据Stability AI最新发布的报告显示，在AI作图训练领域，英特尔Gaudi 2计算卡展现出了令人瞩目的性能。在参数为2B的MMDiT模型训练中，采用24层的深度学习和BFloat16混合精度时，Gaudi 2计算卡的训练速度可达到每秒1254张图片。这一数据相比同类产品的性能，彰显了英特尔在AI计算领域的竞争优势。
具体来看，与英伟达当前的H100计算卡相比，Gaudi 2在相同容量下的训练速度有了显著提升。以H100-80GB为例，其训练速度为每秒595张图片，而Gaudi 2则能达到每秒927张，提高了55%。即便是与更早一代的A100-80GB相比，Gaudi 2的性能也是其2.43倍，每秒训练速度可达每秒703张。
这一突破性的进展，不仅体现了英特尔在硬件设计上的创新，也意味着在AI作图这一高计算强度应用场景中，英特尔Gaudi 2计算卡提供了更高效的处理能力。这一性能的提升，将为研究人员、开发者和企业带来巨大的价值，使得他们在图像生成、编辑和处理等AI应用中，能够以更快的速度达到更高的实验和生产效率。
值得注意的是，随着AI技术的不断发展和应用的广泛扩展，计算卡的性能已成为推动AI进步的关键因素之一。Stability AI的这份报告，不仅为市场提供了一个重要的性能比较参考，也为行业竞争格局带来了新的变化指标。未来，英特尔与英伟达在AI计算领域的竞争将更加激烈，而消费者和企业也将能从中获得更多创新和高效的产品选择。
英语如下：
**News Title:** Intel Gaudi 2 Outperforms NVIDIA H100: AI Image Training Speed Increased by 55%
**Keywords:** AI Image Training, Intel Gaudi 2, NVIDIA H100
**News Content:** ### Intel Gaudi 2 Demonstrates Strong Performance, Surpassing NVIDIA H100 in AI Image Training Speed
According to the latest report released by Stability AI, Intel Gaudi 2 computing cards have shown remarkable performance in the field of AI image training. When training the MMDiT model with 2B parameters using a 24-layer deep learning architecture and BFloat16 mixed precision, the training speed of the Gaudi 2 card reaches up to 1,254 images per second. This figure highlights Intel’s competitive edge in AI computing.
Specifically, when compared to NVIDIA’s current H100 computing card, Gaudi 2 shows a significant increase in training speed at the same capacity. For example, the H100-80GB card trains at a rate of 595 images per second, while Gaudi 2 reaches 927 images per second, an increase of 55%. Even when compared to the earlier generation A100-80GB, Gaudi 2 performs 2.43 times better, with a training speed of 703 images per second.
This breakthrough progress not only reflects Intel’s innovation in hardware design but also means that in high-computational-intensity application scenarios like AI image generation, the Intel Gaudi 2 computing card offers more efficient processing capabilities. This performance improvement will bring tremendous value to researchers, developers, and enterprises, enabling them to achieve higher experimental and production efficiency in AI applications such as image generation, editing, and processing.
It is worth noting that with the continuous development of AI technology and the widespread expansion of applications, the performance of computing cards has become one of the key factors driving AI progress. Stability AI’s report not only provides an important performance comparison reference for the market but also brings new changes to the industry’s competitive landscape. In the future, the competition between Intel and NVIDIA in the AI computing field will be more intense, and consumers and enterprises will also be able to benefit from more innovative and efficient product choices.
【来源】https://www.itheat.com/view/45945.html
Views: 2