Who is the king of AI chips? Qualcomm and NVIDIA have their own strengths

Qualcomm's AI chip beat NVDA in two of three energy efficiency categories, while Taiwan-based Neuchips won the remaining category, according to test results released by MLCommons on Wednesday.

Specifically, Qualcomm's AI 100 beat NVIDIA's flagship H100 chip in image classification, with Qualcomm's chip achieving 197.6 server queries per watt compared to NVIDIA's 108.4 queries per watt, and Neuchips topping the list with 227 queries per watt.

Qualcomm also beat out NVIDIA in target detection, scoring 3.2 per watt on its chips compared to 2.4 per watt on NVIDIA chips. Target detection can be used to analyze applications such as video of retail stores to see where shoppers go most often.

However, in a natural language processing test, NVIDIA won first place in both absolute performance and energy efficiency. Natural language processing is the most widely used artificial intelligence technology in systems such as chatbots. NVIDIA achieved 10.8 samples per watt, with Neuchips in second place at 8.9 samples per watt and Qualcomm in third place at 7.5 samples per watt.

According to the report, MLCommons is a third-party organization that measures the speed of artificial intelligence processing, and its MLPerf is the most authoritative and influential international AI performance benchmark test.

NVIDIA dominates the market for training AI models with massive amounts of data. These AI models are trained to be widely used for so-called "inference", such as performing tasks like generating text responses to prompts.

Analysts believe that the market for data center inference chips will grow rapidly as companies apply AI technology to their products, but companies such as Google are already exploring ways to control the resulting additional costs.