My workstation has a 3990x.

Our "world" build is slightly faster on my M1 Max.

https://twitter.com/kiratpandya/status/1457438725680480257

The 3990x runs a bit faster on the initial compile stage but the linking is single threaded and the M1 Max catches up at that point. I expect the M1 Ultra to crush the 3990x on compile time.

Are both cases compiling to the same target architecture? If not, you may well be comparing the relative performance of different compiler backends instead of comparing the performance of your CPUs.

(+ now I see it's rust: how parallel is your build, really?)

> now I see it's rust: how parallel is your build, really?

Not the OP but I install a lot of Rust projects with Cargo and recently did some benchmarking on DigitalOcean's compute-optimized VMs. Going from 8 cores to 32 cores was a little disappointing:

Bat (~40 crates): 68s -> 61s

Nushell (486 crates): 157s -> 106s

Compilation starts out highly parallel and then quickly drops down to a small number of cores.

The target arch doesn’t matter for the change-build-test loop d vs do. All that matters is how fast can you compile your code to test it.

If the final x86 production build takes longer it doesn’t matter - that happens on the cloud anyway.

Edit: Rust builds are very parallel until linking. No different than any other LLVM build.

> The target arch doesn’t matter for the change-build-test loop d vs do.

It matters when comparing CPU performance, which is what this benchmark is being used for.

that the arm backend is faster to compile for is a legit advantage for the developer though, even if its distinct from cpu perf

> The 3990x runs a bit faster on the initial compile stage but the linking is single threaded and the M1 Max catches up at that point.

Isn't linking IO-bound?

https://github.com/rui314/mold would suggest otherwise. Massive speedups by multithreading the linker. I think traditional linkers just aren't highly optimised.

> https://github.com/rui314/mold would suggest otherwise.

Does it, though?

I mean, if you read that link you'll notice it boasts the linker's performance by comparing it with cp and how it's "so fast that it is only 2x slower than cp on the same machine."

Is cp supposed to be CPU-bound?

Yes, but that's for mold, which is multithreaded. The original context of this thread being the question of whether a linker would see speedups from multithreading. Most people are using traditional single-threaded linkers which are an order of magnitude slower than mold. The fact that mold is so much faster suggests that a linker does indeed see big speedups from multithreading.

The data the linker is running on will generally largely be in memory (in their posted example, which is a warm compile, completely in memory).

Exposing my limited understanding of that level of the computing stack - it is but Apple seems to have very very good caching strategies - filesystem and L1/2/3.

https://llvm.org/devmtg/2017-10/slides/Ueyama-lld.pdf

There is a breakdown in those slides discussing what parts of lld are single threaded and hard to parallelize so I suspect single thread performance plays a big role too. I generally observe one core pegged during linking.

> Exposing my limited understanding of that level of the computing stack - it is but Apple seems to have very very good caching strategies - filesystem and L1/2/3.

That would mean that these comparisons between Threadripper and the M1 Ultra do not reflect CPU performance but instead showcase whatever choice of SSD they've been using.

L1/2/3 are CPU caches, not SSD. Though there is a good chance these are mostly firmware optimizations, not hardware. So still not an apples-to-apples comparison of cpu design.

By firmware you mean microcode, but I don't think either of those actually use microcode to control this.

> L1/2/3 are CPU caches, not SSD.

Why did you omit the reference to "file system"?

Are we supposed to ignore the fact that a linker's main job is reading object files and write the output to a file?

I find this sort of argument particularly comical given a very old school technique to speed up compilation is to use a RAM drive to store the build's output.

For a clean build and a reasonably specced machine, all the intermediate artifacts will still be in the cache during linking.

Do you have any benchmarks of the two to share against a mean to ensure its built optimally? Maybe something like https://opendata.blender.org ? It's painful hearing random anecdotes only to learn the person didn't apply thermal paste to their cpu.

From the tweet reply you used sccache with hot cache, which would probably be mostly single-threaded since it's just fetching and copying things from cache.

I’m not trying to compare the details of SOC performance.

Just that with the same hot caches, the average change-build-test loop that developers do 100+ times a day is just faster on the M1 Max.

Try mold.

>Try mold

Curiosity got the better of me:

https://github.com/rui314/mold

We plan to move to it once MacOS support lands (for the laptops).

Intel was the single-threaded king until Zen3, so that's no real surprise.

Try the same thing with mold.

My 1st gen 16 core Threadripper is barely faster than an M1 Pro/Max at kernel builds, so a 64 core TR3 should handily double the M1 Ultra performance.

But you know, I'm still happy to double my current build perf in a small box I can stick in my closet. Ordered one :-)

How many threads are actually getting utilized in those kernel builds? I don't work on the kernel enough to have intuition in mind but people make wildly optimistic assumptions about how compilation stresses processors.

Also 1st gen threadrippers are getting on a bit now, surely. It's a ~6 year old microarchitecture.

Kernel has thousands of compilation units. Each of them is compiled by a separate compiler process. Only the linking at the end doesn't parallelise, however it should take a much smaller part of the time. The proportions change of course, if you develop kernel and do incremental builds lots of times. Then the linking stage might become a bottleneck.

The above statement should also relate to most other C/C++ projects.

Kernel is C code though, which doesn’t require as much cpu to compile as say c++. It could be more IO bound. There should be a lot of more legit data available on this than my armchair speculations.

C++ adds a few time-consuming stuff, like templates, however my belief was that what uses the most time is optimization passes, which won't be much faster for C. The actual benchmarks should be done however.

Yes, it'd be interesting to see this comparison made with current AMD CPUs and a full build that has approximately the same price.

I am curious whether there is a real performance difference?

I do lots of computing on high-end workstations. Intel builds used to be extremely expensive if you required ECC. They used that to discriminate prices. Recent AMD offerings helped enormously. I wonder whether these M1 offerings are a significant improvement in terms of performance, making it worthwhile to cope with the hassle of switching architectures?

but do the m1s have ecc?

All of them. The CPU graph is pegged for most of the compilation.

Kernel compilation can be heavily parallelized.

I wouldn’t automatically expect a linear decrease in compile time with growing core count. That would have to be tried.

Seems like for things that are: 1. Perfectly parallel 2. Not accelerated by some of the other stuff that's on the Apple Silicon SoC's ...it will be a toss-up.

Threadripper 3990X get about 25k in Geekbench Multicore [1]

M1 Max gets about 12.5k in Geekbench Multicore, so pretty much exactly half [2]

Obviously different tasks will have _vastly_ different performance profiles. For example it's likely that the M1 Ultra will blow the Threadripper out of the water for video stuff, whereas Threadripper is likely to win certain types of compiling.

There's also the upcoming 5995WX which will be even faster: [3]

[1] https://browser.geekbench.com/processors/amd-ryzen-threadrip...

[2] https://browser.geekbench.com/v5/cpu/search?utf8=%E2%9C%93&q...

[3] https://www.amd.com/en/products/cpu/amd-ryzen-threadripper-p...

Also of note is that half of the Mac Studio's case is dedicated to cooling. Up to this point, all M1 Max benchmarks are within laptops while all Threadripper benchmarks are in desktops. The M1 Max in the Mac Studio will probably perform better than expected.

This is sound logic and probably be the case but I wonder if this effect will be less than what we have seen in the past because of the reduced TDP of the M1 processors in general.

Maybe the cooling and power delivery difference between laptop formfactors and PC formfactors will be less with these new arm based chips.

Maybe; every chip hits a point where feeding in more power doesn’t make it go any faster.

If I was to guess, the increased cooling probably helps the Studio sustain similar boost clocks as the laptops, but for longer.

Although it’s possible these are on N4x, which might increase the attainable boost.

Geekbench is extremely sensitive to the OS. Like the same CPU on Windows & Linux score wildly different on Geekbench. For example the 3990X regularly hits 35k multicore geekbench when run on Linux: https://browser.geekbench.com/v5/cpu/11237183

Something is seriously fishy about those geekbench results.

24-core scores 20k, 32-core scores 22.3k, and 64-core score 25k. Something isn't scaling there.

Many GB5 (and real world) tasks are memory bandwidth bottlenecked, which greatly favors M1 Max because it has over double a Threadripper's memory bandwidth.

Sort of. The CPU complex of the M1 Max can achieve ~200 GB/s, you can only hit the 400 GB/s mark by getting the GPU involved.

At the same time the Threadrippers also have a gargantuan amount of cache that can be accessed at several hundred gigabytes per second per core. Obviously not as nice as being able to hit DRAM at that speed.

That cache is not uniform time access. It costs over 100ns to cross the IO die to access another die's L3, almost as much as going to main memory. In practice you have to treat it as 8 separate 32 MB L3 caches.

Also, not everything fits into cache.

Yeah, it’s the real world tasks that GeekBench tries to simulate that don’t tend to scale linearity with processor count. A lot of software does not take good advantage of multiple cores.

That's certainly true. But if that's your workload you shouldn't be buying a 64-core CPU...

I use a few 32 and 64 core machines for build servers and file servers, and while the 64-core EPYCs are not twice as fast as the 32-core ones due to lower overall frequency, they're 70% or so faster in most of the things I throw at them.

Does Geekbench actually attempt to simulate that in their multi-core score? And how?

I was under the impression that all of their multi-core tests were "run N independent copies of the single-threaded test", just like SPECrate does.

> A lot of software does not take good advantage of multiple cores.

It sounds pointless to come up with synthetic benchmarks which emulate software that is not able to handle hardware, and then use said synthetic benchmarks to evaluate the hardware performance.

It has a very specific point: communicating performance to people who don't know hardware.

Most consumers are software aware, not hardware aware. They care what they will use the hardware for, not what they can use it for. To that end, benchmarks that correlate with their experience are more useful than a tuned BLAS implementation.

Probably it's the thermals that don't scale. The more the cores, the lower the the peak performance per core.

Having not seen benchmarks, I would imagine that claimed memory bandwidth of ~800 GB/s vs Threadripper's claimed ~166 GB/s would make a significant difference for a number of real-world workloads.

Someone will probably chime in and correct me (such is the way of the internet - Cunningham's Law in action) but I don't think the CPU itself can access all 800 GB/s? I think someone in one of the previous M1 Pro/Max threads mentioned that several of the memory channels on Pro/Max are dedicated for the GPU. So you can't just get a 800 GB/s postgres server here.

You could still write OpenCL kernels of course. Doesn't mean you can't use it, but not sure if it's all just accessible to CPU-side code.

(or maybe it is? it's still a damn fast piece of hardware either way)

On an M1 Max MacBook Pro the CPU (8P+2E) cores peak at a combined ~240GB/s the rest of the advertised 400GB/s memory bandwidth is only useable by the other bus masters e.g. GPU, NPU, video encoding/decoding etc.

So now the follow-on question I really wanted to ask: if the CPU can't access all the memory channels does that mean it can only address a fraction of the total memory as CPU memory? Or is it a situation where all the channels go into a controller/bus, but the CPU link out of the controller is only wide enough to handle a fraction of the bandwidth?

It's more akin to how on Intel, each core's L2 has some maximum bandwidth to LLC, and can't individually saturate the total bandwidth available on the ring bus. But Intel doesn't have the LLC <-> RAM bandwidth for that to be generally noticeable.

Fascinating!

Linking this[1] because TIL that the memory bandwidth number is more about the SoC as a whole. The discussion in the article is interesting because they are actively trying to saturate the memory bandwidth. Maybe the huge bandwidth is a relevant factor for the real-world uses of a machine called "Studio" that retails for over $3,000, but not as much for people running postgres?

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

I'm obviously going to reserve judgement until people can get their hands on them. Apple makes good stuff but their keynote slides are typically heavily cherrypicked (e.g. our video performance numbers compare our dedicated ASIC to software encoding on a different architecture even though competing ASICs exist kinds of things).

Next-gen Threadripper Pro was also announced today: https://www.tomshardware.com/news/amd-details-ryzen-threadri...

Bit of a wet fart though, even Charlie D thinks it's too little too late. OEM-only (and only on WRX80 socket), no V-cache, worse product support.

https://semiaccurate.com/2022/03/08/amd-finally-launches-thr...

The niche for high clocks was arguable with the 2nd-gen products but now you are foregoing v-cache which also improves per-thread performance, so Epyc is relatively speaking even more attractive. And if you take Threadripper you have artificial memory limits, half the memory channels, half the PCIe lanes, etc, plus in some cases it's more expensive than the Epyc chips. It is a lot to pay (not just in cash) just for higher clocks that your 64C workloads probably don't even care about.

AMD moved into rent-seeking mode even before Zen3 came out. Zen2 threadripper clearly beats anything Intel can muster in the segment (unless they wanted to do W-3175X seriously and not as a limited-release thing with $2000 motherboards) and thus AMD had no reason to actually update this segment when they could just coast. Even with this release, they are not refreshing the "mainstream" TRX40 platform but only a limited release for the OEM-only WRX80 platform.

It was obvious when they forced a socket change, and then cranked all the Threadripper 3000 prices (some even to higher-levels than single-socket Epyc "P" skus) what direction things were headed. They have to stay competitive in server, so those prices are aggressive, but Intel doesn't have anything to compete with Threadripper so AMD will coast and raise prices.

And while Milan-X isn't cheap - I doubt these WRX80 chips are going to be cheap either, it would be unsurprising if they're back in the position of Threadripper being more expensive for a chip that's locked-down and cut-down. And being OEM-only you can't shop around or build it yourself, it's take it or leave it.

The PRO Threadrippers are not cut down, they have 8 memory channels and 128 PCIE 4.0 lanes. I think the only limitation compared to Epyc is that you can have 1 socket only.

And WRX80 has dedicated chipset lanes (https://www.amd.com/en/chipsets/wrx80), so effectively more PCI-e than EPYC, and bootable NVME RAID support on top.

It's bad for competition that only Apple gets to use TSMC's 5nm process. Though what's really bad is that Intel and Samsung haven't been able to compete with TSMC.

AMD will be on TSMC N5P next year, which will give them node parity with Apple (who will be releasing A15 on N5P this year), and actually a small node lead over the current N5-based A14 products. So we will get to test the "it's all just node lead guys, nothing wrong with x86!!!" theory.

Don't worry though there will still be room to move the goalposts with "uhhh, but, Apple is designing for high IPC and low clocks, it's totally different and x86 could do it if they wanted to but, uhhh, they don't!".

(I'm personally of the somewhat-controversial opinion that x86 can't really be scaled in the same super-wide-core/super-deep-reorder-buffer fashion that ARM opens up and the IPC gap will persist as a result. The gap is very wide, higher than 3x in floating-point benchmarks, it isn't something that's going to be easy to close.)

We've already seen x86 draw even with Intel 12th gen: https://www.youtube.com/watch?v=X0bsjUMz3EM

So in that one, you've got a 20-thread Intel part (6+8C/20T) at probably 3.5 GHz going against a 10-thread Apple part (8+2C/10T), at probably 3 GHz, and the Apple part still comes out on top by ~5% in Cinebench R23 MT. And that's with Intel having 50% more high-performance threads available.

Work out the IPC there - the Intel has a 2x thread count advantage, a 17% clock advantage, and Apple comes out 5% ahead. So the IPC gap there is about 2.46x.

It's not a perfect comparison of course, since we're mixing SMT and big/little cores, but in basically every area Intel should (on paper) have more resources available and Apple is coming out on top anyway by sheer IPC.

That's what I'm saying - you can't really do that approach with x86. It's not power-advantageous or transistor-advantageous to go super wide on the decode or reorder buffer like that on x86. And regardless of the tricks x86 uses to mitigate it, you've still got a 2.5x IPC gap at the end of the day. A 2.5x IPC gap will not be closed up by just a single node shrink.

And that's looking at MT, where your task scales perfectly. See where I'm going with this? Intel is using 2x the number of threads, and 3x the number of efficiency cores to get there. Apple can deliver that punch across a much lower number of threads - meaning ST-bottlenecked tasks will scale much much better on Apple.

With a single-threaded test, the M1 is pulling 7W vs 33W for the Alder Lake intel. Obviously that tells us nothing about efficiency, since we'd need to know the scores, but that's the downside, is for normal, poorly-threaded tasks, like surfing the web or editing code, the 12900HK is going to be boosting high to reach the same performance levels the M1 does at 3 GHz. And that's exactly what you see in the power figures there.

In short: you will likely see x86 able to keep up in one metric or another. You can win on performance if you just go nuclear on power. You can match on power on perfectly-threadable tasks that allow the x86 to deploy twice the threads (sharing instruction cache/etc). You can match on single-threaded battery life if you accept lesser performance. But the overall performance of the M1 derives from the massive IPC it generates, and that's something that x86 can't match nearly as easily.

Going ham on a single metric just to claim victory isn't nearly the same thing as the level of all-round performance and efficiency that Apple has achieved there.

(see also, putting a 128-thread Threadripper 3990WX workstation up against a 10-thread M1 Max laptop just to win at rendering... and people here thought that disproved that Apple was great hardware lol)

I don't think there's anything specifically different about the reorder buffer between x86 and ARM.

The reorder buffer size is just a logical consequence of the frontend width.

And yes, scaling an Aarch64 frontend is dead simple compared to x86 due to the fixed instruction width. The disadvantage of x86 is serious, but I don't know if we can count it out quite yet. This is the first time Intel and AMD got any serious pressure on that front. I'm sure they're taking the challenge seriously, and it'll take some years before we'll see the results.

x86 and ARM have different memory access assumptions which affects the reorder buffer in material ways. Apple added something to their chips to use the x86 memory model when flagged for Rosseta2.

How much little cores and HT contribute at this workload tho? If I say turn them off, could I claim the opposite, losing even 20% but using 40% less threads?

There is a third variable: Apple is putting ram much closer to the CPU than AMD. This has the advantage that you get lower latency (and slightly higher bandwidth), but the downside that you're currently limited to 128gb of ram, compared to 2tb for threadripper (4tb for epic). Amd's 3d cache that they're launching in a few months will be interesting since it lets the L3 go up a ton.

Latency to RAM is not any better. Bandwidth is extremely much better but not for this reason.

why so smug? who cares whichever company makes a better processor?

Well compare that to a 400$ CPU like a 5900x, the first M1 is slower than this one and cost 2x the price.

Apple's ARM chips can process a metric ton of ops per cycle due to the architecture of the chip: https://news.ycombinator.com/item?id=25257932

But the answer to the question is still "no".

Doesn't have to though. A Threadripper 3990X uses barrels of electricity, generates plenty of heat, comes with no GPU, has worse single-threaded performance, and still costs $4000 by itself without any of the parts needed to make it actually work.

The question is in relation to Apple's claim that it's "the world’s most powerful and capable chip for a personal computer".

It might also be reasonable to say that the threadripper is a workstation chip, not a chip for personal computers.

Edit: even AMD themselves call their threadripper lineup workstation chips, not personal.

https://www.amd.com/en/processors/workstation

Threadripper Pro is the workstation chip. Regular Threadripper (non-Pro) was not aimed at workstations, it was aimed at the "HEDT" market. Strictly speaking it's considered a consumer market (albeit for the enthusiasts of enthusiasts)

I'd call them personal chips. When I think of non-personal chips I think IBM POWER or Ampere Altra.

Why do you think of Altera as non-personal chips?

Not Altera, Ampere Altra. This: https://amperecomputing.com/processors/ampere-altra/

If the purchase page says to "contact sales" and doesn't list a price then it is not for consumers.

Click where to buy tab, then there is a list of distributors. Including ones which sell workstation versions with a configurator and pricing. https://store.avantek.co.uk/ampere-altra-64bit-arm-workstati...

Depends on what you define "capable" as. Remember, they specify that it is the most powerful and capable chip, not necessarily complete system.

There's no other chip that has the power of an RTX 3090 and more power than an i9-12900K in it - after all, Threadripper doesn't have a lick of graphics power at all. This chip can do 18 8K video streams at once, which Threadripper would get demolished at.

I'm content with giving them the chip crown. Full system? Debatable.

It's a fantastic chip but that wasn't the question. I love my M1 Max and I love my Threadripper workstation, each has their own strengths and that's alright.

Through you would need to compare it to the coming threadripper 5000WX(?) or better the soon coming Ryzen 7000 CPUs (which seen to have integrated graphics).

I mean they all are CPUs coming out this year as far as I know.

Only if the only thing you compare is CPU performance - adding a big GPU on die adds a certain amount of ‘power’ by any measure.

It doesn't matter. Speaking as an Apple cult member imo Threadripper is better value if you're not using the machine for personal use.

It doesn’t support your argument when we’re talking about a massive processor like a threadripper vs. a M1 Ultra.

The performance per watt isn’t in the same universe and that matters.

The article claims that the chip is "the world’s most powerful and capable chip for a personal computer". It's reasonable to ask whether it genuinely is faster than another available chip, it's not an implicit argument that it's not powerful.

The M1 Ultra is by a very wide margin the bigger of the two. According to Tom's Hardware [1], top-of-the-line Epycs have 39.54 billion transistors. That is about a third of the 117 billion in the M1 Ultra. Apple builds bigger than anyone else, thanks largely to their access to TSMC's best process.

The M1 Ultra is a workstation part. It goes in machines that start at $4,000. The competition is Xeons, Epycs, and Threadrippers.

That’s not really a fair comparison. Apples chip spends most of that on their GPU, and the neural engine takes a chunk too. Threadripper is only a CPU.

> The performance per watt isn’t in the same universe and that matters.

I couldn’t give less of a shit about performance-per-watt. The ONLY metric I care about is performance-per-dollar.

A Mac Studio and Threadripper are both boxes that sit on/under my desk. I don’t work from a laptop. I don’t care about energy usage. I even don’t really care about noise. My Threadripper is fine. I would not trade less power for less noise.

This is what some folks miss.

One hour of my time is more expensive than an entire month of a computer electricity bill.

Some people just want tasks to perform as fast as possible regardless of power consumption or portability.

Life's short and time is finite.

Every second adds up for repetitive tasks.

The only reason I've ever cared about watts is that generally speaking 120 watt and 180 watt processors require more complicated cooling solutions. That's less true today than it ever was. Cases are designed for things like liquid cooling, and they tend to be pretty silent. The processors stay cool, and are pretty reliable.

I personally stick to the lower wattage ones because I don't generally need high end stuff, so I think Apple is going the right direction here, but it should be noted that Intel has also started down the path of high performance and efficiency cores already. AMD will find itself there too if it turns out that for home use, we just don't need a ton of cores, but instead a small group of fast cores surrounded by a bunch of specialist cores.

wattage doesn't really tell you how difficult it is to cool a part anymore. 11th-gen Intel is really easy to cool despite readily going to 200W+. Zen3 is hard to cool even at 60W.

Thermal density plays a huge role, the size of the chips is going down faster than the wattage, so thermal density is going up every generation even if you keep the same number of transistors. And everyone is still putting more transistors on their chips as they shrink.

Going forward this is only going to get more complicated - I am very interested to see how the 5800X3D does in terms of thermals with a cache die over the top of the CCD (compute die). But anyway that style of thing seem to be the future - NVIDIA is also rumored to be using a cache die over the top of their Ada/Lovelace architecture. And obviously 60W direct to the IHS is easier to cool than 60W that has to be pulled through a cache die in the middle.

The Zen 3 stock coolers work pretty well. I've never had a problem.

Looking it up though I do see a lot of concerns with the heat they generate. I can only conclude I don't push my chip very hard (which, honestly, I probably don't)

I've been happy with the AMDs I purchased over the past 4 years, we'll see how they hold up and how this next gen comes out. I did see that the recent Intels are quite competitive which is good for everybody.

yup, the correct answer here is people need to stop being worried about the thermals as a metric in themselves, and look at the performance their chip is generating. If your chip is running at 90C, but you're hitting 100 ScoreMarks, and attaching a 5hp chiller to it lets you hit 105 ScoreMarks, that's not really worth it.

Yeah, longevity, blah blah, but laptop chips are designed to sit above 90C under load, it's fine.

Just saying that "how hard it is to cool" doesn't solely depend on power consumption anymore. Heat density is making that harder and harder, even if power consumption stays the same.

What does improve though is how much heat it pumps into your room. Yeah, a Rocket Lake at 200W might be roughly as hard to cool as an AMD at 90W or whatever... but one is still putting 200W into your room and the other is still putting 90W. Temperatures are not the same thing as power dissipation either. I don't like having my gaming PC running in my room during the summer, and I'm actually looking at maybe running cables through the walls to have it in the basement instead. I also have a 5700G and some NUCs that are much lower power that I prefer to use for surfing and shitposting.

> wattage doesn't really tell you how difficult it is to cool a part anymore

Sure it does. Reading the rest of your post I think you're more talking about temperature than cooling requirements, but a 200W CPU needs 200W of heat dissipation, while a 60W CPU only needs 60W of heat dissipation. It's literally a 1:1 relationship since CPUs don't do any mechanical work, so power in == heat out.

Keeping temperatures below some arbitrary number does then include things like density, IHS design, etc... But that only matters for something like Intel's "Thermal Velocity Boost" where it's really important to stay under 70C specifically instead of just avoiding thermal throttling.

Air coolers can handle 300 watts without any complexity. Just a big block of fins on heat pipes.

I mean, yeah, but then you've got case issues and whatnot. I appreciate your point though.

The power is only relevant because it makes the machine quite in a compact form. If you've got a bit of space, then a water cooled system accomplishes a lot of the same thing. For some people there is an aesthetic element.

Power does make a big difference in data centers though - it's often the case that you run out of power before you run out of rack space.

Where power for a computer might make a difference could be in power-constrained (solar/off grid) scenarios.

I don't know if I've ever heard anyone make an argument based on $$$.

Did you know noise pollution causes dementia?

https://www.theguardian.com/society/2021/sep/09/transport-no...

Ouch. I wonder what Apple thinks about that after selling millions of noisy intel macbooks.

As for desktops, watercooling makes computers dead silent.

Yep, it's bad that those products were like that.

I personally bought a Ryzen 5950(?) instead of a Threadripper because I figured I'd accidentally spill water all over it or however it works. There are not many watercooled OEM products as far as I know.

The vast majority of developers today has a laptop as their main machine. Performance-per-watt is absolutely crucial there.

I'll just never understand this. Chances are you are at the same desk day in and day out. You probably have monitors and external keyboard/mouse hooked up because hunching over a laptop and using a touchpad is unnecessary torture for a fixed workspace. Given that why would you hamstring yourself with a thermally constrained and overpriced-because-miniaturization-isn't free setup?

Until maybe these M1's (and I'm not entirely convinced) I've not in the 20 years I've been computing seen a reasonably configured desktop (eg not just a laptop on a stick ala iMac but an ACTUAL desktop) ever not smoke the pants off of every single laptop you could put up against it. It's hard to beat the one-two punch of lots of power and room to cool it. If you are sitting at at desk why the heck wouldn't you leverage that?

I was a desktop diehard until performance reached the point where laptops are good enough. Being able to take the same computer around with you and use it seamlessly in a different place is a big improvement.

I still have a proper desk-based working environment hooked up to a docking station though. I really wouldn't want to use a laptop that doesn't have a first-party dock as my primary machine.

Meetings, working from home, travel, changing desks every few months. Not having a jungle of cables. Retina screen (still hard to find on external monitors). The vast majority of devs also don’t need an insanely powerful machine. If Docker on Mac wasn’t dog slow, I could easily get by with a $999 MacBook Air.

The only reason I also went up to 32GB RAM(almost bought the 64GB one) was because of Docker.

That’s cool. I am not the vast majority of developers. I am me. I use a desktop for high-end game development. I don’t give a shit about web development or laptop development. I care about compiling large C++ projects as fast as possible.

I agree that most developers are web/mobile developers who use a laptop. That’s great. I am an increasingly niche developer.

The root comment was a comparison against Threadripper. Normal developers should not waste money on a Threadripper. If someone is a niche developer that warrants a Threadripper then pointing out that most developers don’t need a Threadripper is a waste of time.

I care about power only because it means louder fans, and I like quiet.

Same. That's why I use watercooling to keep the room silent.

It is worth noting that it has, at least according to Apple's graphs, slightly more than an RTX 3090 in graphics performance.

So, even if it doesn't quite beat Threadripper in the CPU department - it will absolutely annihilate Threadripper in anything graphics-related.

For this reason, I don't actually have a problem with Apple calling it the fastest. Yes, Threadripper might be marginally faster in real-world work that uses the CPU, but other tasks like video editing, graphics, it won't be anywhere near close.

It wont be even close to RTX 3090 looking at m1 max and using same scaling maximum it can be close to 3070 performance.

We all need to take Apple claims with grain of salt as they are always cherrypicked so i wont be surprise if it wont be even 3070 performance in real usage.