NVIDIA’s next-generation Blackwell GB100 GPUs for HPC and AI customers are rumored to be fully compatible with a microchip design based on Copite7kimi.

NVIDIA is rumored to be going all-in on the Chiplet train with its next-generation Blackwell AI GPUs, the GB100, to deliver big changes.

The latest rumors indicate two things, the first being that NVIDIA is now expected to use its first chipset design for the modern data center segment. So, as a summary, the NVIDIA Blackwell GPUs were initially expected to be the first of the family to go the microchip route until the company was rumored to have decided against it and was supposed to use a more standard monolithic design. Chiplet and Monolithic designs have their advantages and disadvantages, but given the cost and efficiency required to achieve performance improvements today, Chiplet and other advanced packaging technologies are used by competitors such as AMD and Intel.

NVIDIA GPU Roadmap. (Image source: Nvidia)

So far, NVIDIA has proven that the industry can move forward without using the microchip with its Hopper and Ada Lovelace GPUs excelling at delivering the best performance per watt and highest profit margins the company has ever seen. But moving forward, this will change, and with the start of Blackwell, we may see NVIDIA’s first integrated chipset design. Blackwell GPUs are so far scheduled to be released in 2024 for the data center and AI sector.

Kopite7kimi focuses on data center and AI-powered GPUs when talking about Blackwell. This shows that NVIDIA may not yet be moving to small chips for gaming GPUs codenamed “Ada-Next” but is incorporating a level of distinct packaging technologies within its data center and AI GPUs to maximize chip production. As we mentioned above, small chips have their drawbacks, and those drawbacks usually lie in using the right factories to package these chips.

TSMC’s CoWoS is one of the main packaging technologies available to GPU customers like AMD and NVIDIA but it looks like both companies may be competing to get access to the best of TSMC’s technologies. The battle is usually about who can offer the most money, and the green team is currently swimming in AI money. Furthermore, there are other key components to acquire depending on the level of chipset-based integration that NVIDIA wants to use. Both AMD and Intel are doing some advanced chipset packages, combining multiple IPs into a single chipset package, so it will be interesting to see how far NVIDIA’s design of the first generation microchip architecture advances on Blackwell.

The second part of the story concerns the internal architecture of Blackwell GPUs. It is reported that the number of modules inside Blackwell GPUs such as GPCs and TPCs has not changed much from Hopper but the internal module architecture can indicate that the number of SM/CUDA/Cache/NVLINK/Tensor/RT has changed significantly. Previously, we have seen leaks of at least two GPUs which include the Blackwell GB100 and GB102. The second part could be either a data center or a gaming GPU, but Kopite7kimi has already said that the consumer (gaming) parts will fall under the GB200 series and not the GB100 series.

There is, too Rumors NVIDIA is evaluating Samsung’s 3GAA (3nm) node which may enter mass production in 2025 although Kopite7kimi believes NVIDIA may not change its plans and stick with TSMC for next-gen GPUs. The same leaker previously reported that Blackwell will not be using a 3nm process node. Given the advances in AI and data center that NVIDIA has made since its Pascal GPUs and huge success with its Ampere and Hopper GPUs, Blackwell will represent a major advancement in NVIDIA’s chipset line, propelling the company into the next era of AI and computing.

Leave a Reply

%d bloggers like this: