According to the latest data from the first MLPerf benchmark to set the standard for generative AI, NVIDIA H100 Tensor Core GPUs achieved the best AI performance.
The company’s hardware achieved new records on all eight tests in the latest released MLPerf training benchmarks. The highlight is the new test for generative AI.
During testing, a commercially present cluster of 3,584 H100 GPUs operated by CoreWeave and developed by startup Inflection AI was able to complete the GPT-3-based training benchmark in less than a minute via a service provider. in cloud.
“Our customers are building next-generation generative AI and LLMs at scale, thanks to our thousands of H100 GPUs on fast, low-latency InfiniBand networks. Our joint MLPerf submission with NVIDIA clearly demonstrates the excellent performance our customers enjoy.”
Brian Venturo
Co-founder and CTO of CoreWeave
Inflection AI used demonstrated performance to create advanced LLM through the first personal AI, Pi. The company’s intention is to operate as an AI studio to create personal formats aimed at simpler and more natural interaction by users.
“Anyone can experience the power of a personal AI based on our large next-generation language model that was trained on CoreWeave’s powerful network of H100 GPUs.”
Mustafa Suleyman
CEO of Inflection AI
When comparing the data obtained in the MLPerf benchmarks, the H100 GPUs showed the highest performance in indicators such as large language models, recommendations, computer vision, medical imaging and speech recognition.
Furthermore, the NVIDIA hardware is the only chip to pass all eight tests. According to NVIDIA, this capability demonstrates the versatility of its platform.
H100 GPUs have achieved new records of performance at scale for AI training. According to NVIDIA, the near-linear performance scaling in the LLM test was made possible by optimizations across the suite.
“We are delighted to announce that the H100 GPUs have consistently set new records for performance at scale for AI training, which is testament to their excellence in performing at scale.”
Marcio Aguiar
Head of NVIDIA’s Enterprise Division for Latin America
NVIDIA further added that nearly a dozen companies showed results on the company’s platform in this round. Entries from system manufacturers include ASUS, Dell Technologies, GIGABYTE, Lenovo and QCT, with over 30 submissions running on H100 GPUs.
What is your assessment of the benchmarks achieved with NVIDIA components in generative AI? Interact with us!
