Computer Hardware

NVIDIA RTX PRO Blackwell - 96GB of VRAM per GPU?!

April 4, 2025 • 6 min read

SPC-Blog-NVIDIA-RTX-Pro-Blackwell-96gb-ram-gpu.jpg

Introduction

During GTC 2025, NVIDIA announced the newest generation of professional RTX GPUs, fittingly dubbed the RTX PRO Blackwell Series GPUs for tackling HPC workloads such as AI training/inferencing, engineering simulation, big data analytics, molecular dynamics/life science research, and more. Featuring the all-new Blackwell Architecture, we see a leap in performance and memory capacity.

We will go over the technical specifications of the various GPUs in the family. SabrePC has plans to support the new NVIDIA RTX PRO Blackwell Series GPUs in our workstations, servers, and HPC solutions, selectable in our configurator.

NVIDIA RTX PRO Blackwell Lineup

SabrePC has already put the RTX PRO Blackwell GPUs on our site and will plan to continue to stock these GPUs. Contact our team for lead time availability. The lineup includes:

Here are the specifications for each GPU. We also included last generation RTX 6000 Ada at the end of the table for comparison:

Specification

RTX PRO 6000 Server Edition

RTX PRO 6000 Workstation Edition

RTX PRO 6000 Max-Q Workstation Edition

RTX PRO 5000

RTX PRO 4500

RTX PRO 4000

RTX 6000 Ada

Generation

Blackwell

Blackwell

Blackwell

Blackwell

Blackwell

Blackwell

Ada Lovelace

Host Interface

PCIe 5.0 x16

PCIe 5.0 x16

PCIe 5.0 x16

PCIe 5.0 x16

PCIe 5.0 x16

PCIe 5.0 x16

PCIe 4.0 x16

Standard Memory

96GB GDDR7

96GB GDDR7 ECC

96GB GDDR7 ECC

48GB GDDR7 ECC

32GB GDDR7 ECC

24GB GDDR7 ECC

48 GB GDDR6 ECC

Bus Width

512-bit

512-bit

512-bit

384-bit

256-bit

192-bit

384-bit

Memory Bandwidth

1.6 TB/s

1.8TB/s

1.8TB/s

1.34 TB/s

896.0 GB/s

672.0 GB/s

960.0 GB/s

CUDA Cores

24,064

24,064

24,064

14,080

10,496

8,960

18,176

Tensor Cores

752 5th Gen

752 5th Gen

752 5th Gen

440 5th Gen

328 5th Gen

280 5th Gen

568 4th Gen

Ray Tracing Cores

188 4th Gen

188 4th Gen

188 4th Gen

110 4th Gen

82 4th Gen

70 4th Gen

142 3rd Gen

Single Precision FP32 (TFLOPS)

117.3

125

110

73.7

54.9

46.9

91.06

Power

600W

600W

300W

300W

200W

140W

300W

Cooler Design

Passive

Double Flow Through

Active

Active

Active

Active

Actve

Slot Dimensions

FHFL Dual Slot

5.4H 12"L Dual Slot

FHFL Dual Slot

FHFL Dual Slot

FHFL Dual Slot

FHFL Single Slot

FHFL Dual Slot

Three Different NVIDIA RTX PRO 6000 Blackwell GPUs

This generation NVIDIA has made three different versions of the RTX PRO 6000 Blackwell each with their own unique heatsinks/coolers and performance.

SPC-Blog-NVIDIA-RTX-Pro-Blackwell-96gb-ram-gpu-2.jpg
  • NVIDIA RTX PRO 6000 Blackwell Workstation Edition
    • Utilizes a dual flow-through cooling system identical to the RTX 5090 Founder's Edition design. With its 600W power capacity, it's ideal for high-performance multi-GPU workstations requiring maximum memory and processing power. However, the non-server-standard 10.5" height makes this GPU incompatible with all standard server chassis double-wide slots.
  • NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition
    • The Max-Q features the black and gold active cooler design, maintaining compatibility with standard server and workstation chassis thanks to its FHFL 9.5" dual-slot form factor. While power-limited to 300W, it offers excellent performance for both server and workstation environments.
  • NVIDIA RTX PRO 6000 Blackwell Server Edition
    • Successor to the NVIDIA L40S, the Server Edition uses an all-gold passive cooler design. It has a configurable 600W power draw for servers only allowing for a little more performance headroom than the Max-Q at 300W, but not quite as much as the Workstation Edition at 600W.
  • The Rest of the RTX PRO Blackwell GPUs
    • The RTX PRO 5000, 4500, and 4000 Blackwell all feature the same black active cooler design. Workloads that don't need peak performance but can benefit from the additional VRAM can rejoice as the RTX PRO 5000 comes with 48GB while the RTX PRO 4500 and RTX PRO 4000 come with 32GB and 24GB of GDDR7 ECC memory, respectively.

Takeaways and Conclusions

The NVIDIA RTX PRO Blackwell lineup of GPUs’ biggest upgrade is the additional memory. As workloads trend towards larger and larger model sizes, the RTX PRO Blackwell 96GB of GDDR7 memory is a big advantage over the last generation RTX 6000 Ada 48 GB. With 96GB of memory, here are some workloads that will benefit greatly:

  • Running Local Complex AI Models: Businesses or even enthusiasts that deploy local LLMs and Agentic AI models can now run larger parameter foundational models on a single GPU. Models that required two RTX 6000 Ada due to memory constraints can now be consolidated to run on a single RTX PRO 6000 Blackwell.
  • High Cell Count Molecular Dynamics: Molecular dynamics simulations like AMBER and GROMACS don’t have multi-GPU support for the same simulation calculation, limiting the cell count of specific simulation use cases unless you upgrade GPUs. Serious research labs that want to perform MD simulations on extremely large systems need to upgrade to the expensive data center GPUs H100 and H200. With additional memory of the RTX PRO 6000, researchers can evaluate even larger-scale molecular models, or opt for the more cost-effective RTX PRO 5000 Blackwell now with 48GB GDDR7.

SabrePC will support the NVIDIA RTX PRO Blackwell series of GPUs in our Workstation and Server offerings. Customize your systems for your specific workload in our intuitive configurator found on our Workstation page or Servers Page. Contact our team for more information on how we can help.

SPC-Blog-Banner-ad-deep-learning-solutions.png

Tags

rtx pro

gpu

nvidia

blackwell



Related Content