Nvidia may be widely known for its gaming-centric graphics cards, but behind the scenes, the company is actually making most of its money from its advances in artificial intelligence. In fact, it’s one of the biggest AI companies in the world.
The Nvidia Blackwell B200 is the latest AI-centric GPU from the company, launching at GTC in March 2024. But how does this new chip differ from previous entries, and who is its target audience? We’ll answer all of those questions and more in this explainer.
What is the Nvidia Blackwell B200?
The Nvidia Blackwell B200 is a new GPU chip that specialises in AI performance.
Nvidia has been creating AI chips for a few years now, with the Nvidia H100 being the most recent before 2024. The H100 was based on the Hopper architecture which sat parallel to Ada Lovelace – the architecture used in the Nvidia 40 Series graphics cards for gamers.
Nvidia has now moved onto a new architecture called Blackwell, and the B200 is the very first graphics card to adopt it.
Nvidia has labelled the B200 as the world’s most powerful chip, as it packs a whopping 208 billion transistors. For reference, the current most powerful GeForce graphics card – the Nvidia RTX 4090 – only has 76 billion transistors, emphasising the absurd power on offer with the B200.
If that’s not enough power for you, Nvidia has also upgraded its NVLink technology to allow more GPUs to communicate with each other in a data centre setup. Nvidia claims that the new 5th Generation NVLink now allows high-speed communication for up to 576 GPUs.
Of course, this is overkill for the average person. The Nvidia Blackwell B200 is instead intended for companies rather than individuals, with the likes of Amazon Web Services, Dell Technologies, Google, Meta, Microsoft, OpenAI, Oracle, Tesla and xAI all confirmed to adopt the AI chip.
The Blackwell GPU architecture is also expected to be used in the upcoming generation of GeForce graphics cards, with the RTX 5000 Series (including the flagship RTX 5090) tipped to become the most powerful chips for gaming.
What is the Nvidia Blackwell GB200 superchip?
The Nvidia Blackwell GB200 was also announced at GTC, which is essentially the combination of two G200 chips, using the 900GB/s NVLink technology to connect them together.
Nvidia claims the GB200 is capable of delivering 30x the performance for LLM inference workloads when compared to the preceding H100. What’s more, Nvidia claims the new superchip is far more efficient, reducing the cost and energy consumption by up to 25x.
