At its recent Vision 2024 event, Intel made a significant stride in the artificial intelligence hardware arena by officially unveiling its Gaudi 3 AI accelerator. Designed as a formidable challenger, the Gaudi 3 chip is poised to directly compete with Nvidia’s current industry-leading H100 GPU and its highly anticipated Blackwall B200 series.
Intel asserts that Gaudi 3 offers substantial performance improvements over its predecessor, Gaudi 2, and aims to outperform Nvidia’s H100 in key metrics for large language model (LLM) training and inference. The company highlighted that Gaudi 3 provides 50% better inference throughput and 40% better average power efficiency compared to Nvidia’s H100, alongside a 1.7x boost in training throughput for some LLMs. These claims, if borne out in real-world applications, could offer a compelling alternative for enterprises and cloud providers heavily investing in AI infrastructure.
A key differentiator for Intel is its emphasis on an open software ecosystem, contrasting with Nvidia’s proprietary CUDA platform. Intel hopes that offering greater flexibility and choice through its open standard approach will attract developers and companies looking to avoid vendor lock-in and potentially reduce costs. The Gaudi 3 chip is expected to be available to original equipment manufacturers (OEMs) like Dell, HP, Lenovo, and Supermicro in Q3 2024.
This launch underscores Intel’s renewed commitment to regaining a significant footprint in the data center and AI segments, areas where Nvidia has established a near-monopoly due to the explosion of generative AI. While Nvidia’s lead is substantial, Intel’s aggressive pricing strategy and focus on an open ecosystem could carve out a crucial niche. The competition is heating up, promising more innovation and potentially more accessible AI compute power for the entire industry.
English 
Deutsch