Intel Gaudi 3 AI Accelerator Built on TSMC 5nm Node, Faster Than Nvidia H100

Intel recently unveiled their latest AI accelerator, the Gaudi 3, built on the TSMC 5nm node. This new chip is said to be faster than NVIDIA’s H100, making it one of the most powerful accelerators currently available.

The Gaudi 3 is a CPU-based accelerator that can provide up to 20 TOPS (trillion operations per second) of AI performance. It also features Intel’s Deep Learning Boost and DLBoost Extensions, which further enhance its compute capabilities. The processor is designed to handle tasks such as image and video recognition, natural language processing, speech recognition, autonomous driving, machine learning, and other AI-related activities.

In addition to its increased performance, the Gaudi 3 also has improved power efficiency. It is capable of operating at less than 200W of power, compared to the 300W required by the H100. This means that the Gaudi 3 can provide more compute performance than the H100 without needing to draw more power.

To make sure that the Gaudi 3 is compatible with the current AI workloads, Intel included several key features. It supports data parallelism, which allows it to process data in parallel, thus allowing for faster completion times. It also supports distributed training and can distribute data across multiple nodes, enabling larger and more complex models. Lastly, it supports OpenVINO, a set of open-source computer vision libraries from Intel.

The Gaudi 3 is an impressive piece of hardware that provides a significant increase in performance for AI applications. With its capability to operate at lower power levels than the H100, while still providing more compute performance, it should prove to be a viable choice for those looking for more efficient computer vision solutions.

Read more here: External Link