Why multi-chip GPUs are the future of graphics power
Multi-chip module (MCM) graphics cards may well be the future, taking next-generation GPUs to a whole new level. Combining multiple dies instead of relying on a single chip could provide better performance without being hindered by current hardware limitations.
According to recent leaks, AMD and Intel may currently be working on MCM graphics cards. AMD’s interpretation of this technology may be imminent – in upcoming RDNA 3 GPUs.
Patent: Position-Based Rendering Apparatus and Method for Multi-Die/GPU Graphics Processing – Intel
Intel MCM GPUs are coming…
More details: https://t.co/GIkfwrXGzV pic.twitter.com/sXGt9nbJ1S
— Underfox (@Underfox3) February 3, 2022
The term “multi-chip module” describes the architecture used to assemble electronic components, such as semiconductor chips or other types of chips. MCM involves integrating multiple GPU modules into a single package. This could potentially result in significantly more GPU power without significantly increasing die size, hopefully improving manufacturability and power consumption.
Today’s leaks come from two different sources and both point (somewhat strongly) to AMD and Intel who are both exploring MCM architecture for their upcoming consumer graphics cards. A new Intel patent has been discovered by sub fox on Twitter and later shared by Wccftech, talking about using multiple GPUs combined via MCM for image rendering and the benefits it would bring.
Intel’s patent talks about using tile-based checkerboard rendering to achieve much more efficient scaling on multi-chip graphics cards. While it’s unlikely (if not impossible) to see this technology in first-generation Intel Arc Alchemist graphics cards, we might see Intel’s MCM GPUs in one of its next-gen discrete graphics cards.
The AMD leak (first shared by PCGamer) comes from an unsuspecting AMD engineer who shared his current projects on his LinkedIn profile and was spotted by blueispurple on Twitter. The engineer is a senior technical staff member and works on Infinity Fabric. The information taken from the profile strongly indicates that some of the next generation AMD RDNA 3 graphics cards will feature an MCM design.
that’s why on some of my previous twit
i said rdna3 would probably start with 5nm
amd pssstlinkedin pic.twitter.com/ZfdfrvgwTO
— blue nugroho (@blueisviolet) February 4, 2022
Multi-chip module GPU technology is something Intel and AMD have dabbled in before, but Nvidia is no stranger to architecture either. As early as 2017, Nvidia published an article titled “Multi-chip-module GPUs for continuous performance scalability”. Since then, rumors have emerged indicating that Nvidia’s next-gen Hopper graphics cards will feature multi-chip GPU designs.
AMD has released an MCM graphics card before, the monstrous Instinct MI200 HPC GPU, designed for high-performance applications (such as data centers). The card offers up to 3.2 TB/s of bandwidth, 128 GB of HBM2e memory and up to 14,080 processors. However, it is likely that MCM technology will now enter the mainstream market with AMD’s upcoming RDNA 3 graphics cards. Intel also has a data center MCM GPU in the works. Called Ponte Vecchio, the card has not yet been released.
The MCM architecture certainly has a lot to offer, and while we’re not quite there yet, it looks like we may soon start to see its benefits in the mainstream market as part of the best graphics cards. AMD RDNA 3 graphics cards are expected to launch later this year. The same can be said of Intel Arc Alchemist, and while we don’t have specific release dates for either, Intel is expected to launch the first Alchemist GPUs before the end of this quarter. Meanwhile, AMD is also rumored to release a refresh of previous RDNA 2 cards in the coming months.