Apple M1 Pro Review: Is it really faster than Intel/AMD?

[Stiqy]

That's a legacy "I won't learning nothing new" opinion - most OS agnostic consumers if exposed to both would immediately choose an M1 computer over a competing Windows computer because of the system's snappiness and efficiency (long battery life).

We've been told for over 30 years that people will choose Apple over Windows if given the choice. People have had the choice for over 30 years and yet Apple still has 15% market share.

May have to re-examine that bit of "consumer" mythology.

 

[hahahanoobs]

That's a legacy "I won't learning nothing new" opinion - most OS agnostic consumers if exposed to both would immediately choose an M1 computer over a competing Windows computer because of the system's snappiness and efficiency (long battery life).

And no, these laptops playing in rather stratospheric space - they successfully compete against the top tier Windows laptops speced up to their level, but that's not where most of the computer market lives.

Most machines are waaayyyy down lower in the territory that the M1 occupies. The lower end M1 (and soon to be M2) computers are what should worry AMD and Intel - especially with efficiency mandates coming down the pike like the <50 kWh/year mandate as has appeared in California. These models are rapidly gaining market share - shortly after introduction, the M1 Mini became the best selling PC in Japan.

The vast majority of the computers sold are not high end Ryzen or core-i9s - they're middling pretty crappy computers like low end chrome books or ultrabooks or their desktop equivalents. The M1s easily overpower these models in pretty much all common consumer metrics.

Even enterprises are abandoning the traditional Windows client/server model because it's just too hard to keep end-user machines working properly between user installed software, lousy Windows maintenance, and iffy CPU vulnerability mitigations - instead they're going for web apps and thin clients tied to SMB/CIFS shares - usual based on linux kernels. Machines they can rapidly wipe and replace should something go wrong.

It's M1 on Mac.
It's no x86 killer.

 

[kiwigraeme]

I think this report also uses some selective testing to make Intel/AMD look better. If you're comparing CPUs, why introduce a GPU for gaming benchmarks, why not use Intels built-in graphics, just like the M1? Of course I expect an Nvidia 3080 to blow away the GPU in the M1 SoC. Duh.

When comparing CPU only tasks the Apple SoC performs very well against Intel and AMD. And we're into 1.5 gen silicon at this point compared to what, 10th or 11th generation Intel? I'd say Intel better be looking over their shoulder because Apple is hot on their heels. Also, how do we know Apple won't integrate an external GPU, Nivida or AMD down the road?

At the end of the day, I think Apple has breathed some new life into their laptop business. After having used a MacBook Air M1 for just shy of a year now I can say this laptop is more than adequate for everyday personal and work tasks for a large percentage of people. As you said, creatives probably need something different but that is also true of Windows creative users. I don't think many creative people are using high powered laptops in mobile applications all that often due to the poor battery performance plus the need for external screens and other peripherals.

Some great points - I'm really glad Apple is pushing this - It pushes everyone else - like big/little cores that Intel have just done - is great for just idling along .
My point is this is great for Apple users with the compatibility caveat . It's also great for those with money who want a luxury chromebook - ie it will maintain itself - they can browse the web/social media , consume media , do banking etc , appletime with family etc - it's overkill but if you have lots of disposable income and a iphone ( or not ) it's a great buy .
However it's not a Linux/Windows PC killer .

Plus one of the interesting things I find about a lot Apple Users is "The United Front" = yet the few times I have landed on a apple user website - and lets say they are complaining about the keyboard , then you will often see a litany of other complaints about there about Apple - I loss my own itunes library and had to set ip up again - when they went from sabertooth to bobcat it borked my ..... etc etc

 

[Verne Arase]

FTFY
The gains here are no more significant than a normal "tick or tock" release on any silicon. Apple just got their latest and greatest out of the lab a few months before INTELs next release. We see the same see saw in ATI / NVIDEA card releases.

It's hardly revolutionary or earthshattering as the review shows. It's a good processor; better in some cases, and not in some others.

The high performance Firestorm cores in the A14/M* processors are wider than any x86 processor - this means that they have a 690 instruction execution queue, a massive reorder buffer, and eight decoders.

The out-of-order execution unit can execute up to eight instructions *simultaneously*. They are easily the most sophisticated processors commercially available.

The only way that x86 vendors can outdo their single core performance is by boosting clocks exorbitantly high, and paying for it with exponentially raised heat generation. The end product is that you need a beefy cooling system to cool both the shamefully wasteful CPU and the equally wasteful GPU, all contributing to extreme heat generation and electrical usage.

Apple Silicon is thirteen years in the making, financed by a company with very deep pockets. I can give a short historical perspective if you're interested.

 

[Verne Arase]

I don't necessarily disagree but my only comment is that you are overstating the importance of a strong GPU quite a bit: yes for gamers is crucial. Also for video editors you've gotta have a strong GPU.

But for everybody else? Either integrated graphics like Vega 3 to 8 on AMD or Iris Pro on intel is sufficient or is actually CPU dependent: Lots and lots of professionals depend on CPU workloads in fact a lot more than GPU based workloads.

M1 is of course exceedingly good at CPU task as well, not without caveats but pretty excellent so this isn't a direct knock against them, but it is possible to thrive with a laptop that doesn't has a dedicated GPU.

If you're just counting on CPU speeds, the lowly M1 will trash just about anything on the x86 side core-for-core. The only exceptions are desktop machines pushing clocks and heat waaayyy up with beefy coolers to remove all that excess heat.

 

[Dimitriid]

If you're just counting on CPU speeds, the lowly M1 will trash just about anything on the x86 side core-for-core. The only exceptions are desktop machines pushing clocks and heat waaayyy up with beefy coolers to remove all that excess heat.

I already said that: just re-read my post a bit more slowly where it says "M1 is of course exceedingly good at CPU task as well, not without caveats but pretty excellent so this isn't a direct knock against them"

Something tells me you probably won't hear how there's any caveats about M1 in just processing, in fact something tells me you're here just looking to get angry at anyone not unquestionably praising the M1.

Sorry I won't do that, it's good but it's not the best option for everybody and I wouldn't characterize options like Ryzen 3 "So far behind OMG is like not even a contest, it's ridiculous, just throw that x86 stuff in the garbage!" like you want to apparently.

 

[Verne Arase]

I already said that: just re-read my post a bit more slowly where it says "M1 is of course exceedingly good at CPU task as well, not without caveats but pretty excellent so this isn't a direct knock against them"

Something tells me you probably won't hear how there's any caveats about M1 in just processing, in fact something tells me you're here just looking to get angry at anyone not unquestionably praising the M1.

Sorry I won't do that, it's good but it's not the best option for everybody and I wouldn't characterize options like Ryzen 3 "So far behind OMG is like not even a contest, it's ridiculous, just throw that x86 stuff in the garbage!" like you want to apparently.

No, just saying the Firestorm cores in the A14 and M1 class processors are much wider than what can be achieved in x86 due to the variable instruction length of x86.

You just can't get much wider than what x86 offers, so your only option to increase single core speed is to up the clocks which leads to exponential heat generation.

Of course Intel and AMD could go to a fixed length instruction architecture to produce wider CPUs, but then it wouldn't be x86 any longer.

 

[HardReset]

No, just saying the Firestorm cores in the A14 and M1 class processors are much wider than what can be achieved in x86 due to the variable instruction length of x86.

You just can't get much wider than what x86 offers, so your only option to increase single core speed is to up the clocks which leads to exponential heat generation.

Of course Intel and AMD could go to a fixed length instruction architecture to produce wider CPUs, but then it wouldn't be x86 any longer.

And why not? Current CPU's count almost everything on micro ops, not x86. Also "width" of CPU is much more complex than just amount of decoders or amount of execution units.

Apple designed their CPU to be very wide and to use low clock speeds. AMD and Intel develop cores to scale from few watts to over 100 watts.

All in all, there is nothing really special about Apple M1. It's just designed for one role and abandons virtually all backwards compatibility.

 

[Verne Arase]

And why not? Current CPU's count almost everything on micro ops, not x86. Also "width" of CPU is much more complex than just amount of decoders or amount of execution units.

Apple designed their CPU to be very wide and to use low clock speeds. AMD and Intel develop cores to scale from few watts to over 100 watts.

All in all, there is nothing really special about Apple M1. It's just designed for one role and abandons virtually all backwards compatibility.

You make it sound like Moses scaled Mount Sinai and came down with clay tablets documenting the x86 instruction set.

x86 is not the be all end all instruction set. Apple elected to use the ARM instruction set - specifically the ARMv8 instruction set it requested ARM to design with 64 bit mode in mind: AArch64. Unlike x86 which just growed over time, AArch64 is designed to be optimized.

And yes, to grow Intel and AMD will have to grow and evolve to an instruction set capable of being optimized. Their very attachment to x86 is what is holding them and the performance and efficiency of their CPUs back.

 

[HardReset]

You make it sound like Moses scaled Mount Sinai and came down with clay tablets documenting the x86 instruction set.

x86 is not the be all end all instruction set. Apple elected to use the ARM instruction set - specifically the ARMv8 instruction set it requested ARM to design with 64 bit mode in mind: AArch64. Unlike x86 which just growed over time, AArch64 is designed to be optimized.

And yes, to grow Intel and AMD will have to grow and evolve to an instruction set capable of being optimized. Their very attachment to x86 is what is holding them and the performance and efficiency of their CPUs back.

Partially agreed. Of course it's easier to create whole new instruction set abandoning all backwards compatibility than using instruction set that is around 40 years old.

But, it's around 30 years now since x86 was supposed to be EOL. And still about only things x86 really have some kind of problems is translating x86 instructions into something that is easier to calculate and maintaining backwards compatibility. Both are pretty small problems. x86-64 already basically abandoned x87 instruction set and probably there will be something similar on integer side too.

 

[Verne Arase]

Partially agreed. Of course it's easier to create whole new instruction set abandoning all backwards compatibility than using instruction set that is around 40 years old.

But, it's around 30 years now since x86 was supposed to be EOL. And still about only things x86 really have some kind of problems is translating x86 instructions into something that is easier to calculate and maintaining backwards compatibility. Both are pretty small problems. x86-64 already basically abandoned x87 instruction set and probably there will be something similar on integer side too.

Never said x86 is dead - simply that if Intel and AMD want to advance, they're going to have to go with another architecture.

There is no "backwards compatibility" - once you've changed to a fixed length instruction set architecture, it's no longer x86. The only thing you can do is build a software bridge to be used while developers convert to the new architecture. Apple's successfully done it three times, proving it's not impossible.

There are two paths to faster execution speed: host a plethora of cores like servers do to service discrete tasks like file servers do for multiple users. Higher core counts for consumer level machines can improve multithreaded speed for those resource intensive workloads which can be easily multithreaded like video transcoding.

For the most part though, hosting a zillion cores on a consumer workstation just results in a lot of idle cores. Great for bragging specs, but not great for end user productivity.

The real important metric for consumer workstations is single core speed. That's where the end user sees the most benefit since most tasks are not easily multithreaded.

RISC basically in its ideal form runs one instruction per clock. That has its limits as raising clock speeds increases energy use and heat exponentially, and this heat is something that must be removed using beefy cooling systems which use even more energy.

The real key is parallelism - scanning plain old inline code to see what instructions can be scheduled to be run in parallel. To do that, you have to peer deep into the execution queue and locate opportunities where you can run multiple instructions simultaneously without affecting the results.

A14/M1 Firestorm cores can peer 690 instructions into the execution queue and find such opportunities, and using a massive reorder buffer can schedule up to eight instructions to run in parallel (simultaneously). This essentially allows the processor to achieve the holy grail - convert inline code into multithreaded code (but without the necessity of scheduling code on multiple cores). This is why I say that the Firestorm cores are among the most sophisticated in the world.

Of course, most times they don't hit that eight instruction ideal - I'm pretty sure the Apple Silicon Team (capitalized out of respect) raised a glass every time they could get eight instructions to run simultaneously. But this is how you get the single core speeds that A14/M1 gets without breaking the energy bank. This also means you have to have the arithmetic units to back all that up, and the logic to do all that isn't simple.

Including IP blocks, GPUs, CPUs and all the controller and caches M1 Max has something like 55 billion transistors.

This isn't something Apple came up with overnight - they've been building their own custom silicon and optimizing it for 13 years now. It was only the bugginess of Skylake and the failure of Intel to deliver on promises which drove them to bridge the gap from iPhone to Mac, and convert the Macintosh line to the ARM ISA and their own custom silicon.

 

[HardReset]

Never said x86 is dead - simply that if Intel and AMD want to advance, they're going to have to go with another architecture.

There is no "backwards compatibility" - once you've changed to a fixed length instruction set architecture, it's no longer x86. The only thing you can do is build a software bridge to be used while developers convert to the new architecture. Apple's successfully done it three times, proving it's not impossible.

But why should AMD or Intel abandon x86? I have heard for decades that x86 makes it impossible to create faster CPU's. And for decades AMD and Intel have proven otherwise.

Apple's solutions are much slower than native support from CPU, something happened on x86 very rarely.

There are two paths to faster execution speed: host a plethora of cores like servers do to service discrete tasks like file servers do for multiple users. Higher core counts for consumer level machines can improve multithreaded speed for those resource intensive workloads which can be easily multithreaded like video transcoding.

For the most part though, hosting a zillion cores on a consumer workstation just results in a lot of idle cores. Great for bragging specs, but not great for end user productivity.

The real important metric for consumer workstations is single core speed. That's where the end user sees the most benefit since most tasks are not easily multithreaded.

RISC basically in its ideal form runs one instruction per clock. That has its limits as raising clock speeds increases energy use and heat exponentially, and this heat is something that must be removed using beefy cooling systems which use even more energy.

The real key is parallelism - scanning plain old inline code to see what instructions can be scheduled to be run in parallel. To do that, you have to peer deep into the execution queue and locate opportunities where you can run multiple instructions simultaneously without affecting the results.

A14/M1 Firestorm cores can peer 690 instructions into the execution queue and find such opportunities, and using a massive reorder buffer can schedule up to eight instructions to run in parallel (simultaneously). This essentially allows the processor to achieve the holy grail - convert inline code into multithreaded code (but without the necessity of scheduling code on multiple cores). This is why I say that the Firestorm cores are among the most sophisticated in the world.

Of course, most times they don't hit that eight instruction ideal - I'm pretty sure the Apple Silicon Team (capitalized out of respect) raised a glass every time they could get eight instructions to run simultaneously. But this is how you get the single core speeds that A14/M1 gets without breaking the energy bank. This also means you have to have the arithmetic units to back all that up, and the logic to do all that isn't simple.

Yes, different uses and here we have real problem: Apple's M1 have virtually one use, low power devices. x86 CPUs scale for virtually everything.

Or then not. Firestorm cores are designed for only low clock speeds. Just adding more stuff into CPU architecture does not necessarily make it more sophisticated. It's perhaps more complex but considering Apple does not maintain backwards compatibility and do not need to care about clock scaling, it makes me wonder how sophisticated it really is.

Main question still remains: what Apple M1 does what Intel/AMD cannot do with x86? RISC has some advantages vs x86 and yes Apple has very wide CPU design. But. As we saw with Alder Lake, Intel also made core much wider than Skylake. And Alder Lake's Golden Cove is 2018-2019 design...

Including IP blocks, GPUs, CPUs and all the controller and caches M1 Max has something like 55 billion transistors.

This isn't something Apple came up with overnight - they've been building their own custom silicon and optimizing it for 13 years now. It was only the bugginess of Skylake and the failure of Intel to deliver on promises which drove them to bridge the gap from iPhone to Mac, and convert the Macintosh line to the ARM ISA and their own custom silicon.

And that's huge waste. Meaning yields are very low, Apple just makes component shortage worse. Large majority of those transistors are used for something else than CPU however. Also I disagree that integrating memory into SOC is good idea. It wastes tons of die space and makes memory expansion impossible because Apple didn't want to support it.

 

[Verne Arase]

But why should AMD or Intel abandon x86? I have heard for decades that x86 makes it impossible to create faster CPU's. And for decades AMD and Intel have proven otherwise.

Apple's solutions are much slower than native support from CPU, something happened on x86 very rarely.

Yes, different uses and here we have real problem: Apple's M1 have virtually one use, low power devices. x86 CPUs scale for virtually everything.

Or then not. Firestorm cores are designed for only low clock speeds. Just adding more stuff into CPU architecture does not necessarily make it more sophisticated. It's perhaps more complex but considering Apple does not maintain backwards compatibility and do not need to care about clock scaling, it makes me wonder how sophisticated it really is.

Main question still remains: what Apple M1 does what Intel/AMD cannot do with x86? RISC has some advantages vs x86 and yes Apple has very wide CPU design. But. As we saw with Alder Lake, Intel also made core much wider than Skylake. And Alder Lake's Golden Cove is 2018-2019 design...

And that's huge waste. Meaning yields are very low, Apple just makes component shortage worse. Large majority of those transistors are used for something else than CPU however. Also I disagree that integrating memory into SOC is good idea. It wastes tons of die space and makes memory expansion impossible because Apple didn't want to support it.

True ... they could have two forks - legacy and higher performance.

I'm just saying that if they want to improve single core speeds without burning down the barn, they'll need to find another method of speeding up processing - right now they're both increasing clocks which drastically increases energy use.

Take Alder Lake - sure it's faster, but they achieved it by turning up the heat. Every implementation I've seen now requires water cooling and beefier cooling. As far as I know, that's the only way Intel's produced a CPU with a CPU core which can outrun Firestorm.

You keep talking about x86 like it's the One True CPU (background chorus). It isn't. I've been in the biz for a lot of years, and the One True CPU has been IBM System/360/370/390/3090, Sparc, Motorola 68K, PowerPC, RS/6000, and x86. Your reverence for the architecture probably stems from the fact that you've used it for your entire computing career, but that's only because you've encased yourself in your local consumer computing bubble.

I worked for a major midwestern medical center for 45 years before retirement, and have lived with and administered a wide variety of computing platforms.

ALL CPUs have one use case: computing. x86 is not the beginning of computing and it won't be the end of computing. It won for a while because of it's commodity price - not because of its great efficiency or sterling design.

Apple designed the Apple Silicon SoCs as a large monolithic SoC because they can run 16 and 32 channel memory at 200 GB/sec and 400 GB/sec respectively for the M1 Pro and M1 Max SoCs. Your traditional DIMM based memory is accessed at something like 50 or 80 GB/sec. All IP blocks including the GPU, NPU, and IP blocks including the media engine can all access the memory at the speed of the SoC.

Here's a video of a comparison between an Alder Lake desktop system compared with the M1 Max:

16" MacBook Pro M1 Max VS Intel 12th Gen i7 / RTX 3070 with 8K Video! This is SHOCKING! - YouTube

Apple has avoided a traditional Wintel architecture to bypass some of the Wintel overheads inherent in the CPU <- PCIe bus -> Discrete GPU architecture.

 

[Verne Arase]

True ... they could have two forks - legacy and higher performance.

I'm just saying that if they want to improve single core speeds without burning down the barn, they'll need to find another method of speeding up processing - right now they're both increasing clocks which drastically increases energy use.

Take Alder Lake - sure it's faster, but they achieved it by turning up the heat. Every implementation I've seen now requires water cooling and beefier cooling. As far as I know, that's the only way Intel's produced a CPU with a CPU core which can outrun Firestorm.

You keep talking about x86 like it's the One True CPU (background chorus). It isn't. I've been in the biz for a lot of years, and the One True CPU has been IBM System/360/370/390/3090, Sparc, Motorola 68K, PowerPC, RS/6000, and x86. Your reverence for the architecture probably stems from the fact that you've used it for you entire computing career, but that's only because you've encased yourself in your local consumer computing bubble.

I worked for a major midwestern medical center for 45 years before retirement, and have lived with and administered a wide variety of computing platforms.

ALL CPUs have one use case: computing. x86 is not the beginning of computing and it won't be the end of computing. It won for a while because of it's commodity price - not because of its great efficiency or sterling design.

Apple designed the Apple Silicon SoCs as a large monolithic SoC because they can run 16 and 32 channel memory at 200 GB/sec and 400 GB/sec respectively for the M1 Pro and M1 Max SoCs. Your traditional DIMM based memory is accessed at something like 50 or 80 GB/sec. All IP blocks including the GPU, NPU, and IP blocks including the media engine can all access the memory at the speed of the SoC.

Here's a video of a comparison between an Alder Lake desktop system compared with the M1 Max:

16" MacBook Pro M1 Max VS Intel 12th Gen i7 / RTX 3070 with 8K Video! This is SHOCKING! - YouTube

Apple has avoided a traditional Wintel architecture to bypass some of the Wintel overheads inherent in the CPU <- PCIe bus -> Discrete GPU architecture.

Here's a comparison with an Alder Lake CPU that beats M1 Max which you might think is more fair.

It beats the M1 Max by a fair bit, but look at the system in question.

 

[Danny101]

X86 architecture is inherently inefficient. But performance is achieved by brute force and software tricks at the expense of efficiency. ARM will win out eventually, I believe. Imagine if more developers, and gaming developers started programming more for Linux and Linux on ARM. Don't use Apple. Never will. But they are on the right side I believe with Linux and ARM. I'm still using x86/Windows, but my mindset is shifting and will be thinking about this more before my next platform upgrade. Hopefully, by then software creators will more onboard with the Linux/ARM train. Old habits are difficult to change for sure. I get it. x86/Window is a money ecosystem. So is the Linux/ARM ecosystem as Google and Apple has proven. It also leaves room for the low to no-cost solutions as well. It's a big table, and transparent.

 

[danny102]

I was hoping for a serious review. While apples to apples sometimes works, in this case, it was not particularly helpful. Comparing CPUs to CPUs showed that M1 was a clear winner, and with lower power consumption. Meaning that per watt, you're getting more juice, out of the M1 Pro and less power consumed in that process. That being said, you compared a top-level GPU in the mobile platform, with the M1 Pro's entry level GPU. That was not a fair comparison. Apple's claim was that the M1 Max, with 32/64 Gb Memory, together with 32 GPU cores would be a more substantial comparison with the Nvidia RTX GPUs. Apple's guiding perspective is that after a 5-7 year spell, most people upgrade their computer, and the industry changes that often enough they'll buy a new processor. In which case, they'll buy a new laptop. So upgradability is not necessarily an issue. I'm not saying this is right or wrong. Making PCs totally upgradeable seems like a consumer benefit, but it comes at a cost of power efficiency. In a desktop this is less of an issue, and overall it would make sense that customizability and upgradability are more important in this genre. However, from a perspective of apple users, if you care more about other factors, then these issues are less crucial. Having a computer that lasts for 10 years, without upgrades is more important than customizing and upgrading in that time. It's a different perspective based on a value judgment. You don't have to agree, but it is a differing perspective. It is a well known fact that many software stacks do not work with Mac. This has always been the case, and it probably won't go away. That being said, when I evaluated my workflow I found, that all my apps run natively. I'm a web developer and invest heavily in having local server and dev environments. I also periodically do graphic design and non-intensive video. The most frustrating thing is having to wait while you're working on a graphic/video or in coding. Waiting for compression/encoding is not as frustrating since it's a reality that no matter what you do, you have to wait. You can pick your battles. I use my Mac as my daily driver, and having it instantly start and run programs is more important to me. I'm willing to pay for that, and while some people may regard upgrades as expensive, I have been extremely happy with my mac purchases over the years. Apple is more expensive, yes, per dollar, but I don't have issues. I have worked with Intel based machines for decades, and I find them to have more issues. I have used both stacks. I find that reviews who don't know how to benchmark apple hardware, sometimes miss the boat. Disappointed that there was no M1 Max (10 Core CPU) with 64 GB RAM and 32-core GPU as a comparison. Despite it being expensive in the reviewer's mind, if we would image that in five years, he would have needed to upgrade his laptop anyway, did it matter that he waited 5 years to pay the extra $400 dollars? No.

 

[Verne Arase]

I was hoping for a serious review. While apples to apples sometimes works, in this case, it was not particularly helpful. Comparing CPUs to CPUs showed that M1 was a clear winner, and with lower power consumption. Meaning that per watt, you're getting more juice, out of the M1 Pro and less power consumed in that process. That being said, you compared a top-level GPU in the mobile platform, with the M1 Pro's entry level GPU. That was not a fair comparison. Apple's claim was that the M1 Max, with 32/64 Gb Memory, together with 32 GPU cores would be a more substantial comparison with the Nvidia RTX GPUs. Apple's guiding perspective is that after a 5-7 year spell, most people upgrade their computer, and the industry changes that often enough they'll buy a new processor. In which case, they'll buy a new laptop. So upgradability is not necessarily an issue. I'm not saying this is right or wrong. Making PCs totally upgradeable seems like a consumer benefit, but it comes at a cost of power efficiency. In a desktop this is less of an issue, and overall it would make sense that customizability and upgradability are more important in this genre. However, from a perspective of apple users, if you care more about other factors, then these issues are less crucial. Having a computer that lasts for 10 years, without upgrades is more important than customizing and upgrading in that time. It's a different perspective based on a value judgment. You don't have to agree, but it is a differing perspective. It is a well known fact that many software stacks do not work with Mac. This has always been the case, and it probably won't go away. That being said, when I evaluated my workflow I found, that all my apps run natively. I'm a web developer and invest heavily in having local server and dev environments. I also periodically do graphic design and non-intensive video. The most frustrating thing is having to wait while you're working on a graphic/video or in coding. Waiting for compression/encoding is not as frustrating since it's a reality that no matter what you do, you have to wait. You can pick your battles. I use my Mac as my daily driver, and having it instantly start and run programs is more important to me. I'm willing to pay for that, and while some people may regard upgrades as expensive, I have been extremely happy with my mac purchases over the years. Apple is more expensive, yes, per dollar, but I don't have issues. I have worked with Intel based machines for decades, and I find them to have more issues. I have used both stacks. I find that reviews who don't know how to benchmark apple hardware, sometimes miss the boat. Disappointed that there was no M1 Max (10 Core CPU) with 64 GB RAM and 32-core GPU as a comparison. Despite it being expensive in the reviewer's mind, if we would image that in five years, he would have needed to upgrade his laptop anyway, did it matter that he waited 5 years to pay the extra $400 dollars? No.

A lot of Silicon Valley mainstays are giving fully speced out M1 Max systems to their engineers - they realize that as expensive as they are, an engineer's time is much more expensive.

Heck, by doing this so publicly, they're using 'em as recruitment posters. Come work for Twitter or Uber or Shopify and you'll get one of these.

 

[Verne Arase]

X86 architecture is inherently inefficient. But performance is achieved by brute force and software tricks at the expense of efficiency. ARM will win out eventually, I believe. Imagine if more developers, and gaming developers started programming more for Linux and Linux on ARM. Don't use Apple. Never will. But they are on the right side I believe with Linux and ARM. I'm still using x86/Windows, but my mindset is shifting and will be thinking about this more before my next platform upgrade. Hopefully, by then software creators will more onboard with the Linux/ARM train. Old habits are difficult to change for sure. I get it. x86/Window is a money ecosystem. So is the Linux/ARM ecosystem as Google and Apple has proven. It also leaves room for the low to no-cost solutions as well. It's a big table, and transparent.

Apple uses the ARM ISA, but not the silicon.

They've been working on their own silicon and improving it year by year for 13 years, so don't expect Qualcomm or Samsung to catch up quickly since both use standard ARM cores. And really, they have no reason to - their primary market is smartphones, not PCs.

Qualcomm is at least trying to satisfy Microsoft who is doing their own silicon like Google is - but neither has the silicon design chops and are seconding their efforts to their respective silicon proxies of Qualcomm and Samsung. Qualcomm who (like Samsung) - basically takes standard ARM designed cores and rearranges them on a SoC with some cache configuration changes - has acquired Nuvia which was established by ex-Apple silicon engineers hoping to get some silicon design expertise.

I'm sure Qualcomm is hoping that Nuvia brought with 'em some of that Apple Silicon design, but if they try to use Apple trade secrets and designs they may find themselves locked in litigation.

Really, the great white hope outside of Apple is Intel and/or AMD who do do atomic level silicon design, but again even concentrating focus on an optimizable RISC architecture (of which ARM is one) is a slow and iterative process. I wouldn't expect even these - whose expertise has been in improving x86 performance - to come blazing out the door with industry leading RISC designs; their focus has been in other areas and they'd need to acquire the expertise in low power, very wide CPU design. Just bumping clocks on a RISC design would put them right back where they are in terms of heat and efficiency.

Right now, ironically, Apple has the only low power super wide CPUs around - and you can't just go down to the Apple Silicon store and pick up a CPU to create a competing product.

Even more ironically, that lifestyle brand from Cupertino has some of the best silicon design chops in the industry right now.

I guess those 13 years of silicon design weren't a waste after all .

 

[HardReset]

True ... they could have two forks - legacy and higher performance.

I'm just saying that if they want to improve single core speeds without burning down the barn, they'll need to find another method of speeding up processing - right now they're both increasing clocks which drastically increases energy use.

Take Alder Lake - sure it's faster, but they achieved it by turning up the heat. Every implementation I've seen now requires water cooling and beefier cooling. As far as I know, that's the only way Intel's produced a CPU with a CPU core which can outrun Firestorm.

But why? I have been hearing this "x86 is not good enough against X, Y and Z" for centuries. So far, x86 has held.

Alder Lake was designed to reach at least 5 GHz. Maintaining high IPC while keeping pipeline long enough to reach those speeds is Much harder than staying on 3 GHz range. For that reason alone, Alder Lake is much more impressive than Firestorm.

You keep talking about x86 like it's the One True CPU (background chorus). It isn't. I've been in the biz for a lot of years, and the One True CPU has been IBM System/360/370/390/3090, Sparc, Motorola 68K, PowerPC, RS/6000, and x86. Your reverence for the architecture probably stems from the fact that you've used it for your entire computing career, but that's only because you've encased yourself in your local consumer computing bubble.

I worked for a major midwestern medical center for 45 years before retirement, and have lived with and administered a wide variety of computing platforms.

ALL CPUs have one use case: computing. x86 is not the beginning of computing and it won't be the end of computing. It won for a while because of it's commodity price - not because of its great efficiency or sterling design.

No, I have used many other architecture CPU's than x86 too. There were architectures that were at least some time much better architectures but x86 still was more than good enough to win over all of those.

That also makes my question to stand still: what makes it impossible for x86 CPU to be on par with others?

Apple designed the Apple Silicon SoCs as a large monolithic SoC because they can run 16 and 32 channel memory at 200 GB/sec and 400 GB/sec respectively for the M1 Pro and M1 Max SoCs. Your traditional DIMM based memory is accessed at something like 50 or 80 GB/sec. All IP blocks including the GPU, NPU, and IP blocks including the media engine can all access the memory at the speed of the SoC.

Here's a video of a comparison between an Alder Lake desktop system compared with the M1 Max:

16" MacBook Pro M1 Max VS Intel 12th Gen i7 / RTX 3070 with 8K Video! This is SHOCKING! - YouTube

Apple has avoided a traditional Wintel architecture to bypass some of the Wintel overheads inherent in the CPU <- PCIe bus -> Discrete GPU architecture.

Yeah, buy doing that, Apple ditches memory expansion options and wastes huge amount of wafers (yields are low). That's nothing impressive. There are worldwide shortage of semiconductors and Apple makes it's best to waste 5nm wafers as much as possible

Another question is, what makes it impossible to do same Apple did but with x86?

Here's a comparison with an Alder Lake CPU that beats M1 Max which you might think is more fair.

It beats the M1 Max by a fair bit, but look at the system in question.

And desktop computer with LN2 cooling overclocked can beat even worlds fastest supercomputer on certain loads. Nothing special there.

 

[Verne Arase]

But why? I have been hearing this "x86 is not good enough against X, Y and Z" for centuries. So far, x86 has held.

x86 has held because of it ubiquity - not because of its excellence.

Alder Lake was designed to reach at least 5 GHz. Maintaining high IPC while keeping pipeline long enough to reach those speeds is Much harder than staying on 3 GHz range. For that reason alone, Alder Lake is much more impressive than Firestorm.

Alder Lake has a lot of problems, not the least of which is heat (which it does nothing to alleviate and much to exacerbate).

It seeks to implement a very complex microarchitecture, but leaves dispatching of this complex microarchitecture to the host operating system of another company.

Seems fine now, right? How about next year when the number of SKUs double, or the year after? How long before Microsoft or another kernel vendor screws up because the parameters change?

We've already seen performance degradation of AMD processor speed due to dispatching errors in the Microsoft kernel. How long before this becomes yet another maintenance headache with long PE chains?

That also makes my question to stand still: what makes it impossible for x86 CPU to be on par with others?

x86 is a variable length instruction set, and as such each decoder has to guess where each instruction in the execution queue begins. Guess wrong, and the long logic chain of predictive execution has to be thrown away and started over.

And desktop computer with LN2 cooling overclocked can beat even worlds fastest supercomputer on certain loads. Nothing special there.

We're entering an era when you have to consider the energy efficiency of your computing hardware - I don't know if you've been behind a server rack recently, but it can be unbearable, and consumer computers need to be made more efficient - not less.

Six states have already mandated computer efficiency standards, especially in the western states which are susceptible to wildfires and vulnerable power grids or declining levels of hydroelectric power. There's a world-wide energy shortage rearing its ugly head, and increasingly large and needless expenditure of energy is becoming more and more unwelcome.

Needless to say, it's not just environmentally unfriendly, but expect power rates to spike up as shortages continue, and evacuating heat from your computer chassis just moves it into your room where during the summer you have to spend even more to remove that excess heat your house or office.

California adopts first energy-efficiency standards for PCs in US

Rules set by one of the largest states in the US should have national impact.

arstechnica.com

These mandates call for less than 50 kWh/year energy expenditure for home computers, and with current x86 architectures, higher PCIe speeds and memory, and higher performance discrete GPUs and the power supplies required to power them and cooling required to keep them from burning up you're probably nearing 800 or more watts at peak load. Assuming a 500 watt nominal load and you're talking 100 hour runtime per year to fit under the mandate cap. If you run your computer 24 hours for overnight maintenance tasks and you're talking a little over four days runtime.

 

[Verne Arase]

Except that the M1 laptops cost north of $2000.... and a chromebook costs south of $400.... don't tell me they are in competition....

Actually I've seen them on sale for $850 with a coupon code.

And that will buy you a computer which will last a while and retain resale value, whereas if that chromebook lasts two years we'll know you've sacrificed a virgin to achieve that goal - and it'll be landfill when you're done with it.

 

[Squid Surprise]

Actually I've seen them on sale for $850 with a coupon code.

And that will buy you a computer which will last a while and retain resale value, whereas if that chromebook lasts two years we'll know you've sacrificed a virgin to achieve that goal - and it'll be landfill when you're done with it.

Lol - chromebooks are garbage - I agree with that… but they certainly don’t compete in any way with MacBooks…
It’s like saying a Ferrari competes with a Ford Pinto

 

[HardReset]

x86 has held because of it ubiquity - not because of its excellence.

There were quite many alternatives that made more or less ground (PowerPC, ARM, Sparc, Alpha, Motorola, even Intel's own Itanic etc). Many of those were supposed to be better than x86, at least on supercomputers where backwards compatibility means essentially nothing. Well, x86 is still dominating even there.

Alder Lake has a lot of problems, not the least of which is heat (which it does nothing to alleviate and much to exacerbate).

It seeks to implement a very complex microarchitecture, but leaves dispatching of this complex microarchitecture to the host operating system of another company.

Seems fine now, right? How about next year when the number of SKUs double, or the year after? How long before Microsoft or another kernel vendor screws up because the parameters change?

We've already seen performance degradation of AMD processor speed due to dispatching errors in the Microsoft kernel. How long before this becomes yet another maintenance headache with long PE chains?

Alder Lake problems with Windows is another question. With Alder Lake Intel showed that creating wider x86 core seems to be quite simple thing.

Microsoft always screws up with AMD CPU's, at least on beginning.

x86 is a variable length instruction set, and as such each decoder has to guess where each instruction in the execution queue begins. Guess wrong, and the long logic chain of predictive execution has to be thrown away and started over.

x86 instructions are usually translated into uops before entering actual decoders. Manufacturers are very silent about those uops but I expect them to be fixed length. Basically current x86 CPU's are outside CISC, inside RISC. That solved problem that Should have meant there is no more speed to be gain using x86.

Another solution is to create more instruction sets that "replace" x86 after some time. Like SSE2 on x86-64 instruction set.

We're entering an era when you have to consider the energy efficiency of your computing hardware - I don't know if you've been behind a server rack recently, but it can be unbearable, and consumer computers need to be made more efficient - not less.

Six states have already mandated computer efficiency standards, especially in the western states which are susceptible to wildfires and vulnerable power grids or declining levels of hydroelectric power. There's a world-wide energy shortage rearing its ugly head, and increasingly large and needless expenditure of energy is becoming more and more unwelcome.

Needless to say, it's not just environmentally unfriendly, but expect power rates to spike up as shortages continue, and evacuating heat from your computer chassis just moves it into your room where during the summer you have to spend even more to remove that excess heat your house or office.

California adopts first energy-efficiency standards for PCs in US

Rules set by one of the largest states in the US should have national impact.

arstechnica.com

These mandates call for less than 50 kWh/year energy expenditure for home computers, and with current x86 architectures, higher PCIe speeds and memory, and higher performance discrete GPUs and the power supplies required to power them and cooling required to keep them from burning up you're probably nearing 800 or more watts at peak load. Assuming a 500 watt nominal load and you're talking 100 hour runtime per year to fit under the mandate cap. If you run your computer 24 hours for overnight maintenance tasks and you're talking a little over four days runtime.

Energy efficiency should also apply on manufacturing. Otherwise we soon have computer chips that spend bit less on idle but manufacturing them wastes huge amount of electricity and other resources. Apple's M1 is good example of that. Yields cannot be high considering how huge that chip is and it wastes tons of 5nm die area on wafer.

No wonder Apple have supply problems despite manufacturing wafers dies like crazy.

 

[Danny101]

Apple uses the ARM ISA, but not the silicon.

They've been working on their own silicon and improving it year by year for 13 years, so don't expect Qualcomm or Samsung to catch up quickly since both use standard ARM cores. And really, they have no reason to - their primary market is smartphones, not PCs.

Qualcomm is at least trying to satisfy Microsoft who is doing their own silicon like Google is - but neither has the silicon design chops and are seconding their efforts to their respective silicon proxies of Qualcomm and Samsung. Qualcomm who (like Samsung) - basically takes standard ARM designed cores and rearranges them on a SoC with some cache configuration changes - has acquired Nuvia which was established by ex-Apple silicon engineers hoping to get some silicon design expertise.

I'm sure Qualcomm is hoping that Nuvia brought with 'em some of that Apple Silicon design, but if they try to use Apple trade secrets and designs they may find themselves locked in litigation.

Really, the great white hope outside of Apple is Intel and/or AMD who do do atomic level silicon design, but again even concentrating focus on an optimizable RISC architecture (of which ARM is one) is a slow and iterative process. I wouldn't expect even these - whose expertise has been in improving x86 performance - to come blazing out the door with industry leading RISC designs; their focus has been in other areas and they'd need to acquire the expertise in low power, very wide CPU design. Just bumping clocks on a RISC design would put them right back where they are in terms of heat and efficiency.

Right now, ironically, Apple has the only low power super wide CPUs around - and you can't just go down to the Apple Silicon store and pick up a CPU to create a competing product.

Even more ironically, that lifestyle brand from Cupertino has some of the best silicon design chops in the industry right now.

I guess those 13 years of silicon design weren't a waste after alI

Apple uses the ARM ISA, but not the silicon.

They've been working on their own silicon and improving it year by year for 13 years, so don't expect Qualcomm or Samsung to catch up quickly since both use standard ARM cores. And really, they have no reason to - their primary market is smartphones, not PCs.

Qualcomm is at least trying to satisfy Microsoft who is doing their own silicon like Google is - but neither has the silicon design chops and are seconding their efforts to their respective silicon proxies of Qualcomm and Samsung. Qualcomm who (like Samsung) - basically takes standard ARM designed cores and rearranges them on a SoC with some cache configuration changes - has acquired Nuvia which was established by ex-Apple silicon engineers hoping to get some silicon design expertise.

I'm sure Qualcomm is hoping that Nuvia brought with 'em some of that Apple Silicon design, but if they try to use Apple trade secrets and designs they may find themselves locked in litigation.

Really, the great white hope outside of Apple is Intel and/or AMD who do do atomic level silicon design, but again even concentrating focus on an optimizable RISC architecture (of which ARM is one) is a slow and iterative process. I wouldn't expect even these - whose expertise has been in improving x86 performance - to come blazing out the door with industry leading RISC designs; their focus has been in other areas and they'd need to acquire the expertise in low power, very wide CPU design. Just bumping clocks on a RISC design would put them right back where they are in terms of heat and efficiency.

Right now, ironically, Apple has the only low power super wide CPUs around - and you can't just go down to the Apple Silicon store and pick up a CPU to create a competing product.

Even more ironically, that lifestyle brand from Cupertino has some of the best silicon design chops in the industry right now.

I guess those 13 years of silicon design weren't a waste after all .

Well, I suppose I could target the Windows on ARM laptop that Microsoft and Qualcomm have been collaborating on and install Linux on it. At least it's a start. If more people start doing this, then it could lead to the shift. Programming has to improve. 2 cores at 2.4 Ghz should be equal to 1 at 4.8 Ghz. The reliance on single thread performance in games is abysmal holding things back. More cores at lower speeds should mean more efficiency.

 

[Danny101]

There were quite many alternatives that made more or less ground (PowerPC, ARM, Sparc, Alpha, Motorola, even Intel's own Itanic etc). Many of those were supposed to be better than x86, at least on supercomputers where backwards compatibility means essentially nothing. Well, x86 is still dominating even there.

Alder Lake problems with Windows is another question. With Alder Lake Intel showed that creating wider x86 core seems to be quite simple thing.

Microsoft always screws up with AMD CPU's, at least on beginning.

x86 instructions are usually translated into uops before entering actual decoders. Manufacturers are very silent about those uops but I expect them to be fixed length. Basically current x86 CPU's are outside CISC, inside RISC. That solved problem that Should have meant there is no more speed to be gain using x86.

Another solution is to create more instruction sets that "replace" x86 after some time. Like SSE2 on x86-64 instruction set.

Energy efficiency should also apply on manufacturing. Otherwise we soon have computer chips that spend bit less on idle but manufacturing them wastes huge amount of electricity and other resources. Apple's M1 is good example of that. Yields cannot be high considering how huge that chip is and it wastes tons of 5nm die area on wafer.

No wonder Apple have supply problems despite manufacturing wafers dies like crazy.

Hmm. With all the instruction sets, it seems that one way to deal with the inefficiencies is to do away with the older ones or at least be able to disable the ones you're not using. At that point, x86 becomes a name and not a descriptor. Simplifying the work of the chips. Every new instruction set should be separate instead of encapsulating the old as well. Thereby making it easier to disable unused instruction sets. Just an idea.