Grace Blackwell

Extracted Attributes

• Type

  GPU Microarchitecture

• Developer

  Nvidia

• Named After

  David Blackwell (statistician and mathematician)

• Successor To

  Ada Lovelace microarchitecture

• Key Components

  Blackwell GPUs, Grace CPUs (Arm architecture)

• Primary Applications

  AI training, AI inference (large language models), robotics (Physical AI), electronic design automation (EDA)

• Interconnect Technology

  NVIDIA NVLink-C2C

• GB200 NVL72 Configuration

  36 GB200 Grace Blackwell Superchips (36 Grace CPUs, 72 Blackwell GPUs) in a rack-scale design

• GB200 NVL72 Inference Speed

  30X faster for trillion-parameter large language models

• GB200 Superchip HBM3e Memory

  Up to 372 GB

• GB10 Superchip Unified System Memory

  128 GB

• GB10 Superchip Model Parameter Support

  Up to 200 billion parameters

• GB10 Superchip AI Performance (FP4 precision)

  Up to 1 petaflop

• GB200 Superchip FP8 Compute (without sparsity)

  10 petaflops

Timeline

• The name of the Blackwell architecture was leaked or rumored. (Source: User Summary, Wikipedia)

  2022

• B40 and B100 accelerators based on Blackwell were confirmed in an official Nvidia roadmap during an investors presentation. (Source: User Summary, Wikipedia)

  2023-10

• The Blackwell microarchitecture was officially announced at Nvidia's GTC 2024 keynote. (Source: User Summary, Wikipedia)

  2024-03-18

• Nvidia began mass-producing Grace Blackwell systems, with reports of 1,000 GB200 systems per week. (Source: Web Search)

  Ongoing

• Synopsys and Nvidia are collaborating to accelerate electronic design automation (EDA) workloads using the Grace Blackwell platform, projecting up to 30x speedup for certain tasks. (Source: Web Search)

  Ongoing

Blackwell (microarchitecture)

Blackwell is a graphics processing unit (GPU) microarchitecture developed by Nvidia as the successor to the Hopper and Ada Lovelace microarchitectures. Named after statistician and mathematician David Blackwell, the name of the Blackwell architecture was leaked in 2022 with the B40 and B100 accelerators being confirmed in October 2023 with an official Nvidia roadmap shown during an investors presentation. It was officially announced at Nvidia's GTC 2024 keynote on March 18, 2024.

Web Search Results

• Nvidia just dropped a game-changer — CEO Jensen ...

  Image 2: Whatsapp Follow ChannelFollow us Share "Email this article") Font Size Abc Small Abc Medium Abc Large Save Print Comment Synopsis Nvidia's CEO Jensen Huang introduced the GB200 Grace Blackwell superchip at VivaTech 2025, calling it a technological miracle. This superchip combines two Blackwell B200 GPUs and a Grace CPU, designed for both training and inference of large language models. Nvidia is now mass-producing these systems at an unprecedented scale of 1,000 per week. [...] Image 3: Nvidia just dropped a game-changer — CEO Jensen Huang calls it a miracle for AI supercomputing Nvidia CEO Jensen and co-founder Huang recently unveiled a “miracle” in AI and technology, the GB200 Grace Blackwell superchip at the GTC Paris Keynote during VivaTech 2025, as reported by Barchart. Nvidia's Technological Milestone [...] He emphasized that it was a milestone achievement for his company and even for the broader tech industry, saying, “Grace Blackwell systems. It is really a miracle. It's a miracle from a technology perspective,” as quoted in the report. Mass Production on a New Level

• New Amazon EC2 P6e-GB200 UltraServers accelerated by ...

  The NVIDIA Grace Blackwell Superchips connect two high-performance NVIDIA Blackwell tensor core GPUs and an NVIDIA Grace CPU based on Arm architecture using the NVIDIA NVLink-C2C interconnect. Each Grace Blackwell Superchip delivers 10 petaflops of FP8 compute (without sparsity) and up to 372 GB HBM3e memory. With the superchip architecture, GPU and CPU are colocated within one compute module, increasing bandwidth between GPU and CPU significantly compared to current generation EC2 P5en [...] AWS Blogs ## AWS News Blog # New Amazon EC2 P6e-GB200 UltraServers accelerated by NVIDIA Grace Blackwell GPUs for the highest AI performance | | | Voiced by Polly | Voiced by Polly

• NVIDIA Puts Grace Blackwell on Every Desk and at ...

  Grace Blackwell AI Supercomputing Within Reach With the Grace Blackwell architecture, enterprises and researchers can prototype, fine-tune and test models on local Project DIGITS systems running Linux-based NVIDIA DGX OS, and then deploy them seamlessly on NVIDIA DGX Cloud™, accelerated cloud instances or data center infrastructure. [...] Project DIGITS features the new NVIDIA GB10 Grace Blackwell Superchip, offering a petaflop of AI computing performance for prototyping, fine-tuning and running large AI models. With Project DIGITS, users can develop and run inference on models using their own desktop system, then seamlessly deploy the models on accelerated cloud or data center infrastructure. [...] The GB10 Superchip is a system-on-a-chip (SoC) based on the NVIDIA Grace Blackwell architecture and delivers up to 1 petaflop of AI performance at FP4 precision.

• Synopsys Accelerates Chip Design with NVIDIA Grace ...

  Synopsys and NVIDIA are advancing a multi-year collaborative effort to accelerate electronic design automation (EDA) workloads. Synopsys is further applying NVIDIA accelerated compute architectures, including the NVIDIA GB200 Grace Blackwell Superchip, to achieve significant, projected runtime gains for workflows including circuit simulation, computational lithography, Technology Computer-Aided Design (TCAD), physical verification, and materials engineering. These accelerated workflows include: [...] SUNNYVALE, Calif., March 18, 2025 /PRNewswire/ -- Synopsys Inc. (Nasdaq: SNPS) today announced the next phase of its work with NVIDIA to accelerate chip design up to 30x with the NVIDIA Grace Blackwell platform. [...] Circuit Simulation: Synopsys PrimeSim™ SPICE simulation workloads are projected to achieve a 30x speed up utilizing the NVIDIA Grace Blackwell platform. Today, customers can achieve up to 15x speed up utilizing NVIDIA GH200 Superchips. NVIDIA accelerated computing architectures enable the simulation of challenging circuits to achieve signoff with SPICE-level accuracy, reducing runtimes from days to hours.

• NVIDIA Blackwell Architecture

  DGX Spark brings the power of NVIDIA Grace Blackwell™ to developer desktops. The GB10 Superchip, combined with 128 GB of unified system memory, lets AI researchers, data scientists, and students work with AI models locally with up to 200 billion parameters. Learn More ## Unlock Real-Time, Trillion-Parameter Models With the NVIDIA GB200 NVL72 [...] ### NVIDIA DGX Station Unlike any AI desktop computer before, this system features NVIDIA Blackwell GPUs, the Grace CPU Superchip, and large coherent memory, delivering unparalleled compute performance. Learn More ### NVIDIA DGX Spark A compact, personal AI supercomputer with the NVIDIA GB10 Grace Blackwell Superchip, delivering high-performance AI capabilities and support for models up to 200 billion parameters. Learn More ### NVIDIA HGX B300 [...] The NVIDIA GB200 NVL72 connects 36 GB200 Grace Blackwell Superchips with 36 Grace CPUs and 72 Blackwell GPUs in a rack-scale design. The GB200 NVL72 is a liquid-cooled solution with a 72-GPU NVLink domain that acts as a single massive GPU—delivering 30X faster real-time inference for trillion-parameter large language models. Learn More About the NVIDIA GB200 NVL72 ## NVIDIA NVFP4 Technical Blog