NVIDIA appears to be getting ready to make a fundamental jump in the way premium and top-end GPUs are manufactured. Interestingly, the company could adopt the same technique that AMD has successfully used to make powerful Threadripper and Ryzen Series CPUs. The process essentially does away with the single die architecture that NVIDIA and AMD had been using for several decades. Apparently, the next-gen NVIDIA GPU that would succeed the current generation of technology is called Hopper, which is named after Grace Hopper who was one of the pioneers of computer science.
Clearly understanding the increasingly constricting limitations of the current manufacturing process of GPUs and CPUs, AMD advanced to a completely different evolutionary process. Essentially, AMD has moved beyond the ‘Sing Chip Module’, which basically consists of a single processor. With the latest generation of Threadripper and Ryzen CPUs, AMD adopted the multiple dies in a single package methodology, which is also called as Multi-Chip-Module or MCM. The NVIDIA’s Hopper MCM GPU, which is supposedly going to succeed Ampere and will consist of a family of powerful graphics cards with multiple dies in a single package.
After Ampere, the next codename of GeForce is Hopper, in memory of Grace Hopper.
— kopite7kimi (@kopite7kimi) June 11, 2019
NVIDIA Next-Gen Hopper MCM GPU Could Offer Several Enhanced Features On a Single Graphics Card:
An MCM-based design is justifiably the next step in GPU evolution. This is primarily because GPUs and CPUs are now being increasingly limited by the reticle size of most EUV scanners. Simply put, AMD has already successfully migrated to the 7nm fabrication process. Meanwhile, Intel is struggling with 10nm production. The problems with shrinking die sizes are forcing companies to seek external expertise. One of the world’s biggest semiconductor manufacturer, TSMC, is already actively moving on to even smaller sized die. Sooner rather than later, companies may not be able to shrink the size any further.
Understanding the limitations of the current manufacturing processes, and after observing AMD’s successful transition, NVIDIA’s next-gen Hopper GPU could be based on the MCM architecture. Incidentally, GPUs are much more “parallel devices” that process information concurrently, as compared to CPUs, and this step only seems logical. As compared to a monolithic die, MCM based approach should offer massive yield gains at the onset itself.
GeForce of Hopper will use MCM module to build giant cores. pic.twitter.com/2bevDtg6rG
— kopite7kimi (@kopite7kimi) June 12, 2019
Incidentally, NVIDIA’s next-gen Hopper GPU being based on MCM architecture is still a rumor. A tweet that has now been redacted, claimed to reveal NVIDIA’s future plans. Hence it is quite possible that NVIDIA might just stick to the current manufacturing process, simply because the techniques have been finetuned over the last few decades.
Nonetheless, NVIDIA is surely capable of creating an MCM based GPU. Additionally, the company would even get some serious yield benefits if it chooses to run with this for Hopper GPUs. Switching to an MCM based design would allow NVIDIA to build large and powerful GPUs with relatively affordable price tags.
With MCM, AMD basically handed down the humongous processing power of servers and Xeon processors to average desktop users. The new Threadripper and Ryzen Series CPU packs several cores, and still sport an attractive price tag. NVIDIA has always been perceived as the maker of premium or expensive graphics cards. Hence, the company could surely benefit immensely with the MCM approach for its next-gen Hopper GPUs.