Wired
Google’s Tensor chip is said to have powerful AI performance, but GeekBench’s AI benchmark scored inferior to Apple’s A15 Bionic.
However, the ML Perf score outperformed the A15 Bionic, making an overwhelming difference, especially in language processing.
Tensor chip surpasses A15 Bionic in ML Perf
Performed by AnandTech is an AI performance comparison using MLPerf, a benchmark for measuring the performance of running machine learning applications.
In benchmarks such as image recognition, object detection, and image segmentation, the Tensor chip scored better than the A15 Bionic.
However, these benchmarks are losing to Qualcomm’s Snapdragon 888.
On the other hand, in language processing tests, the Tensor chip outperformed other chips.
The Pixel 6 series equipped with the Tensor chip is equipped with unique functions using AI related to languages such as real-time translation and transcription, and it seems that it can be said that it is the result of the performance behind it.
GeekBench scores inferior to the A15 Bionic
On the other hand, GeekBench’s AI processing benchmark showed that the Tensor chip was inferior to the A15 Bionic, similar to previous results.
However, while the A15 Bionic uses CoreML, the Tensor chip uses the GPU / NNAPI, so a simple comparison may not be possible.
According to the results of AI Benchmark 4 using the same NN API, the Tensor chip outperforms other Android chips.
However, as Google itself states, how to use AI is more important than benchmarking.
Good benchmarking doesn’t mean you get a good user experience.
Source: AnandTech via 9to5Google
(Hauser)
Source: iPhone Mania