Next iPhone faster than M1 Ultra Mac Studio

[UBS28]

It just goes to show that single core performance is pretty irrelevant in a normal computing environment. I have no idea why some people think it's so important.

Single core performance is extremely important. If 1-core is maxing out, it can screw up everything.

Try running VPS Avenger in Logic Pro with the M1 Max and hear the audio crackle sometimes because 1-core is overloaded (and the other cores having plenty of headspace).

I have to adjust my project settings to not get this plugin to crap out.

Upgrading to the M1 Ultra won’t change a thing because it has the same single core performance as my M1 Max.

 

[l0stl0rd]

Perhaps it will be but the architecture/ cores are 2 generations ahead of the M1 which is basically an A14 ?‍♂️

Also it will throttle anyway.

 

[bobcomer]

Try running VPS Avenger in Logic Pro with the M1 Max and hear the audio crackle sometimes because 1-core is overloaded (and the other cores having plenty of headspace).

Not something I run. I'm just a business type and don't do any video/audio stuff that one would pack into a single thread.

Interesting observation though, I'll have to think about that. I never hit anything like that with my workload. (it's always multiple processes and easier to spread across cores...)

 

[JouniS]

People using the highest performing (but still less raw performance than the other vendors) macOS systems for awhile already know that performance is better with another rig, but the macOS experience is what they prefer.

macOS is a boring generic easily replaceable OS, and that's how things should be. Surely it's a bit nicer than Windows and Linux, but that's rarely the decisive factor.

I've been primarily a Mac user since 2007, but I've also struggled with the narrow product range most of that time. Half the time there are no laptops worth buying, and half the time the desktops are not worth buying. Sometimes those two periods overlap, like in the late 2010s. I almost switched back to Linux at that time, but my employer gave me a high-end MBP, which was a decent enough laptop even if it wasn't worth its price. The 2020 iMac was great again and easily competitive with other desktops in the same price range, as long as you didn't care too much about the GPU. I bought one, but I often use a Ubuntu VM in it, because many things are more convenient in Linux than in macOS.

Apple Silicon reversed the situation. The laptops are great and easily worth buying, while the desktops are underwhelming. If you already have a laptop, there is no benefit in buying a desktop in the same price range, because it's the same hardware in a different package. Apple doesn't even try to take advantage of the desktop form factor. The Mac Studio looks promising, but it's crippled by it's low memory and internal storage capacity.

 

[mr_roboto]

I've been overclocking chips for 15 years. Any chip can be overclocked and you'd likely see extra performance. IE. Back in the days, my AMD Athlon XP would run at 1.8Ghz but you could overclock it to 2.5ghz. I doubt Firestorm is an exception.

But doing official binning is extra work for the chip makers.

Forget about the extra work. Being able to overclock a chip doesn't mean it's actually a good idea for that chip's manufacturer to release an official version which runs at that frequency.

The fundamental thing most (not all) overclockers don't realize is that you are always risking incorrect program execution and other potentially harmful side effects, even when you swear you've found a stable OC. The people on the other side of the fence run analysis tools on the chip's netlist and physical layout, get characterization data from the fab on that process and validate it while bringing up engineering sample chips, do reliability analysis to figure out what the safe voltage and temperature limits are for the desired operating lifespan, and so forth. They actually know.

It literally doesn't matter that, in principle, you can OC a M1 chip. Of course you can. What matters is whether Apple's silicon engineering team has done the work to validate that those frequencies (and/or the voltages required to make them work) are safe and good to use in a product which won't let users down. Designing the actual bin sort would not be a huge extra effort, they know the chip's critical paths and already need to create a test program to characterize them. (Even with no speed grades, you're always sorting into at least two bins - works/doesn't work at target frequency.)

 

[ChromeCloud]

Good to see that an interesting discussion has emerged from rather silly initial considerations.

My two cents: Apple does not "overclock" the M1, even in its Pro, Max and Ultra configurations because... they simply don't have to.

The M1 is clocked to hit the sweet spot between performance and power consumption, and it happens to deliver already great ST performance in those conditions, so why push it more?

The M1 that debuted in late 2020 still has great ST performance by today's standards, which see the top-scoring Intel Core i9-12900K being only 13% faster than M1 in ideal conditions.

Yes, that is in ideal conditions.

As soon as you execute more than one intensive task at a time, the Intel CPU is unable to offer its peak ST performance because of thermal limitations: peak clock speed (what Intel calls "Turbo Boost" frequency) drops down significantly when multiple cores are engaged for a sustained amount of time (EDIT: unless you have a really big cooler, like the one you could fit in a machine that is the same size of a Mac Pro).

This does not happen on any variation of M1. Regardless of how many cores are engaged, ST performance remains the same, which in turn makes the system more responsive under load. And that, I think, is ultimately something that a professional user would benefit from.

 

[nintendoswitch]

That would definitely be funny!

(dark mode turns on while typing ?)

 

[quarkysg]

Not something I run. I'm just a business type and don't do any video/audio stuff that one would pack into a single thread.

It'll be an interesting experiment if you could maybe limit your computer's CPU to 1 GHz (for maybe a 8/16 core CPU) and see how it fares running mainstream OSes. Likely you will feel that it is sluggish even when it is idling.

 

[leman]

Yeah, I shouldn't have mentioned interrupts, but I still disagree, it just doesn't make sense to me given what I know about how OS's work on modern hardware.

I do wonder what you mean by main task though as I never have just one main task and what I do takes more cores to feel "snappy".

@ADGrant already explained this. You can search for “main thread” to find more info. Even in modern UI frameworks, the main thread abstraction is still vital. An app even processing and drawing/layout loop is fundamentally single-threaded, even if modern software does often use auxiliary tasks to process non-immediate work (like fetching data from the internet etc.). But that is orthogonal to the task of updating the UI itself.

 

[bobcomer]

It'll be an interesting experiment if you could maybe limit your computer's CPU to 1 GHz (for maybe a 8/16 core CPU) and see how it fares running mainstream OSes. Likely you will feel that it is sluggish even when it is idling.

Of course it would. I bet I'd get more of my job done in that situation than someone using only a single thread.

I have very different computing needs and experience than it appears the rest of the people here do.

 

[quarkysg]

Of course it would. I bet I'd get more of my job done in that situation than someone using only a single thread.

I have very different computing needs and experience than it appears the rest of the people here do.

I think you are intentionally circling around the issue being discussed. I don't think anyone has claimed that they only need one execution thread. Folks are arguing that we need fast execution threads to create the effect of fast response. I don't think anyone said they only needed one execution thread.

I think from your response you agree that a slow CPU, regardless of how many cores will make any system feel sluggish for modern OSes. Regardless of computing experience, I would think everyone, including yourself, value a system with fast responses. This is what folks are arguing with you about, but I guess you already knew that.

 

[bobcomer]

@ADGrant already explained this. You can search for “main thread” to find more info. Even in modern UI frameworks, the main thread abstraction is still vital. An app even processing and drawing/layout loop is fundamentally single-threaded, even if modern software does often use auxiliary tasks to process non-immediate work (like fetching data from the internet etc.). But that is orthogonal to the task of updating the UI itself.

Of course I understand that, but that's really not a problem for my environment. So my snappy is being able to switch between tasks easily and fast, and there's no saturated cores to worry about. (part of my programming is to make sure there's never saturated cores) I do now understand what you guys mean though.

 

[Unregistered 4U]

People do choose machines based on multiple factors, which also includes performance. For example, there appears to be a not an insignificant amount of developers who purchased an M1 based laptop recently because of it's incredible performance in many development workflows.

We’re in agreement. While many in forums will look at a single metric from a single benchmark and start a thread like “Apple’s M1 can’t beat (random AMD/Intel processor)”, in the real world, that metric doesn’t add up to sales either gained or lost in any significant way. The folks focused in that way on raw performance are few and far between.

I HAVE learned in this thread that there are architectural benefits to Apple Silicon that means some cross platform tasks DO perform better on Apple Silicon at this time, even versus the highest end desktop Intel/AMD solutions. Understanding this, then, there IS a peak performance space that exists for cross-platform applications. It may be small, but these specialized cases are where the performance picture points directly at Apple Silicon in the near term for best performance.

This still feeds into the idea that whatever’s happening with Intel/AMD just doesn’t matter to anyone in the Mac world (save for those more interested in comparing numbers than just using their systems). They either don’t care because they need to use macOS, don’t care because they need a macOS app OR don’t care because whatever numbers are being shown for “raw performance” it’s understood that Intel/AMD can’t pull off what Apple has with memory bandwidth and cache.

 

[Unregistered 4U]

Apple Silicon reversed the situation. The laptops are great and easily worth buying, while the desktops are underwhelming. If you already have a laptop, there is no benefit in buying a desktop in the same price range, because it's the same hardware in a different package. Apple doesn't even try to take advantage of the desktop form factor. The Mac Studio looks promising, but it's crippled by it's low memory and internal storage capacity.

Apple Silicon didn’t reverse the situation, people’s experience with AMD/Intel makes them assume things about what they should expect. For example, those “underwhelming” desktops? They still outperform the desktops they replace, just as expected. What’s REALLY amazing is that Apple’s provided the single threaded performance of those desktops down their entire line. Prior to Apple Silicon, that was unheard of.

So, someone thinking “Laptops should always have lower single threaded performance than desktops” (as taught by AMD/Intel) will look at Apple’s desktops and be underwhelmed. And, they’ll likely continue to be underwhelmed with Apple Silicon systems as new iterations follow the same basic structure… impressive single threaded performance across the ENTIRE line, and better multi-threaded performance on higher end laptops/desktops.

 

[leman]

Apple Silicon didn’t reverse the situation, people’s experience with AMD/Intel makes them assume things about what they should expect. For example, those “underwhelming” desktops? They still outperform the desktops they replace, just as expected. What’s REALLY amazing is that Apple’s provided the single threaded performance of those desktops down their entire line. Prior to Apple Silicon, that was unheard of.

So, someone thinking “Laptops should always have lower single threaded performance than desktops” (as taught by AMD/Intel) will look at Apple’s desktops and be underwhelmed. And, they’ll likely continue to be underwhelmed with Apple Silicon systems as new iterations follow the same basic structure… impressive single threaded performance across the ENTIRE line, and better multi-threaded performance on higher end laptops/desktops.

Exactly. It’s a false dichotomy based on preconceived notions. It’s like “ARM CPUs can’t be fast” (because for a while they’ve been only used in low-power budget appliances), integrated GPUs can’t be fast (same reason) etc.

Of course, a point can (and should) be made that one would expect more performance from a premium desktop, but then again, this is gen 1 and we yet have to see what Apple is planning down the road.

 

[leman]

BTW, something just came to my mind: predictable peak per-core performance is also beneficial for development. It you know that your basic algorithm is fast enough on your beefy workstation, you know that it will also work fine on an iPhone. This can simplify things in some cases.

 

[Unregistered 4U]

Of course, a point can (and should) be made that one would expect more performance from a premium desktop, but then again, this is gen 1 and we yet have to see what Apple is planning down the road.

The point can be made, but more performance IS there, in multithreaded, GPU, neural cores, almost everything else other than single threaded. If one were to flip the release schedule, so that the Ultra comes out first, the Max’s and Pro’s and THEN the Air, folks would be expecting lower single threaded performance as the release goes on and would be utterly astounded how the Air’s have similar single threaded performance to the Ultra.

The point that SHOULD be made is that Intel and AMD should be able to provide the same single threaded performance on THEIR lowest end processors.

 

[JouniS]

Apple Silicon didn’t reverse the situation, people’s experience with AMD/Intel makes them assume things about what they should expect. For example, those “underwhelming” desktops? They still outperform the desktops they replace, just as expected. What’s REALLY amazing is that Apple’s provided the single threaded performance of those desktops down their entire line. Prior to Apple Silicon, that was unheard of.

You are focusing too much on CPU performance. From my point of view, it's one of the least interesting aspects of a computer, because the CPU is rarely the bottleneck.

Intel/AMD consumer chips support up to 64 GB of RAM in a laptop. If you put effectively the same chip into a desktop, the limit increases to 128 GB. Laptops don't support as much memory, because memory chips are quite large and there is usually not enough physical space inside the laptop. Apple didn't increase the memory capacity in their desktops, which makes them underwhelming for my purposes. This is particularly important for the M1 Ultra, which competes against the cheaper workstation chips from Intel and AMD. Increasing the memory limit to 512 GB or even just 256 GB would have made the high-end Mac Studio a much better computer.

Desktops tend to have internal expansion capacity, which allows you to put many of the ugly peripherals and their cables out of sight. At the very least, there should be a few slots for additional storage. Apple doesn't support it, probably as a way of assigning blame. If your setup looks ugly because of peripherals and cables, it's your fault. If a Mac looks ugly due to internal expansion capacity, it's Apple's fault.

Desktops can be larger than laptops, which provides many benefits. Apple doesn't bother taking advantage of them, which makes their current desktop lineup underwhelming.