I made a similar transition as you earlier this year from a i7 8700k/GTX 1080 Ti Windows desktop to a Mac Studio and much like you I'm very happy with that decision for a lot of the reasons you already described.

One of the more interesting aspects of the switch I hadn't considered though is that it actually (positively) impacted my power bill - I'm paying about $5-$10 less per month since the switch.

The difference in form factor is welcome as well - some days I find myself astonished at how much power fits in such a (relatively) small box on my desk.

The Mac Studio was a bit of a sleeper. For AI inferencing at 128gb it is higher performance than any affordable consumer solution with that much memory available, it's quiet, it's small, it's power efficient, it supports 5 displays. It costs a lot less than an equivalent MBP.

It's the desktop a lot of us having been waiting for... no built in screen, no forced purchase of hundreds of dollars in "magic (i.e. garbage) keyboard/mouse), and ports out the wazoo... even the base model I have has 4x Thunderbolt 4, 4x USB 3 (two Type C on the front, 2 Type A on the back), 10Gb eth, even a headphone jack and SD slot.

Interesting comment about the Apple keyboard. I have bought Apple Magic keyboards for all my Windows machines. Need a correct driver to make the fn key work and remap the delete key, but otherwise I have not yet found a better TKL keyboard.

Hotswap mechanical or nothing for me. I hate low travel keys (and flat keycaps).

I actually just bought this last week for the work machine..

https://www.keychron.com/products/keychron-k8-pro-qmk-via-wi...

Not as good as the board on my personal PC, but that is a highly tweaked out, kit built, full metal gasket mount board that I have mumble hundreds of dollars in to...

People are different. I grew up with clunky and noisy IBM keyboards. Really love the low key travel.

Model Ms are terrible. A nice modern mechanical with light linear switches is neither.

I used to like apple keyboards but I got RSI using them. I was probably holding it wrong but switching to mechanical keyboards fixed it for me.

I buy mechanical keyboards for all my macs and skip the flat feel of the chiclet keyboards. It is too bad that such an experience can’t be very portable.

Still the performance vs price seems pretty bad. From what I see their 10 core 500GB SSD + 32GB RAM costs as much as a 7950X (16 cores) + RX 7900XT + 64GB RAM + 1TB SSD which should have better performance and you can actually play AAA games (at very high performance and quality, also 4k).

I like Apple stuff (I have an iPhone, MBP) but their high-end stuff is really overpriced for what it offers.

Well, let's try to compare like for like...

7950x = $599 7900XT = $800 64GB of RAM = $150

Thats' $1550 right there, before motherboard, storage, case, cooling, powersupply, etc. By the time you add all that (plus labor, either in dollars or in time) I bet you're in the $3500 range with those specs.

You have to consider the total package... the Studio is 7" x 7" x 4"... that's smaller than a microATX board, never mind some giant dual slot graphics card (Which also draws 300w btw, over 3x what the entire Studio draws).

Lack of noise, heat, and power efficiency has real value. Some of us are at a point where we want stuff that just works, not to fiddle with components and BIOs settings. Plus, frankly, nobody on the PC side has anything even close to Applecare.

I think you went way overboard with the costs - a good AM5 MOBO is around $200, case $150, AIO cooling or good fan $100 and a good PSU would be $200. That's a total of $650. Add to that assembly and sanity checking for let's say $100 (many times it comes free for these upper end setups) and we get $750.

That's a total, with your estimates for the CPU and GPU prices, of $2300 which is around $1000 lower than the Studio while having more power. I agree the power efficiency is worse but also note that the Mac simply doesn't have access to the same compute power. Also the CPU and GPU are redlined by default and going down to almost 50% TDP only causes a 5% drop in performance so it is not that bad.

Of course there are advantages that you mentioned (form factor, better support) but, at least for me, that is not worth a $1k+ premium. Also using MacOS for me at least is a pain compared to Linux.

You're comparing against the ultra. I'm using (and talking) about the base model Max, which is $1999 all in. (The Ultra is 20 core, btw. It's essentially two Maxes fused together.)

I will agree that the Ultra ($3999) is probably not a great value for most people, since outside of synthetic benchmarks, it's usually more like 10-20% faster, not 100%, as outside of editing 8k video or AI, there really isn't much that scales well to that many threads.

Another thing I'll mention, and that I think really is a big part of the special sauce, is the insane memory bandwidth.

A 7950x has a maximum memory bandwidth of 83.7GB/sec. An M1 Max has 409GB/sec.

It's really hard to outrun RAM that's essentially soldered directly to the CPU.

I’m generally a pretty big fan of self-built systems in general and Intel in particular (what can I say, every couple years they pulled a rabbit out of their hat during my childhood, it was really magical).

But that memory bandwidth is some envy inducing stuff.

The ram is regular package on package stuff, it's not the reason for the high bandwidth.

The bandwidth is due to apple giving more area and pins to more channels.

I have a similar 8700k system (I’ve got a weaker GPU, though). I already kind of feel like anything that gets even close to stressing it would be much happier running off on some cluster or server somewhere.

Chess Engines I've got to say are an argument against Apple Silicon, at the same power draw any AMD or Intel solution is going to beat the pants off of it.

Seems super fast to me? Benchmarking now...

bench 1024 8 26 in latest stockfish (well, whatever brew installed two days ago)

Total time (ms) : 128108 Nodes searched : 1605433485 Nodes/second : 12531875

12,531,875 NPS is competitive with a 2950x threadripper, which is an $1100 cpu with a TDP of 180w. The M1 Max has a power draw under load of about 90w max (and that's the full SoC), but under typical high loads it's more like 50-60w.

2950x is a weird comparison when it was two generations older than what AMD had available at the time. Also the 2950x still did 37422972 NPS which is still roughly 3x higher. I think it goes without saying that if you look back enough generations you will eventually find an AMD/Intel product that gives less performance/watt.

The 5950x was also available for sale at the time, a system based on it was cheaper, and it does 54029460 NPS in stockfish which is over 4x faster.

It's not like the Mac Studio is too slow for chess engines to be usable for purposes like analysis, but the thing is if you make that argument than why not get a much cheaper Ryzen 5 or Intel i5 based system which is still faster than the Mac Studio? Say a Ryzen 5600g which is also pretty power efficient and can fit in a nice little box and push a few displays?

https://openbenchmarking.org/test/pts/stockfish&eval=7596252...

Because running chess engines is like 0.5% of what I use a computer for.

How does this description of your hardware purchases relate to the post you replied to? Or to the news article?

Yes, the 10900k as the biggest 14nm consumer chip was a power guzzler and is bad compared to Intel/AMD/Apples current line up on new processes.

These 10900k or 11 series anecdotes always crack me up. I switched from M1 Pro to Ryzen 6800H and it feels just as fast as the much more expensive apple laptop. It's not even using TSMC 5nm like the new Zen 4 parts.

I think the majority of what made the M1 lineup so successful, was that they were on TSMC's 5nm before anyone else, and had that exclusivity.

November 2020 the m1 mac mini was released... September 2022 was when Zen4 came along, the next CPU to use the same node. Nearly a full 2 years.

Cinebench scores... The M1 Ultra got 24189 multithreaded... not that impressive but its tdp is 60w... at 65w the 7950x (eco mode) scored 31308 in the same test, down from 38291 at full power usage (170w).

It will be interesting to see if apple pushes for another exclusivity deal with TSMC for 3nm.

AMD is yet to release a TSMC 5nm mobile part (low power 5nm), so the exclusive foundry lead Apple's been paying for is more like 2.5 years. It is my understanding that things are pretty much the same with TSMC's 3nm node. Apple's lead is all but guaranteed for at least a couple more years.

> It will be interesting to see if apple pushes for another exclusivity deal with TSMC for 3nm.

It’s pretty much guaranteed as it’s part of the deal Apple has with TSMC. They bankroll a significant part of costs for a new process node, in exchange they get exclusive access to the node for a number of years.

> Even for gaming it's a lot more competent than you'd expect, especially for games that have been ported to Silicon. WoW runs at over 100fps (NB: 1440p, 165hz gaming monitor) with settings turned up pretty high. Factorio runs awesome, and just got a silicon port that bumped performance another 20% or so.

WoW doesn't need much hardware to run. What is the fps on the gaming rig for comparison?

About the same. It’s cpunlikited not gpu. But either is plenty of gps combined with an adaptive refresh rate monitor… even the studio hits the monitors 165hz cap indoors, and it’s 100+ in most out door zones… some of the ones with tons of fog get it down into maybe the 80s, but it’s still really smooth… no jitter or microstuttering.

I'm curious if anyone has a similar story but running Linux instead of macOS? I have a DIY AMD 5600X system (128 GB RAM, two M.2 drives) that is fast but warm. In the summer it can be uncomfortable in my upstairs office, despite A/C. I could be tempted into trying an ARM workstation if I thought it would make a big difference on heat.

I don't really do gaming, except occasional Minecraft with the kids. My GPU is a 10+ years old GeForce GTX 560, and it seems fine. (But maybe that only works with x64? I don't know.)

I've already had a taste of Apple's ARM speeds with my M1 Macbook Air. Running `make clean && make` on Postgres is practically instant. I don't understand what's going on there to make it so fast. But for my daily work I'd rather be in Xubuntu. Is anyone out there doing something similar?

No experience myself but I've heard good things about Asahi Linux (https://asahilinux.org/). Not sure how usable it already is as a daily driver.

For the heat try undervolting the cpu, there’s even a utility to help you get started with the settings by 1usmus, if you are willing to tinker you can get some incredible power efficiency.

Counterpoint to the "but I have a PS5 for that": if you develop video games or work in CAD every day, Macs just aren't a sensible option.

Regarding recording studios, PCs are generally kept in a machine room away from everything else.

Great that it works for you, but for a lot of creative industries, the Mac doesn't make sense in 2023.

Yea, I’m aware how recording studios work. Those of us making amateur recordings at home don’t always have the luxury.

I'm really curious what is holding CAD companies back to port their software to the Mac?

This is a great question that people were asking even 15-20 years ago, likely even before then too, but it was before my time in asking it.

Essentially, the demand for CAD on Mac just isn't high enough to justify it, especially when anybody serious about CAD wants the absolute cutting edge of hardware, and the ability to expand upon it as soon as the rest of the industry is able to take advantage of the latest GPU. Being competitive is everything.

It's never been a case of software needing to be ported. Some CAD apps have come and gone on Mac, but the demand just isn't there in terms of sales to make it worthwhile. That's not to say that there aren't CAD apps on Macs, there are, but the big boy industrial toys are elsewhere.

Keep in mind we used to run Parallels and Bootcamp in the '00s to get Windows software on Macs!

Plus the dominant platform (Soludworks) is totally built using MS UI libraries… ribbon, etc.

I thougt autocad was ported to M1?

No one uses autocad anymore. It's as relevant to the modern CAD world as MS Basic is to us. Even the architects have moved on to stuff like Revit.

(Like, to further clarify, when I did my Engineering Technology degree, we were trained on AutoCAD (But also 3D Cad), and it was obvious then that autocad was antiquated and was soon to be abandoned en masse.

This was in 2004.

I'd assume their rendering enigne is written for single API, such as OpenGL or Vulkan. Porting it to metal would be quite large project.

Isn't there a reasonable OpenGL shim on top of Vulkan?

Cad programs barely work on the systems they’re designed for. I love solidworks, but it’s a real turd if you aren’t using one of the correct “certified” driver for your graphics card.

I'd say that much of the M2 speed comes from its ultrawide RAM interface, only possible because RAM is soldered basically directly to the CPU die(s). It can't scale to server sizes, but it makes perfect sense for a laptop or a smaller desktop.

Yeah, it's insane. Even my M1 Max is 400GB/sec.

> While 243GB/s is massive, and overshadows any other design in the industry, it’s still quite far from the 409GB/s the chip is capable of. More importantly for the M1 Max, it’s only slightly higher than the 204GB/s limit of the M1 Pro, so from a CPU-only workload perspective, it doesn’t appear to make sense to get the Max if one is focused just on CPU bandwidth.

https://www.anandtech.com/show/17024/apple-m1-max-performanc...

It's shared with the GPU, so limited it to CPU-only doesn't seem very fair.

Apple's wide ram interface is an advantage. But they aren't doing anything exotic to achieve it. AMD & Intel could offer consumer cpus with more memory channels but they choose not to, likely for cost and market-segmentation reasons.

Isn’t having the memory integrated into the dies bit exotic for a performance desktop?

"On Package" isn't the same thing as on die. Apple's M1 LPDDR memory setup isn't really any different from what you would find in a normal PC laptop. By putting the memory as close as possible to the CPU it makes it easier to maintain signal integrity, but it's not really any different from anyone else's approach.

To throw a similar experience into the mix... I agree on all counts, particularly around silence & efficiency - especially on the MBPs where the fans will actually turn off when not needed.

I've been a long time Windows user (20+ yrs) and have built my own PCs for nearly as long. My most recent 5900X + 3080 FE setup with 2x Gen 4 NVMes is amazing, but it's not the most efficient. ~95W idle and can top out around 200-400W depending on the use.

The (base) 14" M1 Pro I got launch day has been great overall, just really smooth/snappy in real-world use, where lately I've hit snags on Windows and MS software. The 14" M1 Pro maxes out around 35W or so (including its display). Desktop's more like 500W including displays. Great battery life on all the new M1s, it can do a full day at 8-14+ hours of medium/heavy use depending on your use/multitasking. It doesn't have to throttle down for thermal or power reasons like many recent Intel laptops (including a previous and recent-gen XPS 15 - which was much noisier by comparison).

In terms of quietness, I have the 3080 FE undervolted to typically only top out around 240W with under 2% performance loss. It typically runs around 55-65c in games without the fans being noticeable. 5900X is cooled by a 240mm AIO so it too is able to maintain reasonable temps running nice low fan speeds - only up to around 1050 RPM for sustained max loads which is still quiet compared to most other peoples' desktop PCs I've seen.

The 14" M1 Pro has never been audible above the idle fan speeds on the desktop (which run at ~550-600rpm and are also nice and quiet). The MBP is virtually always silent unless you manually crank the fan. Direct sunlight or somewhere very hot the fans might kick up a little more in response, but for me it's never really noticeable or even on or audible under normal use, and I like quiet devices.