The stuff Hacker News actually recommends.

Desktops and mini PCs · Apple Mac

Apple Mac M3

8owners with an opinion
60%of verdicts positive
3/2positive/negative since 2024
0happy after 3+ years

Praised for performance.

What owners wrote

Likes it
On my M3 mac, it works okay inside a docker container with just CPU. { "model": "strands-decider-2B-hobson-v19", "answers": { "is_urgent": { "type": "noul", "noul": 0.8287 } }, "usage": { "input_tokens": 86, "output_tokens": 1 }, "latency_ms": 1732.17 } This is how I got it running - https://gist.github.com/2891eb0db9ea92c1a4e860d44f556292 There's a lot more to be done if we optimize for MLX & let it run on a Mac mini instead of the docker wrapper.
Neutral
M3 was a weird generation, as they contained fewer transistors than the previous ones. It is slightly faster in single core tasks, and has a few more cores, but they are very close. But in terms of gpu, m3s are quite nerfed esp because they lowered the memory bandwidth, so on llm performance they are on par. I have both an m3 and an M1 Max, one of them from work, so I have tested them extensively (though the m3 is binned and 14”, the m1 full and 16”). M3 had better TTFT but the M1 had a bit higher tokens/s.
Likes it
I can't tell you how much I disagree with this take. Microsoft's AMD64 emulator is slow and buggy compared to Rosetta, and you will need it a lot more, too. Many apps will need to rely on this, including programs many users will immediately try to use, like Visual Studio. Neither Visual Studio nor its compilers support running on an ARM host; it does seem to basically work, but is slow, which is not good considering Visual Studio is already not particularly fast. It will even display a message box warning you that it is not supported during setup, so you couldn't miss it (note that this applies to the Visual Studio compilers needed by Python and Node for installing packages with C extensions). MSys2 gave me a lot of trouble, too; the setup just doesn't seem to work on ARM64. Chocolatey often installs AMD64 or x86 binaries on ARM instead of native ones; sometimes native ones don't exist anyways. Third party thing that needs to load a kernel module? Don't bet on there being an ARM64 build of it; sometimes there is, sometimes there isn't. WinFSP has a build, many hardware drivers I've looked at don't seem to (don't laugh: you can pass through USB devices, there is sense in installing actual hardware drivers.) I just set up a fresh copy of Parallels on an M3 Mac a couple months ago, I'm stopping now to be terse, this paragraph could easily be much longer. It would suffice to say that basic Windows software usage is a lot worse on Parallels than a native Windows AMD64 machine.…
Doesn’t like it
On a 3090 (24gb vram), same prompt & quant, I can report more than double the tokens per second, and significantly faster prompt eval. total_duration: 10530451000 load_duration: 54350253 prompt_eval_count: 36 prompt_eval_duration: 29000000 prompt_token/s: 1241.38 eval_count: 460 eval_duration: 10445000000 response_token/s: 44.04 Fast prompt eval is important when feeding larger contexts into these models, which is required for almost anything useful. GPUs have other advantages for traditional ML, whisper models, vision, and image generation. There's a lot of flexibility that doesn't really get discussed when folks trot out the 'just buy a mac' line. Anecdotally I can share my revealed preference. I have both an M3 (36gb) as well as a GPU machine, and I went through the trouble of putting my GPU box online because it was so much faster than the mac. And doubling up the GPUs allows me to run models like the deepseek-tuned llama 3.3, with which I have completely replaced my use of chatgpt 4o.
Doesn’t like it
macOS is behind. And this is speaking as someone who probably owns one of everything Apple makes, Apple stock, and was exclusively Mac for the last 10 years. I have less than 50 hours use on my Windows 11 machine, a midgrade Lenovo P358 rig I bought renewed because it had plenty of memory and an Nvidia T1000 card. Yet it taught me that the test of an operating system is how quickly you can navigate around, and how well it can find things, given only clues. Windows 11 is just snapper, quicker, than the latest macOS running on a new M3 Mac.
Likes it
You know that RAM in these machines is more different than the same as "RAM" in a standard PC? Apple's SoC RAM is more or less part of the CPU/GPU and is super fast. And for obvious reasons cannot be added to. Anyway, I manage a few M1 and M3 machines with 256/8 configs and they all run just as fast as 16 and 32 machines EXCEPT for workloads that need more than 8GB for a process (virtualization) or workloads that need lots of video memory (Lightroom can KILL an 8GB machine that isn't doing anything else...) The 8GB is stupid discussion isn't "wrong" in the general case, but it is wrong for maybe 80% of users.
Neutral
> Why were you mucking with Grub configs? Because something went wrong. Because I needed to enable a driver, actually, and something on the Arch forums or AskUbuntu said that I could enable it in a boot parameter in my grub config. I cannot remember the specific driver, but it was absolutely not to fix any boot partition issues. Also, I explicitly stated that I didn't know what I was doing in the comment you're replying to. Please don't get me wrong; I really don't have any rose-tinted glasses in regards to Linux. I'm basically a Mac person now; my work computer is a Mac M3, my home computer is a MacBook i9. I do have NixOS dual-booted on my personal computer, and it's been a nightmare to get all the drivers working. Just kidding, they still don't work, because the Linux drivers for the T2 MacBooks are garbage. There's a good chance that I will be nuking the Linux partitions this weekend in all honesty. The only thing that runs Linux full time in my house is my server, which runs a NixOS install on a tmpfs root. It's not like I have a ton of love for Linux, if I did I would probably still be using it for my daily driver, but I just haven't had the dual-boot-partition issues that people complain about. Maybe it's because I so rarely actually went to Windows, but grub more or less did the job. I will acknowledge that getting Linux installed in a secureboot environment continues to be a pain, however.
Likes itHasn’t owned it
Did Jeff Geerling get paid to post misleading headlines and reviews on Intel vs Apple? He's basing this on his own "super computing benchmark" measuring only FP64 performance. Meanwhile, any normal person using these $600 laptops will never run FP64 supercomputing tasks on the CPU. He also only showed perf/watt in this one benchmark but skipped the fact that Apple Silicon is significantly faster. Obviously, all this efficiency is on the low end—the laptop only achieves a little over 120 Gflops, while Apple's faster M3 and M4 Macs are pushing through up to 1.2 Teraflops. I can't believe this crap is getting upvoted. Hopefully this guy loses all his credibility going forward. This kind of post is seriously misleading and smells like like an undisclosed sponsor review.
Likes itHasn’t owned it
It used to be that for locally running GenAI, VRAM per dollar was king, so used NVidia RTX 3090 cards were the undisputed darlings of DYI LLM with 24GB for 600€-800€ or so. Sticking two of these in one PC isn't too difficult despite them using 350W each. Then Apple introduced Macs with 128 GB and more unified memory at 800GB/s and the ability to load models as large as 70GB (70b FP8) or even larger ones. The M1 Ultra was unable to take full advantage of the excellent RAM speed, but with the M2 and the M3, performance is improving. Just be prepared to spend 5000€ or more for a M3 Ultra. Another alternative would be a EPYC 9005 system with 12x DDR5-6000 RAM for 576GB/s of memory bandwidth with the LLM (preferably MoE) running on the CPU instead of a GPU. However today, with the latest, surprisingly good reasoning models like QwQ-32B using up thousands or tens of thousands of tokens in their replies, performance is getting more important than previously and these systems (Macs and even RTX 3090s) might fall out of favor, because waiting for a finished reply will take several minutes or even tens of minutes. Nvidia Ampere and Apple silicon (AFAIK) are also missing FP4 support in hardware, which doesn't help. For the same reason AMD Halo Strix with a mere 273GB/s of RAM bandwidth and perhaps also NVidia Project Digits (also speculated to offer similar RAM bandwidth) might just be too slow for reasoning models with more than 50GB or so of active parameters. On the other hand, if th…
Likes itHasn’t owned it
Both the M3 and the EPYC will be useful for far longer than the time it takes Apple to have the M4 on their next-gen laptops. Computers last a lot longer than they used to. I have a 10 year old Mac Mini that’s still comfortable to use, and, while an M3 Mac is a beast, it’s not that much faster than an M2 (or an i7) to create a qualitative change in my workflows. What is possible now was already possible last year. It’s just faster now. I get a higher return on investment with better keyboards and screens.