8 JANUARY 2025IN MYPINIONIN MYPINIONNEXT-GENERATION AI HARDWAREBy Dr. Nikolai Ardey, Executive Director of VW Group Innovation, Volkswagen AGArtificial intelligence (AI) is exploding. Machine learning (ML) models are mutating, getting bigger, and more complicated. Some of the emerging ML models are beginning to replace current models, potentially rendering current hardware (HW) obsolete. What HW do we need for our cars? Today, we have three basic options: CPUs, GPUs, and ASICs. CPUs, designed for general-purpose computing, are flexible but not powerful enough for AI. GPUs, designed to render triangles, are powerful but not efficient enough for AI. ASICs, by definition, can be hardcoded for specific ML models, making them both powerful and efficient enough for AI. However, they may not be flexible enough to adapt to future AI approaches. AI needs HW that is the best of all three: as flexible as a CPU, as powerful as a GPU, and as efficient as an ASIC.Early AI started on CPUs. Back then, implementing ML models was straightforward, and any ML models coded up for CPUs would just run faster every year. That was until 2005 when CPUs could not be clocked any faster due to heat. This forced AI to transition from CPUs to GPUs. GPUs are similar to CPUs, except they are massively parallel loading thousands of numbers from external memory at a time. This is powerful but not efficient since each trip to external memory requires a tremendous amount of energy. This newfound power of the GPU enabled a whole new class of ML models called Convolutional Neural Networks (CNNs).As these CNNs became popular, AI started transitioning from GPUs to ASICs. An explosion of chips started hitting the market specifically designed for CNNs by hardcoding the dataflow to keep the data on the chip, thus avoiding
< Page 7 | Page 9 >