据热点科技报道，人工智能领域再添新里程碑，Stability AI近期发布的一份报告显示，英特尔的Gaudi 2计算卡在AI图像生成训练方面显著优于英伟达的顶级芯片。Stability AI，作为知名稳定扩散模型（Stable Diffusion）的开发者，对两种芯片进行了详尽对比，结果显示Gaudi 2在特定参数设置下展现出惊人的效率。
以2B参数的MMDiT模型为例，训练深度设定为24，并采用BFloat16的混合精度，Gaudi 2计算卡在最大容量下每秒可训练1254张图片，而在256容量下则可达到927张。相比之下，英伟达的H100-80GB计算卡每秒只能处理595张，而A100-80GB的训练速度为381张。这意味着在相同容量下，Gaudi 2的训练速度比H100快55%，性能提升显著；其速度更是A100的2.43倍，展示了英特尔在AI计算领域的技术飞跃。
这一报告的发布，无疑将引发业界对英特尔Gaudi 2芯片的广泛关注，同时也对英伟达在AI计算市场的领先地位构成挑战。随着AI技术的不断发展，计算效率的提升对于加速模型训练、推动人工智能应用的普及至关重要。英特尔Gaudi 2的出色表现，预示着未来AI领域的竞争将更加激烈。
英语如下：
**News Title:** “Intel’s Gaudi 2 Chip Outperforms NVIDIA in AI Image Generation Training with a 55% Speed Boost”
**Keywords:** Intel Gaudi 2, AI Image Generation, Performance Enhancement
**News Content:**
**Title:** Intel’s Gaudi 2 Chip Demonstrates Superior Performance in AI Image Training, Outpacing NVIDIA’s H100
According to HotTech Vision, a new milestone has been reached in the realm of artificial intelligence. A recent report by Stability AI, the renowned developer of the Stable Diffusion model, reveals that Intel’s Gaudi 2 compute card significantly outperforms NVIDIA’s top-tier chips in AI image generation training. The comprehensive comparison conducted by Stability AI showcases the remarkable efficiency of Gaudi 2 under specific parameters.
Using the MMDiT model with 2 billion parameters, trained to a depth of 24 and utilizing BFloat16 mixed precision, the Gaudi 2 card can train 1,254 images per second at full capacity and 927 images per second at 256 capacity. In contrast, NVIDIA’s H100-80GB card processes 595 images per second, while the A100-80GB manages 381 images per second. This indicates that Gaudi 2 trains 55% faster than the H100 at equivalent capacity, signifying a significant performance boost; it is also 2.43 times faster than the A100, highlighting Intel’s leap forward in AI computing technology.
The release of this report is set to attract considerable attention to Intel’s Gaudi 2 chip and poses a challenge to NVIDIA’s leading position in the AI computing market. As AI technology continues to evolve, enhancing computational efficiency is crucial for accelerating model training and fostering the proliferation of AI applications. Intel’s Gaudi 2’s impressive performance foreshadows an even more intense competition in the AI domain.
【来源】https://www.itheat.com/view/45945.html
Views: 2