Apple’s M3 Max Is Faster Than An RTX 4090 In OpenAI’s Whisper Speech Transcribing Test, But Only When Using An Unoptimized Benchmark

wccftech.com:

The M3 Max is Apple’s most capable 3nm SoC designed for the Mac, with its multi-core performance higher than the M2 Ultra, and its 40-core GPU gives a laptop RTX 4080 a decent run for its money. Despite these improvements, there is still a disparity that exists between laptop GPUs and NVIDIA’s ‘top of the line’ RTX 4090, but in the latest speech transcribing test, the M3 Max outpaces the fastest consumer graphics processor in the world, under the right test conditions that is.

Apple recently announced an open-source framework called MLX, which is specifically designed explicitly for the company’s M-series of chipsets. Developers can use this framework to design, test, and improve their projects, which Oliver Wehrens did with various Macs, along with an RTX 4090 GPU, and provided benchmark results below. The MLX framework uses Whisper, which is OpenAI’s speech recognition model, and Wehrens ran a transcribing test to see how fast various chips performed.

With a 10-minute audio file, the initial test results show the M2 Ultra and M3 Max are surprisingly faster than the desktop RTX 4090, but that is also because an unoptimized tool was run to carry out the benchmark, as shown in the graph. Wehrens provided results using the optimized tool as well, and the results are about as you would expect; no competition for the RTX 4090 as it leaves both the M2 Ultra, M3 Max, and the remaining SoCs in the dust.

In the refresh test, the RTX 4090 completed the audio transcription test in just 8 seconds, whereas the M2 Ultra finished the same test in 95 seconds, followed by the M3 Max in 100 seconds. The updated results show that Apple’s M-series family has a long way to go in terms of raw performance. One area where the custom silicon