All articles tagged with #h100
Intel has launched its Gaudi 3 AI accelerator, aiming to compete with Nvidia’s H100. The Gaudi 3 is claimed to offer 50% faster time-to-train across Llama2 models with better efficiency and at a fraction of the cost of the H100. It will be available to companies like Dell, HPE, and Lenovo in the second quarter of this year.
AMD has responded to NVIDIA's H100 TensorRT-LLM benchmarks by showcasing that its Instinct MI300X GPU still outperforms the H100, even when the H100 is running optimized software. AMD claims that the benchmarks used by NVIDIA were not using the optimized libraries and compared different data types, resulting in a skewed comparison. With the latest figures, the MI300X running on vLLM offers 30% faster performance than the H100 running on TensorRT-LLM. The competition between AMD and NVIDIA in the AI space is expected to continue as both companies strive to innovate and excel in offering better capabilities and faster performance.
Nvidia's H100 processors, the company's latest flagship AI chip, are selling for over $40,000 on eBay due to high demand for chips needed to train and deploy artificial intelligence software. The H100 is being used to build large language models (LLMs), which are at the heart of AI applications like OpenAI's ChatGPT. The more powerful chips will be necessary to build even bigger and more data-hungry models than those that are currently available.