At NVIDIA GTC 2025, the company announced the NVIDIA RTX Pro 6000 Blackwell series. This is not called the B40, like the A40, L40, and similar cards were. Instead, the “Pro” tells us that we have a card designed for professional workloads. This new RTX Pro 6000 Blackwell series comes in three variants to span desktop workstations to server applications.
NVIDIA RTX Pro 6000 Blackwell Family for Desktops and Servers
We are calling this a “family” rather than a single GPU because there are three distinct variants. The Server Edition is the passively cooled successor to the NVIDIA L40, NVIDIA L40S, and NVIDIA A40. The Workstation Edition takes a large cooler more similar to what we would find on a desktop GeForce card and is designed more for having one high-powered GPU in a system. The Max-Q Workstation Edition uses a blower-style cooler design for multiple GPUs, similar to theĀ NVIDIA RTX 6000 Ada Generation.

Each version has 96GB of GDDR7 memory with ECC. Aside from the physical form factors due to cooling, the power consupmtion varies significantly. The Server Edition can be configured from 400-600W. The Workstation Edition is designed to operate as a single 600W GPU. The Max-Q Workstation edition is designed to have sets of four in a workstation.

From what we have been hearing, there is a lot of interest in these cards as the 96GB of memory capacity puts it above the NVIDIA H100 NVL in terms of raw memory capacity. In addition, unlike the H100, H200, B200, and other data-center-only designs, these parts have the RTX cores and are designed for graphics workloads, not just data center (AI) compute. For many who need each GPU in a system to run more than AI 24×7, these along with their relatively lower pricing to the pure data center parts, are becomign very attractive.
Final Words
For the workstation single-GPU market, the Workstation edition looks like a card we have wanted for some time as the RTX 6000 Ada was not a 600W design. Blower-style coolers are still useful in many systems, so the Max-Q variant is always neat. The RTX Pro 6000 Server Edition is changing the power consumption of the 8x GPU systems. Even those PCIe systems, if running the RTX Pro 6000 GPUs at 600W, are now hitting around 10kW, a figure that was reserved for only SXM-based systems not too long ago. Still 768GB of GPU memory in a 4U or 5U server and at a lower price than the HBM parts, is going to be a winning combination.

To our longtime STH readers: We realized that this one was stuck in the draft queue, but we need the announcement piece to link back to for a few pieces in the next few weeks. So, here we are.
Looks cool. But does it have NVLink / NVSwitch? The rack server in the picture doesn’t seem to have one. If there is no interconnection hardware then saying “768GB of GPU” is not valid…
Kim, true, but PCIe5 is quite fast so…