Page 38 - EE Times Europe Magazine – June 2024
P. 38

38 EE|Times EUROPE



         AUTONOMOUS VEHICLES | ARTIFICIAL INTELLIGENCE
        Nvidia GTC 2024: Why Nvidia Dominates AI


        By Egil Juliussen


                 vidia’s Developer Conference has been a major   few years. Each GPU chip has functionality that improves performance
                 event in the GPU industry for more than a    for specific application segments—initially for PC graphics and game
                                                              consoles and recently for autonomous-vehicle functions and AI-centric
                 decade. With the company’s GPUs having       software and systems.
       Nbecome the leading AI chips over the past two           Thus far, the company has introduced a total of 16 GPU microar-
        years, the 2024 Nvidia GPU Technology Conference (GTC)   chitectures, all named after famous inventors and scientists across
                                                              multiple disciplines. The first was the Fahrenheit architecture,
        attracted not just the traditional, engineering-centric   released in 1995. Nvidia’s latest GPU microarchitecture, Blackwell,
        attendees but many Wall Street analysts as well.      debuted in March 2024.
                                                                Table 1 shows how Nvidia’s GPUs have improved since 1995. The
          The AI GPU trend accelerated Nvidia’s market value and revenue   table includes all 16 GPU microarchitectures, with a focus on the
        tremendously in the past two years, and it is now one of the three most   latest versions. The first GPU microarchitecture used in the automo-
        valuable companies in the world based on market capitalization.    tive industry, in 2015, was Maxwell, Nvidia’s ninth GPU architecture,
        Nvidia’s revenue grew from US$10 billion in 2018 to US$61 billion in   released in 2014.
        2023, a spike that explains the influx of Wall Street analysts at    The first Fahrenheit-based GPU had 1 million transistors and was
        GTC 2024.                                             based on a 500-nm manufacturing process technology. By 1999, the
          The growth of generative AI since the launch of ChatGPT has   technology enabled GPU chips with 15 million transistors.
        accelerated the importance of GTC presentations and product   Tesla, the sixth Nvidia GPU microarchitecture, was capable of
        announcements. This article summarizes Nvidia’s key announcements   210 million transistors in 2006 and 1.4 billion transistors in 2010.
        at GTC 2024, offering a perspective that considers Nvidia’s history,   During its four-year manufacturing run, the Tesla GPU chip shrank
        notably the evolution of its GPU chip architecture and technology; the   from 90 nm to 40 nm, even as the transistor count grew 6.7×.
        importance of Nvidia’s CUDA software technology; and why and how   The first Nvidia GPU microarchitecture to offer more than 1 billion
        Nvidia’s GPUs became a key technology for the AI industry.  transistors at introduction was Maxwell, which provided 2.9 billion
                                                              transistors in 2014. Maxwell was also the first Nvidia GPU to be used
        HISTORY OF NVIDIA’S GPU ARCHITECTURE                  in automotive applications. Each new generation added more com-
        The microarchitectures that Nvidia designs for each GPU generation   puting power through computing accelerators for specific calculation
        offer significant improvements with each new version. Multiple GPU   in parallel or simultaneously, which increased the performance of
        chips are designed based on each microarchitecture over a period of a   GPU-based chips.


                                     Table 1: Evolution of Nvidia’s GPU Microarchitecture
         Microarchitecture   Introduction   Microarchitecture   Transistor      Manufacturing      Fab Partner
              Name            Year           Generation          Range           Technology
             Fahrenheit       1995              1st             1M to 15M      250 nm to 500 nm       TSMC
              Celsius         1999              2nd            17M to 25M      150 nm to 220 nm       TSMC
              Kelvin          2001              3rd            29M to 63M          150 nm             TSMC
              Rankine         2003              4th            45M to 135M     130 nm to 150 nm     TSMC, IBM
               Curie          2004              5th            75M to 302M         130 nm             TSMC
               Tesla          2006              6th           219M to 1,400M    40 nm to 90 nm        TSMC
               Fermi          2010              7th           260M to 3,000M       40 nm              TSMC
              Kepler          2013              8th           292M to 7,080M       28 nm              TSMC
              Maxwell         2014              9th            2.94B to 8B         28 nm              TSMC
              Pascal          2016             10th            1.8B to 12B         16 nm              TSMC
               Volta          2017             11th               21.1B            12 nm              TSMC
              Turing          2018             12th            4.7B to 18.6B       12 nm              TSMC
              Ampere          2020             13th            8.7B to 28.3B        7 nm              TSMC
              Hopper          2022             14th               80B               4 nm              TSMC
            Ada Lovelace      2022             15th           18.9B to 76.5B        4 nm              TSMC
             Blackwell        2024             16th               208B              4 nm              TSMC

                                                   (Source: VSI Labs, April 2024)


        JUNE 2024 | www.eetimes.eu
   33   34   35   36   37   38   39   40   41   42   43