NVIDIA to deep dive into the Blackwell GPU architecure at Hot Chips 2024 next week

NVIDIA will be hosting a Hot Chips Talk next week, deep diving into its new Blackwell GPU architecture while reminding the world that its Blackwell GPU has the highest AI compute, memory bandwidth, and interconnect bandwidth ever in a single GPU.

11

At Hot Chips 2024 next week, NVIDIA will go into more detail about the Blackwell GPU architecture while also reminding us that it features not one, but two reticle-limited GPUs merged into one. One of the limitations of lithographic chipmaking tools is that they've been designed to make ICs (integrated circuits) that are no bigger than around 800 square millimeters, which is referred to as the "reticle limit".

NVIDIA has two reticle-limited AI GPUs together (104 billion transistors per chip, 208 billion transistors in total). NVIDIA will discuss building to the reticle limit, and how it delivers on the highest communication density, lowest latency, and optimal energy efficiency during its Hot Chips Talk.

Alongside the 208 billion transistors, Blackwell B200 also has 20 PetaFLOPS FP4 AI performance, up to 192GB of HBM3E memory with 8TB/sec of memory bandwidth from its 8-site HBM3E. There's 1.8TB/sec of bidirectional bandwidth through NVLink, and high-speed NVLink-C2C Link to Grace CPU. There's a LOT going on inside of Blackwell, a LOT.

• Read more: NVIDIA's new Blackwell GB200 AI servers 'component shortage' leading to short supply in Q4 2024
• Read more: NVIDIA hits major roadblocks with Blackwell AI GPU: revised B200A AI GPU is now in the works
• Read more: NVIDIA's new Blackwell AI GPUs 'encountering major issues' which require redesign, big delays
• Read more: NVIDIA's next-gen Blackwell AI GPUs delayed, rumor has it 'design flaws' are to blame
• Read more: NVIDIA to make $210 billion revenue from selling its Blackwell GB200 AI servers in 2025 alone
• Read more: NVIDIA says its Blackwell GPUs will be the 'most successful' product in the company's history

11

• Two reticle-limited GPUs merged into one package
• 208B transistors in TSMC 4NP
• 20 Peta FLOPS FP4 AI
• 8 TB/s Memory Bandwidth
• 8-Site HBM3e memory
• 1.8 TB/s Bidirectional NVLINK Bandwidth
• High-Speed NVLINK-C2C Link to Grace CPU

11

• Read more: NVIDIA B200 Blackwell Ultra AI GPU has 288GB HBM3E, redesigned B200A Blackwell Ultra with 144GB
• Read more: NVIDIA places fresh new orders with TSMC for more Blackwell GB200, B100, B200 AI chips
• Read more: NVIDIA's next-gen Rubin, Rubin Ultra, Blackwell Ultra AI GPUs: also supercharged Vera CPUs
• Read more: NVIDIA's full-spec Blackwell B200 AI GPU uses 1200W of power, up from 700W on Hopper H100