近日，Stability AI公司发布了一份报告，揭示了在人工智能作图训练领域，英特尔的新型Gaudi 2芯片表现出了惊人的性能。该报告显示，在24层深度和混合精度的条件下，Gaudi 2芯片的训练速度可达每秒1254张图片，这一数字在256容量下略降至每秒927张。相比之下，英伟达的H100-80GB芯片的训练速度仅为每秒595张，而上一代的A100-80GB芯片更是只有每秒381张。这意味着，基于英特尔Gaudi 2的训练速度比英伟达H100快了55%，几乎是A100性能的2.43倍。这一突破性的进展或将重塑AI作图领域的竞争格局。
英文标题：Intel Gaudi 2 Outperforms Nvidia H100 in AI Image Generation Training
英文关键词：AI Image Generation, Intel Gaudi 2, Nvidia H100
英文新闻内容：
A recent report by Stability AI has highlighted the impressive performance of Intel’s new Gaudi 2 chip in the field of AI image generation. The report shows that under conditions of 24-layer depth and mixed precision, the Gaudi 2 can train at a speed of up to 1,254 images per second, dropping to 927 images per second at a capacity of 256. In comparison, Nvidia’s H100-80GB chip trains at a speed of only 595 images per second, with the previous-generation A100-80GB chip managing only 381 images per second. This means that training speeds based on Intel’s Gaudi 2 are 55% faster than Nvidia’s H100 and nearly 2.43 times faster than the A100. This breakthrough could reshape the competitive landscape in the AI image generation field.
【来源】https://www.itheat.com/view/45945.html
Views: 1