Mac mini M4 - a revolution in energy efficiency and power? (film, 3m)
In his latest video, Jeff Geerling discusses the performance of the Apple M4 processor, highlighting its remarkable efficiency in comparison to other chips on the market. Having conducted various tests, Geerling decided to showcase the M4 Mini, especially after previously criticizing the M4 Mini against his Raspberry Pi with eGPU setup. While the Raspberry Pi setup is indeed faster for applications like 4K gaming and language models (LLMs), the unparalleled efficiency of the M4 Mini deserves specific attention. Geerling notes that while his tests indicated around 4 GigaFLOPS per watt, the M4 Mini achieves an astonishing 7.57 GigaFLOPS per watt, making it a leader in this category.
It's important to highlight that despite the lower TeraFLOPS performance, the M4 Mini holds no equal in terms of efficiency. Geerling emphasizes that at idle, the M4 Mini consumes just 4 watts, which is comparable to various single-board computers like the Raspberry Pi 5. This demonstrates how Apple has successfully adapted mobile technology to desktop performance. Although his Ampere server delivers 3 TeraFLOPS, its efficiency pales in comparison to the M4 Mini, which provides significantly better energy performance.
Geerling points out the challenges that Intel and AMD face as they attempt to match Apple in efficiency. While Apple has successfully scaled up performance while maintaining low power consumption, competitors continue to struggle to improve their figures. Geerling humorously likens comparing the Ampere server to the M4 Mini to comparing a spaceship to a regular car, firmly believing that the M4 Mini represents the best value for money in 2024.
In his video, Geerling also addresses the attention to detail issues, such as the odd placement of the power button on the M4 Mini. By sharing personal experiences and applying various tests, Geerling effectively showcases how many features are packed in the compact design of the M4 Mini. His recommendations are sure to attract attention from users seeking a blend of performance and efficiency in one device.
Lastly, it's worth mentioning that at the time of writing this article, Jeff Geerling’s video has amassed 209,988 views and 7,549 likes, indicating significant interest and valuable insights provided to viewers. This further confirms that the efficiency and performance of the M4 Mini are at the forefront of desktop computer technology. Geerling can take pride in his research and opinions gaining such recognition within the tech community.
Toggle timeline summary
-
Discussion on pausing work on the Ampere server video due to M4's efficiency.
-
Comparison of M4 Mini's performance with Raspberry Pi setup for 4K gaming and LLMs.
-
M4 Mini's efficiency is highlighted in terms of GigaFLOPS per watt.
-
M4 Mini achieves nearly 300 GigaFLOPS and outperforms competitors' efficiency.
-
M4 Mini's idle power draw is only 4 watts, comparable to a Raspberry Pi.
-
M4 Mini is recognized as the most efficient CPU tested.
-
Apple's focus on power efficiency is contrasted with Intel and AMD's attempts.
-
Potential benefits of running multiple Mac Minis in a compact rack space.
-
Frustration about the power button placement on the Mac Mini.
-
Comparison of M4 Mini's speed to M1 Mac's Mac Studio.
-
Anticipation for M4 Ultra's performance and commentary on Linux support.
Transcription
I had to pause working on my Ampere 192 core server video because after running some more tests, I just had to post on this M4's efficiency. Especially after I posted a couple videos dunking on the M4 Mini with my Pi eGPU setup. Now, that Raspberry Pi setup is faster both for 4K gaming and for LLMs, but that's mostly from the AMD GPUs I was running with it. But Apple, especially with the M4 chip, they got efficiency dialed in. For years, the best ARM chips I tested in my top 500 benchmarks hovered around 4 GigaFLOPS per watt. Now, that test runs FP64 calculations, so 64-bit numbers. When I say something does 100 GigaFLOPS or a TeraFLOPS or whatever, that's not the same as saying your GPU does 20 TeraFLOPS at FP8 or FP16. I just want to make that clear. But that's besides the point. Not only does my base model M4 Mini get nearly 300 GigaFLOPS on the CPU, it puts every other chip's efficiency I've tested in the rear-view mirror hitting 7.57 GigaFLOPS per watt. That even beats this Ampere server that almost topped the charts at 4.82 GigaFLOPS per watt. Now, granted, this server does 3 TeraFLOPS, which is a lot more than 0.3 TeraFLOPS, and it has a power button right on the front. But pound for pound, dollar for dollar, I honestly think the M4 Mac Mini is the best value computer you can buy in 2024. And the most insane thing about all this? When it's not doing anything, the idle power draw is just 4 watts. That's like only a watt more than a Raspberry Pi 5. So we have SBC-level idle power draw on the low side, and full tilt, it's beating CPUs that burn hundreds of watts all by themselves. This uses just 40. For the whole system. It's insane. Ridiculous, even, which is why I put that in the title. The chip isn't the fastest at everything, but it's by far the most efficient CPU I've ever tested. Apple's been able to keep the mobile-first focus on sipping power at idle while scaling it up to desktop and even server-class performance. Intel and AMD are still trying to claw away to hit better idle numbers, but they're not even competing yet, at least on desktop. I mean, even this Ampere server still uses 100 watts at idle, though comparing this to the Mini is like comparing Starship to my Toyota Camry. But imagine, in 1.25U of rack space, I wish they made it just a hair shorter, you could run three Mac Minis idling around 10 watts, giving almost a TeraFLOP of CPU performance. Not to mention, there's a decent GPU and NPU built in, 10 gigabit networking, and high speed Thunderbolt I.O. in the back. If only they didn't put the power button on the bottom. All joking aside, I did 3D print this bracket because I think it's dumb having to lift up the computer just to press the power button. Now, I just press down on the Mac and that's it. I'm not going to do a full review of this thing, there are plenty of those already, but I did run it through my gauntlet of tests, and I've already run my automated Mac setup playbook on it to put it in service at home. It's faster in many ways than the M1 Mac's Mac Studio I use at work, and I can't wait to see how Apple does with the M4 Ultra. If only they made Linux easier to run on it, or supported external graphics cards, or, well, there's a reason I still like my other computers too. I haven't tested the M4 Pro or M4 Max, but all my current benchmark results are linked in the description. Until next time, I'm Jeff Geerling.