-on reviews and benchmarking.
Follow Wccftech on Google to get more of our news coverage in your feeds.
