NVIDIA spotlights GPU dominance in era of AI and supercomputers
Its GPU-based accelerated computing platform has displaced CPUs across AI, science and industry.
NVIDIA’s accelerated computing platform, built around GPUs, has displaced CPUs as the primary engine for AI, scientific computing and large-scale inference.
According to the company, CEO Jensen Huang highlighted the shift at SC25 and said that within the TOP100 list of supercomputers more than 85% of systems now use GPUs, marking a move away from serial CPU architectures.
The company cited Green500 results showing the top five GPU systems averaged 70.1 gigaflops per watt versus 15.5 flops per watt for top CPU-only systems, a 4.5x efficiency gap. It also reported a Graph500 record of 410 trillion traversed edges per second using 8,192 H100 GPUs to process a graph with 2.2 trillion vertices and 35 trillion edges, while the next best result required roughly 150,000 CPUs. “The world has a massive investment in non-AI software. From data processing to science and engineering simulations, representing hundreds of billions of dollars in compute cloud computing spend each year,” Huang said on NVIDIA’s recent earning call.
The company said its platform is a full stack that pairs GPUs with networking, CUDA libraries, memory, storage and orchestration. It highlighted a Snowflake integration that adds A10 GPUs and preinstalled cuML and cuDF libraries, and reported benchmark gains of 5x less time for Random Forest and up to 200x for HDBSCAN on A10 GPUs versus CPUs. The firm said its platform delivered the highest performance on every MLPerf Training test and was the only platform to submit on all tests, and that NVIDIA systems run on leading generative models and handle 1.4 million open-source models. The company described three scaling laws—pretraining, post‑training and test‑time scaling—as the roadmap for AI workflows.
The company showcased Project GR00T for humanoid robotics and said a robotics breakthrough is expected within years. It cited a Morgan Stanley estimate that there could be 1 billion humanoid robots generating $5 trillion in revenue by 2050.
The Recap
- NVIDIA GPU platform has replaced CPUs across supercomputing workloads.
- Top five Green500 systems average 70.1 gigaflops per watt.
- Robotics breakthroughs expected within years, the company said.
