🤓 IBM’s AIU and Nvidia’s AI chips are both designed to accelerate deep learning tasks, but they have some differences in their features and performance. Here are some of the main points of comparison:

• Precision: IBM’s AIU uses approximate computing and supports a range of smaller bit formats, including both floating point and integer representations, to reduce the amount of computation and memory required for AI operations. Nvidia’s AI chips use tensor cores that can perform mixed-precision calculations, such as FP16 and FP32, to boost the speed and accuracy of AI models.
• Speed: IBM’s AIU can perform up to 1.5 trillion operations per second, while Nvidia’s latest AI chip, the H100, can perform up to 2.5 trillion operations per second. However, the speed of an AI chip also depends on other factors, such as memory bandwidth, cache size, and interconnects.
• Power: IBM’s AIU claims to use 10 times less power than a CPU for deep learning tasks, while Nvidia’s H100 claims to use 3 times less power than a CPU for the same tasks. The power consumption of an AI chip also depends on the workload, the cooling system, and the system configuration.
• Price: IBM did not provide a specific price for the AIU, but it was expected to be available in the first half of 2022¹. However, due to the global chip shortage and supply chain disruptions caused by the pandemic, the launch of the AIU was delayed until the third quarter of 2023².