As the M4 Mac series goes on sale overseas, more and more real-world tests have surfaced. Besides strong CPU and GPU performance, the AI hasn’t disappointed either. Recently, a user shared M4 Max Whisper speech-to-text benchmark results, significantly ahead of NVIDIA’s professional discrete GPU, the RTX A5000.
Whisper speech-to-text performance benchmark results: M4 Max vs RTX A5000, with the M4 Max coming out clearly ahead.
A few days ago, a netizen shared on X his real-world test data of Whisper v3 Turbo speech-to-text performance for the M4 Max vs RTX A5000. He said that when using Whisper V3 Turbo via MLX technology, the M4 Max completed transcription of an audio file in just 2 minutes 29 seconds, with power consumption of only 25W.
For comparison, he also tested NVIDIA’s professional-grade graphics card RTX A5000, which took 4 minutes and 33 seconds to complete the same task, with power consumption at 190W. This also shows that the M4 Max not only took half the time, but also consumed nearly 8 times less power—a quite impressive performance:
Holy cow
M4 Max transcribed a 179:23 audio using Whisper V3 Turbo in just 2:29 with MLX
with a mere 25W power drawThe same audio, transcribed with same Whisper V3 Turbo on an RTX 5000,
took 4:33 and consumed 190W https://t.co/8tFgD3AnJ9 pic.twitter.com/rBj73mYm5z— INIYSA (@lafaiel) November 12, 2024
Compared with the RTX A5000, some people might feel it’s a bit unfair, since this professional graphics card was launched in April 2021, and three years have already passed. However, its current market price is still around 80,000 to 90,000. Looking at price alone, choosing the M4 Max would be slightly more expensive, but at least you get a MacBook Pro with quite good specs, while the A5000 only gets you a graphics card.
The base M4 Max also comes with 36GB of unified memory, while the RTX A5000 has 24GB of GDDR6 video memory.
So, looking purely at the Whisper task, the M4 Max is undoubtedly the better value option.
However, it’s not surprising that Whisper was won by the M4 Max, as this chip is equipped with multiple encoders. Unlike most graphics cards that have one or two encoders, the M4 Max features four encoders, including two standard video encoding engines and two Pro Res encoding and decoding engines.
What’s more noteworthy is that this test was conducted in “balanced mode.” If the fan were adjusted to maximum speed, the transcoding time could be shortened even further.
With the launch of M4 Macs and their excellent performance in various AI computing tests, many people are now trying to link multiple M4 Macs together to create a large computing server. However, whether the construction cost is truly more cost-effective compared to using multiple GPUs remains to be seen.
M4 Mac AI Coding Cluster
Uses @exolabs to run LLMs (here Qwen 2.5 Coder 32B at 18 tok/sec) distributed across 4 M4 Mac Minis (Thunderbolt 5 80Gbps) and a MacBook Pro M4 Max.
Local alternative to @cursor_ai (benchmark comparison soon). pic.twitter.com/2BAiABKHpD
— Alex Cheema – e/acc (@alexocheema) November 12, 2024
When will the M4 Mac be released in Taiwan?
Although Apple hasn’t made an official announcement yet, Apple’s community ads previously revealed that the M4 MacBook Pro series would go on sale on December 20th, so the timing makes quite a bit of sense.
There’s no news about the M4 Mac mini and M4 iMac, but they basically shouldn’t be much different from the M4 MacBook Pro, which means we’ll likely have the chance to buy the M4 Mac series by the end of next month.
Based on past experience, the first batch may not have much stock, so if you want to get your hands on it right away, remember to be quick.
Source: KOCPC Chinese