In a recent test of Apple's MLX machine learning framework, a benchmark showed that the new Apple Silicon Mac far outperformed NVIDIA's RTX 4090. Apple announced the release of MLX on December 6, an open source framework designed specifically for Apple chips. The framework is designed for AI developers to build, test, use, and enhance in their projects.
Developer Oliver Wehrens recently shared some benchmark results of the MLX framework on Apple's M1 Pro, M2, and M3 chips with the NVIDIA ARTX 4090 graphics card. It uses OpenAI’s speech recognition model Whisper.
Wehrens used the Whisper model to transcribe speech and measured the time it took to process a 10-minute audio file. The results show that the performance of the M1Pro chip is comparable to that of the NVIDIA GPU. It takes 216 seconds to process audio, while the 4090 only takes 186 seconds.
Newer Apple chips perform much better, though. For example, different people ran the same audio files on an M2 Ultra with 76 GPUs and an M3 Max with 40 GPUs and found that these chips took less time to transcribe the audio than the NVIDIA GPU.
There are also significant differences in power consumption between Apple chips and NVIDIA products. Specifically, if you compare the power consumption of a PC running NVIDIA 4090 and idle, it will increase by 242 watts.
In comparison, a MacBook with 16 M1 GPU cores has a much smaller increase in power consumption between running and idle, with a difference of only 38 watts.
These results highlight Apple's progress in artificial intelligence and machine learning capabilities and may be the beginning of better performance for Apple products. As the MLX framework becomes open source, it paves the way for wider application and innovation by developers.
The NVIDIA 4090 GPU starts at $1,599, excluding a computer, and is the same price as the M3 MacBook Pro launched in 2022, but the prices of the M3 Pro and M3 Max are still increasing rapidly.