NVIDIA Surpasses AMD and Intel with MCM Design

There have been many leaks and rumors about NVIDIA's upcoming Blackwell architecture graphics card. According to leaked information, the new architecture will adopt a multi-chip module (MCM) design to improve processor performance.

Although the specific specifications and features of the Blackwell card are still shrouded in mystery, a hardware leaker using the username kopite7kimi on the X forum has revealed some information for us.

kopite7kimi stated that the GB100 high-performance computing (HPC) chip in the Blackwell architecture will adopt MCM design. This means that GB100 will consist of multiple chip modules working together, rather than relying on a single monolithic chip. While GB100 is intended for data centers, the same MCM design will also be applied to GB200, a consumer-level chip for gaming products, which will be used in NVIDIA's next-generation flagship RTX50 gaming graphics card.

This is not the first time NVIDIA has been rumored to release MCM design chips. The Grace Hopper 100 (GH100) was also expected to adopt MCM design but ultimately opted for a monolithic structure.

NVIDIA's approach to achieving this?

The MCM approach integrates multiple chip modules into a GPU package to improve performance and scalability. NVIDIA's decision to explore this design is a significant breakthrough from their traditional monolithic structure GPU approach. All of this is expected to start with the GB100 HPC chip, originally planned for release in 2026 but now expected to be released before the end of 2025.

The driving force behind this shift is clear: unlocking greater performance potential. Monolithic structure GPUs are limited by the physical constraints of the manufacturing process, which determine the maximum chip size. As transistors become increasingly difficult to shrink further, the only option to enhance GPU performance is to create a larger chip.

This is where chip modules come into play. By dividing the GPU into smaller chip modules and interconnecting them, a single GPU can effectively operate as a larger unit within the manufacturing process's typical allowance. This approach offers significant potential for performance improvement, addressing the challenges posed by the end of Moore's Law, where transistor density on chips doubles approximately every two years.

Another X user shared a rumor that NVIDIA is also testing Samsung's 3nm chips. This suggests that NVIDIA may use these chips to build MCM-designed chips. These chips may also enter mass production starting in 2025. While there is debate over whether TSMC's 5nm chips are better than Samsung's 3nm chips, MCM design will still benefit from smaller sizes and form larger overall architectures.

The future of multi-chip modules (MCM)

It is worth noting that NVIDIA is not the first manufacturer in the GPU industry to explore MCM design. AMD and Intel have already made significant progress in this direction and have benefited from increased performance and scalability. In particular, AMD has achieved notable success with its Ryzen processors and Radeon graphics cards (which adopt chip module architecture) and the success of Instinct MI300.

If NVIDIA adopts MCM design in its Blackwell architecture, it could mark a turning point in the GPU industry. It is worth noting that this shift is not limited to the data center market but could also extend to consumer-level cards, including the RTX 50 series, providing higher performance within smaller form factors.

However, kopite7kimi also stated in another article that the core count will not increase, and only structural improvements will make a difference.

Although the specific details of the Blackwell card are still speculative, the potential for performance and scalability achieved through MCM technology is undoubtedly exciting. NVIDIA's competitors have already demonstrated the advantages of this approach, making it a logical step for the company to consider, if not a necessary one.

While these developments are still rumors and speculation, the precedents set by industry leaders such as AMD and Intel indicate that MCM technology is key to unlocking unparalleled GPU performance and scalability. For NVIDIA, a shift towards MCM may mark a pivotal moment, ensuring the continued development of its leading position in the GPU industry in the coming years.