We spoke to you onDecember 8 of a new framework published by Apple. The latter is called MLX and was designed to help AI computer scientists “develop machine learning on machines using Apple Silicon chips.» In other, less technical terms, it's a tool that helps make Macs with an Apple Silicon chip learn and respond in an automated way.
In order to see what the MLX machine learning framework was capable of, a developer named Olivier Wehrens carried outa test pitting the new Apple Silicon chips against the Nvidia RTX 4090 graphics card.For information, the latter is displayed at the price of 1799 euros on the brand's website. To carry out his test, Olivier used a voice recognition model created by OpenAI.
The reason why Olivier used this model is very simple, he wanted to seehow long each chip takes to transcribe a 10 minute audio recording.Thus, he put the chips into competitionM1 Pro, M2 Ultra, M3 Max, et Nvidia 4090.
Test results
A first test appeared on the web but did not pleaseNvidia, who accused Olivier of having published the results of a test optimized only for Apple Silicon chips, and not for theRTX 4090.
This first test shows that the M1 Pro chipfrom Apple took 216 secondsto transcribe the ten-minute audio recording. Which is the lowest score. The graphics cardNvidia4090does better than the M1 Pro, with186 seconds. But is dethroned, by far, by the M3 Max and M2 Ultra Max chips, which respectively obtained the scores of100 seconds and 95 seconds. But as said above, these results do not reflect the real capabilities of the Nvidia graphics card.
After optimizing the test with Nvidia processes, it emerged that it did much better than the others, taking only 8 seconds to transcribe the audio.
See also:
