NVIDIA RTX PRO 6000 “Blackwell”: High-end workstation GPU with massive performance

NVIDIA is now extending its Blackwell architecture to the workstation and server sector with the new RTX PRO 6000. This GPU is based on the GB202 chip and not only offers high computing power, but also an impressive 96 GB GDDR7 memory capacity. The card is explicitly designed for professional applications and data centers and marks a clear differentiation from the RTX 5090 gaming variant.

More cores, more memory, more performance

The RTX PRO 6000 is based on the GB202 chip, which has 24,064 CUDA cores, around 10.5% more cores than the RTX 5090. In addition, there are 752 tensor cores and 188 RT cores, which optimizes the card especially for AI and ray tracing workloads. The pure FP32 performance is 125 TFLOPs, while the card should deliver up to 4000 AI TOPS. The biggest difference to the RTX 5090 is probably the memory: while the gaming version uses 32 GB GDDR7, the RTX PRO 6000 comes with 96 GB GDDR7 ECC – a huge improvement for applications that need to process large amounts of data. The memory runs at 28 Gbps and offers a bandwidth of 1.8 TB/s.

Three variants for different areas of application

NVIDIA is launching the RTX PRO 6000 in three different variants:

1. Standard variant: a classic dual-fan design optimized for workstations, with a TDP of 600W pushing the power limits of the 12V-2×6 16-pin connector.
2. Max-Q variant: A blower-style design intended for professional workstations with limited cooling space. Here the TDP is reduced to 300W.
3. Server variant: A passively cooled variant for rack systems in data centers.

The different cooling concepts make it clear that NVIDIA is not targeting the consumer market with the RTX PRO 6000, but is focusing exclusively on professional users.

New architecture features for more efficiency

NVIDIA has integrated several architecture upgrades into the RTX PRO 6000:

• Fourth generation RT cores with double the ray tracing performance of the previous generation.
• Fifth generation Tensor Cores with FP4 support and up to 4000 AI TOPS, optimized for DLSS 4 multi-frame generation.
• Latest generation NVIDIA NVENC and NVDEC to accelerate professional video processing.
• PCIe Gen 5 for twice the bandwidth of PCIe Gen 4.
• DisplayPort 2.1 with support for 4K@480Hz and 8K@165Hz.
• Multi-Instance GPU (MIG) for partitioning the GPU into up to four isolated instances.

These functions make the card particularly interesting for AI workloads, ray tracing, professional visualizations and scientific calculations.

Price and availability

The RTX PRO 6000 will be available as a workstation GPU from April, while the server version will be released in May. The prices are, unsurprisingly, beyond the gaming market: initial listings show a price of around 8,300 US dollars, and depending on the supplier, it could even be 10,000 to 15,000 dollars. RTX PRO 5000, 4500 and 4000 models will also be released in the summer, which will presumably offer cheaper alternatives for less computing-intensive applications.

Conclusion: A pure professional card without compromises

With the RTX PRO 6000, NVIDIA has once again shown that although the gaming market is important, it is no longer the main sales driver. Data centers, AI applications and workstations are now the main target group for high-end GPUs – and this is exactly where NVIDIA comes in with the RTX PRO 6000 “Blackwell”. This card is neither intended nor affordable for the average user. But for companies working with massive AI workloads, 3D modeling or high-end video production, it should be a real workhorse – at least if the budget plays along.

Source: NVIDIA, Direct Dial