Blackwell Architecture

NVIDIA Blackwell is a next-generation GPU architecture featuring 208 billion transistors and a second-generation Transformer Engine purpose-built for trillion-parameter AI models.

Blackwell redefines large-scale computing through six core technologies: a custom 4NP TSMC process, fifth-generation NVLink (1.8TB/s bidirectional throughput), and dedicated decompression engines. The dual-die design functions as a single unified GPU, delivering 20 petaflops of FP4 precision to slash LLM inference costs and energy consumption by up to 25x compared to the H100. By integrating a RAS Engine for preventative maintenance and secure enclave capabilities, Blackwell provides the hardware foundation for the next frontier of generative AI and real-time data processing.

https://www.nvidia.com/en-us/data-center/blackwell/