Nvidia, a leader in the graphics processing unit (GPU) industry, is gearing up for the release of its next-generation AI chip, the R100, which is anticipated to begin mass production in the last quarter of 2025. According to industry analyst Guo Mingyi, the highly anticipated chip is set to bring significant advancements in AI computing power while also addressing critical energy consumption issues prevalent in AI server operations and data center construction.
The R100 is expected to utilize the N3 process technology from Taiwan Semiconductor Manufacturing Company (TSMC), incorporating the advanced CoWoS-L packaging method, a technique also found in Nvidia's recently launched B100 Blackwell chip. However, the R100 will differentiate itself with a larger interposer and a reticle design roughly four times larger than its predecessor, suggesting substantial improvements in both scale and performance.
Each R100 chip will be composed of four smaller reticles, creating a unified larger processor. This design will also include eight units of the latest HBM4 memory, positioning the R100 as a formidable player in the high-performance computing (HPC) and artificial intelligence sectors. Additionally, Nvidia plans to pair the R100 with its GR200 Grace CPU, which is also set to be produced using TSMC's N3 process technology, an upgrade from the N5 technology currently used in the GH200 and GB200 Grace superchips.
The shift towards improved power efficiency is a strategic move by Nvidia as it responds to growing concerns over the operational costs associated with AI technologies. Previous generations of Nvidia's AI chips, such as the H100, required substantial power, with training a 1.8 trillion parameter model consuming up to 15 megawatts of electricity. In contrast, Nvidia's Blackwell GPUs have dramatically reduced that consumption to just 4 megawatts, illustrating a significant reduction in energy requirements.
This focus on energy efficiency does not come at the expense of processing power. The R100 is poised to continue Nvidia's legacy of rapidly advancing AI capabilities, with the B100 marking a 1000x increase in computing power over an eight-year period. The R100 aims to push these boundaries even further, catering to the intensive needs of modern AI applications from autonomous vehicles to complex data analytics.
