Nvidia launches new GPUs under the name RTX Pro Blackwell. These cards are the professional equivalent of the previously launched RTX 50 series.
Consistency is not something you can accuse Nvidia of: the GPU specialist can’t launch a new series of professionally oriented GPUs without disregarding the naming conventions of the previous generation. After Quadro and RTX x000 and RTX x000 Ada Generation, there’s now RTX Pro Blackwell. At its GTC convention, Nvidia reveals what these GPUs will look like.
Variations on Variations
Nvidia launches RTX Pro Blackwell for desktop, laptop, and server. The chips are available in different variations and sub-variations. The thousands in the naming give an indication of relative power but are followed by an appendix in the name that indicates for which situation the GPU is suitable.
For example, Nvidia launches an RTX Pro 6000 Blackwell Workstation Edition alongside an RTX Pro 6000 Blackwell Server edition. The complete RTX Pro Blackwell line-up currently looks as follows:
For desktop:
- Nvidia RTX PRO 6000 Blackwell Workstation Edition
- Nvidia RTX PRO 6000 Blackwell Max-Q Workstation Edition
- Nvidia RTX PRO 5000 Blackwell
- Nvidia RTX PRO 4500 Blackwell
- Nvidia RTX PRO 4000 Blackwell
For laptop:
- Nvidia RTX PRO 5000 Blackwell
- Nvidia RTX PRO 4000 Blackwell
- Nvidia RTX PRO 3000 Blackwell
- Nvidia RTX PRO 2000 Blackwell
- Nvidia RTX PRO 1000 Blackwell
- Nvidia RTX PRO 500 Blackwell
For server:
- Nvidia RTX PRO 6000 Blackwell Server Edition
For clarity: all these chips are professional equivalents of the RTX 50 series, which is also based on Blackwell. These RTX 50 chips are also suitable for professional use. What distinguishes these workstation GPUs is mainly the amount of memory. Why Nvidia doesn’t choose RTX 50xx Pro in the naming is very unclear.
96 GB GDDR7
The top model Nvidia RTX Pro 6000 Blackwell Workstation Edition comes with, for example, 96 GB of GDDR7 memory and a memory bandwidth of 1,792 GBps over a 512-bit interface. Nvidia claims 4,000 AI TOPS of AI-specific computing power and 125 TFLOPS of more general single-precision performance. For this, the card needs a substantial 600 watts of power.
The Nvidia RTX Pro 6000 Blackwell supports MIG, which implies that you can split them into smaller virtual GPUs with four times 24 GB memory, or two times 48 GB. The difference with the Server Edition is not immediately clear. The basic specifications are at least identical, although you can configure the server variant with a lower TDP (up to 400 watts) according to the specs.
Looking at the further specifications of the desktop cards, we see 48 GB GDDR7 for the RTX Pro 5000 Blackwell, 32 GB for the RTX Pro 4500 Blackwell, and 24 GB for the RTX Pro 4000 Blackwell.
What About the Laptop?
How the laptop configurations will look is not so clear. Nvidia shares GPU names for laptop and desktop, but for the RTX 50 series, there is indeed a different chip behind those names. That seems to be the case again for RTX Pro Blackwell.
For example, the Nvidia RTX 5000 Pro Blackwell has 48 GB GDDR7 on board and requires 300 watts of power. For the laptop GPUs, the maximum amount of available memory would currently be limited to 24 GB. The TDP will probably also be significantly lower. For indication: an RTX 5080 for desktop has a TDP of 360 watts and drives 10,752 graphics cores, the RTX 5080 in the laptop manages with 80 watts and 8,192 cores. Both GPUs share a name, but little else.
read also
Blackwell Ultra, Vera Rubin and Feynman: Nvidia Unveils GPU Roadmap
We don’t know the memory configurations for laptop-specific models like the RTX Pro 3000 Blackwell at this time.
Competition from AMD
For desktop and server, Nvidia seems to be taking the leading position again here. The manufacturer translates the performant Blackwell architecture into powerful chips tailored for heavy graphics and AI-related workloads.
For laptops, the story is less clear. Here, Nvidia competes with AMD Ryzen AI Max Pro, or Strix Halo among friends. This is an APU with a graphics component that certainly can’t match what Nvidia is launching today purely based on computing power. However, Strix Halo uses a unified memory architecture, giving the GPU access to up to 96 GB of RAM: as much as Nvidia’s top model RTX 6000 Blackwell and considerably more than the 24 GB for Nvidia’s laptop cards today.
read also
AMD surprises with Ryzen AI Max Pro: workstation on a chip
For certain (AI) workloads, memory is more important than absolute GPU computing power. A Strix Halo laptop will thus be able to handle larger AI models than a device with the new Nvidia RTX 5000 Blackwell.
Availability
An additional question is availability: Nvidia’s RTX 50 series launched primarily on paper, with only a handful of GPUs available per retailer and long waiting lists. For these Blackwell chips, the chip specialist is mainly looking at manufacturers of desktops and mobile workstations.
Expect devices from all major professional PC manufacturers, including Dell, HP and Lenovo. For the server variant, Dell and Lenovo are ready again, along with HPE, Cisco and SuperMicro.