Published , by Morgan Shaver
Published , by Morgan Shaver
Technology and artificial intelligence continues its rapid expansion, with companies like Google investing in the creation of things like artificial intelligence supercomputers. Recently, Google published details about one of these AI supercomputers along with assertions that it’s both faster and more efficient than systems seen with its competitors like NVIDIA.
As noted by outlets like CNBC, even though NVIDIA currently dominates the market for AI model training, Google has been hot on its heels deploying AI chips called Tensor Processing Units (TPUs) since 2016. The latest system from Google features over 4,000 TPUs paired with custom components designed to both run and train AI models.
As recent as this technology sounds, it’s been running since 2020 and was even used to help train Google’s PaLM model as a competitor to OpenAI’s GPT. While Google’s latest TPU seems to have an edge over competitors like NVIDIA and its A100, it’s worth keeping in mind that the results weren’t compared to NVIDIA’s latest AI chip, the H100.
According to Google’s researchers, this is because the H100 is both more recent and features more advanced manufacturing technology. Speaking to the power of the H100, NVIDIA CEO Jensen Huang recently stated in a blog post that it delivers 4x the performance of NVIDIA’s A100.
“Today’s MLPerf 3.0 highlights Hopper delivering 4x more performance than A100,” said Huang, as quoted by CNBC. “The next level of Generative AI requires new AI infrastructure to train Large Language Models with great energy-efficiency.
Moving forward, it’ll be interesting to see how Google and NVIDIA are able to compete with each other, and who will come out on top in this competition. For more on Google’s latest TPU, be sure to look over the report shared by researchers. For more on AI, be sure to read through some of our previous coverage, including Italy becoming the first Western country to ban ChatGPT, and Midjourney pausing free trials for AI image generation over deepfake concerns.