根据Stable Diffusion开发商Stability AI的最新报告，英特尔最新的Gaudi 2计算卡在AI作图训练方面展现了显著的性能提升。报告指出，在参数为2B的MMDiT模型下，Gaudi 2计算卡的最高训练速度可达每秒1254张图片，而在256容量下则为每秒927张。相比之下，英伟达的H100-80GB计算卡的训练速度为每秒595张，A100-80GB则为每秒381张。这意味着Gaudi 2的训练速度在相同容量下比H100计算卡提升了55%，是A100性能的2.43倍。这一结果对于需要高效处理大量AI图像数据的行业来说，无疑是一个重要的进步。
Title: Intel Gaudi 2 Outperforms Nvidia H100 in AI Graphic Training Speed
Keywords: AI Chip Performance, Compute Card Comparison, Training Efficiency
News content:
According to a recent report by Stability AI, the developer of Stable Diffusion, Intel’s latest Gaudi 2 compute card has shown significant performance improvements in AI graphic training. The report indicates that at a model parameter of 2B MMDiT, Gaudi 2’s maximum training speed reaches 1,254 images per second, while at a capacity of 256, it is 927 images per second. In contrast, Nvidia’s H100-80GB compute card has a training speed of 595 images per second, while the A100-80GB is 381 images per second. This means that Gaudi 2’s training speed is 55% faster than H100’s at the same capacity and 2.43 times the performance of A100. This result is a significant advancement for industries that require efficient processing of large amounts of AI image data.
【来源】https://www.itheat.com/view/45945.html
Views: 1