NVIDIA introduced the NVIDIA Blackwell platform, which enables generative AI based on large language models (LLMs) with trillions of parameters. And it’s 25 times cheaper and more energy efficient than its previous counterparts.

The Blackwell architecture GPUs contain 208 billion transistors and are manufactured on TSMC’s special 4nm process with two GPU edge arrays connected via a 10Tbps communication channel into a single unified GPU.

The platform operates as a single GPU with 1.4 exascale AI performance and 30 TB of fast memory.

With new microtensor scaling support and NVIDIA’s advanced dynamic range management algorithms integrated into the NVIDIA TensorRT-LLM and NeMo Megatron frameworks, Blackwell will support twice the compute and model sizes with new floating point AI capabilities based on 4-bit computing.

“For three decades we’ve pursued accelerated computing, with the goal of enabling transformative breakthroughs like deep learning and AI,” said Jensen Huang, founder and CEO of NVIDIA. “Generative AI is the defining technology of our time. Blackwell is the engine to power this new industrial revolution. Working with the most dynamic companies in the world, we will realize the promise of AI for every industry.”

The NVIDIA GB200 Grace Blackwell superchip is a combination of the NVIDIA B200 Tensor Core and the NVIDIA Grace CPU.

For the ultimate in AI performance, GB200-based systems can be connected to NVIDIA Quantum-X800 InfiniBand and Spectrum-X800 Ethernet platforms, which provide advanced networking at speeds up to 800 Gbps.

The GB200 platform is a key component of the NVIDIA GB200 NVL72, a liquid-cooled multi-node system designed for the most intensive computing workloads. It combines 36 Grace Blackwell superchips, which include 72 Blackwell GPUs and 36 Grace CPUs, interconnected via fifth-generation NVLink.

Additionally, the GB200 NVL72 includes NVIDIA BlueField-3 data processing units that enable cloud network acceleration, composable storage, security, and elasticity of GPU computing in hyperscale AI clouds.

The GB200 NVL72 delivers up to 30x performance improvement over the same number of NVIDIA H100 Tensor Core GPUs for LLM inference workloads, while reducing cost and power consumption by up to 25x.

Among the many organizations expected to switch to Blackwell are Amazon Web Services, Dell Technologies, Google, Meta, Microsoft, OpenAI, Oracle, Tesla, and xAI.