Intending to launch a new era of accelerated computing and enable businesses to develop and execute real-time generative AI on trillion-parameter big language models, NVIDIA has unveiled its next-generation Blackwell GPU architecture.
Comparing the Blackwell platform to its predecessor, the Hopper architecture, is expected to result in cost and energy savings of up to 25 times. The new GPU architecture, which bears the name of the influential statistician and mathematician David Harold Blackwell, brings six game-changing innovations.
“The technology that defines our era is generative AI. According to Jensen Huang, the founder and CEO of NVIDIA, Blackwell is the engine driving this new industrial revolution. “We will realize the potential of AI for every industry by collaborating with the world’s most active companies.”
A second-generation Transformer Engine that supports twice the compute and model sizes, a fifth-generation NVLink interconnect for high-speed multi-GPU communication, the most powerful chip in the world (with 208 billion transistors), and advanced engines for data decompression, security, and reliability are some of Blackwell’s key innovations.
The NVIDIA GB200 Grace Blackwell Superchip, which combines a Grace CPU and two B200 Tensor Core GPUs via an incredibly quick 900GB/s NVLink link, is the brains of Blackwell. Systems like the liquid-cooled GB200 NVL72 platform, which has up to 72 Blackwell GPUs and 36 Grace CPUs and provides 1.4 exaflops of AI capability, can be built using multiple GB200 Superchips.
Major cloud providers including Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure have already agreed to support NVIDIA’s plan to offer Blackwell-powered instances. Dell Technologies, Meta, Microsoft, OpenAI, Oracle, Tesla, and numerous other partners in the hardware, software, and sovereign cloud spaces are also preparing Blackwell products.
“We are fortunate to have a longstanding partnership with NVIDIA, and look forward to bringing the breakthrough capabilities of the Blackwell GPU to our Cloud customers and teams across Google to accelerate future discoveries,” stated Sundar Pichai, CEO of Alphabet and Google.
Engineering, semiconductor design, scientific computing, and generative AI are just a few of the fields that will witness new advancements made possible by the Blackwell architecture and software stack.
“AI already powers everything from our large language models to our content recommendations, ads, and safety systems, and it’s only going to get more important in the future,” said Mark Zuckerberg, the founder and CEO of Meta.
“We’re excited to use NVIDIA’s Blackwell to support the training of our open-source Llama models and the development of the upcoming Meta AI and consumer products.”
Blackwell has the potential to be the driving force behind the eventual commercialization of real-time trillion-parameter AI thanks to its enormous efficiency and performance benefits.