M2 Pro and M2 Max will be disappointing?

[theorist9]

Unlike Apple, Intel and AMD make big bucks selling powerful server chips. While performance per watt is important for all chips, for some applications (like server chips) absolute performance is at least as critical. Thus Intel and AMD design choices will inevitably be different from Apple choices. On a flips side, this may explain why Apple is having problems with releasing M-based Mac Pro.

To add to what others have said, efficiency is king in server chips. AWS's Graviton 3, which is clocked at only 2.6 GHz, is a great example.

"Amazon has thus chosen to use TSMC’s cutting edge 5 nm process to reduce power consumption. TSMC’s 7 nm process already did wonders for low power designs, and 5 nm would take this even further. While Graviton 3 is a beefier core than N1, it’s nowhere near as ambitious as Intel’s Golden Cove, and should still be considered a moderate design. Such a core running at 2.6 GHz on 5 nm should absolutely sip power. That in turn lets AWS pack three of these chips into a single node, increasing compute density. The final result is a chip that lets AWS sell each Graviton 3 core at a lower price, while still delivering a significant performance boost over their previous Graviton 2 chip."

Graviton 3: First Impressions

In late May of 2022, AWS released Graviton 3 to the general public. Graviton 3 was the first ARM CPU to introduce the SVE instruction set to a widely accessible server CPU. Before Graviton 3’…

chipsandcheese.com

 

[ChrisA]

In single core performance, the M2 chip only scores 1800 to 1900 in geekbench. At the same time, the mobile / laptop Intel’s 13th gen chip can do 2000 in single core performance. So it might be the case that the upcoming M2 Pro and M2 Max might not even beat the current line up of Intel in raw performance. And AMD is releasing their new Ryzen CPU line also.


So the M2 Pro and M2 Max might not be as revolutionary as the M1 Pro and M1 Max were, as PC laptops might be more powerful this time.

You have to look at Apple's target users. What do they do with their computers? Then ask if the computer is a good match to the use case. Of course Apple could make the single core performance better but what woud be the effect on battery life, weight, cost and noise?

In fact what Apple added was a set of "low performance cores" that use much less power. As it turns out most of what we use our computers for is bast done with slower, less powerful cores.

Only a very tiny percent of Apple users are actually creating digital content, like video editing or music production. Most do a little productivity work, watch movies and read email.

Even for those who do video production with 4K files, they might not need faster CPU cores. The processor is called "Apple Silicon" not "Apple CPU" because there are other processors on the chip than just the CPU cores. Some of them can be media processors. So even to edit video, maybe what they do is put on a few "media cores"

If you can off load 80% of the work a CPU cores does to a special media processor then this has the same effect as making the CPU core 5X faster (for that one app) I think where Apple will gain speed relative to Intel is with these other processors they can put on the chip. It will not show up in benchmarks

 

[Boil]

Even for those who do video production with 4K files, they might not need faster CPU cores. The processor is called "Apple Silicon" not "Apple CPU" because there are other processors on the chip than just the CPU cores. Some of them can be media processors. So even to edit video, maybe what they do is put on a few "media cores"

You know the M1 Pro/Max/Ultra and the M2 already have Media Engines specifically designed for video editing, yeah...?

 

[falainber]

To add to what others have said, efficiency is king in server chips. AWS's Graviton 3, which is clocked at only 2.6 GHz, is a great example.

"Amazon has thus chosen to use TSMC’s cutting edge 5 nm process to reduce power consumption. TSMC’s 7 nm process already did wonders for low power designs, and 5 nm would take this even further. While Graviton 3 is a beefier core than N1, it’s nowhere near as ambitious as Intel’s Golden Cove, and should still be considered a moderate design. Such a core running at 2.6 GHz on 5 nm should absolutely sip power. That in turn lets AWS pack three of these chips into a single node, increasing compute density. The final result is a chip that lets AWS sell each Graviton 3 core at a lower price, while still delivering a significant performance boost over their previous Graviton 2 chip."

Graviton 3: First Impressions

In late May of 2022, AWS released Graviton 3 to the general public. Graviton 3 was the first ARM CPU to introduce the SVE instruction set to a widely accessible server CPU. Before Graviton 3’…

chipsandcheese.com

I am not sure I understand your and Leman's arguments. As I understand, Geekbench multicore score for Graviton 3 is 50739. That's way higher than the score for M2 (8965). Thus using Graviton for proving that server chips do not need to be powerfull seems illogical at the very least. And yes, Graviton uses more power than M chips.

 

[bcortens]

I am not sure I understand your and Leman's arguments. As I understand, Geekbench multicore score for Graviton 3 is 50739. That's way higher than the score for M2 (8965). Thus using Graviton for proving that server chips do not need to be powerfull seems illogical at the very least. And yes, Graviton uses more power than M chips.

The performance per watt is what matters here - yes the graviton uses more power but like the Xeon’s it does not win in single threaded performance. Instead it uses many cores to achieve a very high performance per watt.
The point isn’t that performance of graviton or Xeon’s don’t matter, it’s that they achieve that performance at a much lower power draw than if they were configured in the same manner as an Intel enthusiast chip. The Intel enthusiast chip will beat both graviton and Xeon in single-thread tests but be demolished in performance per watt because Intel pushes their enthusiast chips past the performance per watt sweet spot.

 

[theorist9]

I am not sure I understand your and Leman's arguments. As I understand, Geekbench multicore score for Graviton 3 is 50739. That's way higher than the score for M2 (8965). Thus using Graviton for proving that server chips do not need to be powerfull seems illogical at the very least. And yes, Graviton uses more power than M chips.

The OP's original post, and correspondingly, most of this thread, has been about single-core performance. Thus when you wrote...

While performance per watt is important for all chips, for some applications (like server chips) absolute performance is at least as critical.

...we took it to mean that you were arguing per-core performance is critical for server chips. In fact, it's not; server chips tend to sacrifice per-core performance in order to gain higher efficiency. They thus get their high multicore performance, which you cited, by combining lots of slower (and thus more efficient) cores.

I.e., as a rule of thumb, performance increases linearly with clock speed, but power consumption increases quadratically with clock speed.

 

[falainber]

The performance per watt is what matters here - yes the graviton uses more power but like the Xeon’s it does not win in single threaded performance. Instead it uses many cores to achieve a very high performance per watt.
The point isn’t that performance of graviton or Xeon’s don’t matter, it’s that they achieve that performance at a much lower power draw than if they were configured in the same manner as an Intel enthusiast chip. The Intel enthusiast chip will beat both graviton and Xeon in single-thread tests but be demolished in performance per watt because Intel pushes their enthusiast chips past the performance per watt sweet spot.

As I said before: performance per watt matters but that's not the only thing that matters. For certain applications (even on servers) max single core performance matters too (it has to be acceptable). More importantly, total (multicore) performance definitely matters. If, say, someone could design server chip with superior performance per watt but with low absolute multicore performance (say, 1/10 of the typical CPU) such a chip would not be very useful in server farm for one would have to put 10x more chips to get the same performance and most likely server design would require much more space (with negative consequences for infrastructure and power).

 

[maflynn]

As I said before: performance per watt matters but that's not the only thing that matters.

I don't think it matters as much as people here think it does. Even with this era of high energy costs, it may not be the factor that causes someone to choose apple over one of its competitors - just my opinion

 

[jdb8167]

I don't think it matters as much as people here think it does. Even with this era of high energy costs, it may not be the factor that causes someone to choose apple over one of its competitors - just my opinion

How much does it cost to charge a 50W battery? Not too much. I doubt the cost of energy is very motivating for determining what notebook to buy. But an important consequence of having high performance per Watt is longer battery life with a smaller and thus lighter battery. I think that is more likely what a buyer cares about.

 

[leman]

I don't think it matters as much as people here think it does. Even with this era of high energy costs, it may not be the factor that causes someone to choose apple over one of its competitors - just my opinion

Why do you think Apple Silicon is so popular with developers? It’s hard to overstate the value having a compact work laptop that can compile code or run unit tests as fast as a large desktop workstation. Thats what efficiency is all about.

For an average business or education user performance is a secondary concern, but the MBA is again an attractive laptop exactly because if it’s efficiency. You don’t have to sacrifice either the form factor or the battery to get extremely smooth operation. This is what efficiency enables.

 

[leman]

As I said before: performance per watt matters but that's not the only thing that matters. For certain applications (even on servers) max single core performance matters too (it has to be acceptable). More importantly, total (multicore) performance definitely matters. If, say, someone could design server chip with superior performance per watt but with low absolute multicore performance (say, 1/10 of the typical CPU) such a chip would not be very useful in server farm for one would have to put 10x more chips to get the same performance and most likely server design would require much more space (with negative consequences for infrastructure and power).

Talking about performance per watt comparison is only really useful if either the performance or the total power are the same between two products. So while your argument is absolutely accurate it’s also a bit redundant? I mean, if someone comes to you and claims that a snapdragon is more efficient in multicore operation than a Threadripper you’d probably look at them like they are crazy.

So going back to the real world, performance per watt is the central metric for a server CPU precisely because it allows you to have more performance at the same power target. Other metrics of course are performance per area (dollar, core etc.). In the context of Apple Silicon, performance per watt means that you can get the performance of the competitor but at a fraction of power usage, which means compact products.

 

[maflynn]

Why do you think Apple Silicon is so popular with developers? It’s hard to overstate the value having a compact work laptop that can compile code or run unit tests as fast as a large desktop workstation. Thats what efficiency is all about.

My opinion was more about what consumers are looking to buy. That segment imo, is the largest customers apple has, not developers.

Macs have been the preferred computer for developers for so many years and Apple's move to ARM didn't have a material impact on getting developers to switch. I could be wrong, and if you have statistics to show windows/linux developers moving to Macs in mass, I'd love to see it. To put it another way, developers that wanted a Mac were going to buy one either, regardless if Apple used Intel or ARM. Those that were (or are) on Linux or Windows were probably not largely swayed to buy a Mac.

 

[dugbug]

seriously can we get back to synthetic chess benchmark debates

 

[theorist9]

I don't think it matters as much as people here think it does. Even with this era of high energy costs, it may not be the factor that causes someone to choose apple over one of its competitors - just my opinion

By focusing on the reduction in energy costs, I think you're missing what, for most Mac buyers, are the main benefits of increased efficiency.

Let's focus on laptops (as opposed to desktops), since they are both the bigger market segment, and also where efficiency makes the biggest difference. Here, in addition to reduced energy costs, efficiency gives four things:

1.The ability to pack more performance in a smaller form factor.
2. Significantly increased battery life.
3 (and related to the last one): The ability to sustain high performance while on battery.
4. Quietness.

I'm not disimissing energy costs, but I think energy costs are the least of it—particularly for Mac buyers who, because of the cost of the product, probably have a higher average income than the general population. Besides, laptops don't use that much energy regardless, so the difference in cost should be small.

Having said that, while efficiency is big, it is not the biggest thing--that is someone's OS preference. If they strongly prefer Windows, they're not going to switch to Mac because of better hardware (and visa-versa). However, where it becomes interesting is if someone is OS-agonstic, or has been considering switching to the Mac. For those consumers at the margins, this might just be what motivates them to change.

 

[Colstan]

Having said that, while efficiency is big, it is not the biggest thing--that is someone's OS preference. If they strongly prefer Windows, they're not going to switch to Mac becasue of better hardware (and visa-versa). However, where it becomes interesting is if someone is OS-agonstic, or has been considering switching to the Mac. For those consumers at the margins, this might just be what motivates them to switch.

This is the answer. We tech nerds tend to twist ourselves into knots over tech jargon, but it's the operating system that defines the user experience. There are a lot of "concerned" posts around here about Apple not being able to keep up with Intel, AMD, and Nvidia. This forum gets at least one or two a week ever since Apple Silicon was announced, like clockwork.

Apple isn't selling you an M(x) processor; they're selling you a Mac with macOS, with the benefits of the SoC tossed into the mix. We get so distracted by the gewgaws and doodads that it's easy to forget that Apple's vertical integration strategy is the primary pull.

I'm interested in learning about new technologies, but in practical terms, I personally don't give two craps about what the PC companies are offering because I'm never going to switch to their products. I'm a macOS user, I like the Mac experience. Nothing in these posts showing grave concern over Apple's future (while getting a case of the vapors) is going to get me to switch, full stop. Perhaps there are folks around the margins, but they probably aren't visiting an enthusiast forum, this is MacRumors, and most of us have hardened opinions on these subjects.

Besides, considering that Mac revenue increased by 25% last quarter, and around half of Mac purchases were by customers new to the platform, I'd say Apple is doing just fine at attracting people away from Windows PCs. I'm sure that won't stop the PC partisans from plying their trade around these parts, in fact, it will probably just increase in intensity as Apple continues to gain marketshare.

 

[theorist9]

This is the answer. We tech nerds tend to twist ourselves into knots over tech jargon, but it's the operating system that defines the user experience. There are a lot of "concerned" posts around here about Apple not being able to keep up with Intel, AMD, and Nvidia. This forum gets at least one or two a week ever since Apple Silicon was announced, like clockwork.

Apple isn't selling you an M(x) processor; they're selling you a Mac with macOS, with the benefits of the SoC tossed into the mix. We get so distracted by the gewgaws and doodads that it's easy to forget that Apple's vertical integration strategy is the primary pull.

I'm interested in learning about new technologies, but in practical terms, I personally don't give two craps about what the PC companies are offering because I'm never going to switch to their products. I'm a macOS user, I like the Mac experience. Nothing in these posts showing grave concern over Apple's future (while getting a case of the vapors) is going to get me to switch, full stop. Perhaps there are folks around the margins, but they probably aren't visiting an enthusiast forum, this is MacRumors, and most of us have hardened opinions on these subjects.

Besides, considering that Mac revenue increased by 25% last quarter, and around half of Mac purchases were by customers new to the platform, I'd say Apple is doing just fine at attracting people away from Windows PCs. I'm sure that won't stop the PC partisans from plying their trade around these parts, in fact, it will probably just increase in intensity as Apple continues to gain marketshare.

I'm like you. My preference for MacOS over Windows is so strong (and has such an influence on my ease of use and productivity) that I'd probably choose a Mac even if it ran on vacuum tubes.

The twist I'd add is that, even though I personally wouldn't switch to Windows, it is important (at least to me) how Macs perform compared to PC's:

1) Better Mac performance relative to PC's increases Mac conversion, and thus the size of the Mac user base, which ensures the continued robustness of the Mac ecosystem (including the percent of programs that run on Macs, as opposed to being Windows-only).

2) As PC's have gotten speedier, software overhead has increased--that's why it's tough to run new software on older machines, even if they are technically compatible. If Macs don't keep up with that evolution it will become increasingly difficult to run modern software on the Mac, since most software is probably designed with PC capability in mind.

3) I can't deny the psychological element of knowing that, by choosing the Mac, you're not giving up performance (and may in fact be gaining it).

 

[exoticSpice]

This is the answer. We tech nerds tend to twist ourselves into knots over tech jargon, but it's the operating system that defines the user experience. There are a lot of "concerned" posts around here about Apple not being able to keep up with Intel, AMD, and Nvidia. This forum gets at least one or two a week ever since Apple Silicon was announced, like clockwork.

Apple isn't selling you an M(x) processor; they're selling you a Mac with macOS, with the benefits of the SoC tossed into the mix. We get so distracted by the gewgaws and doodads that it's easy to forget that Apple's vertical integration strategy is the primary pull.

I'm interested in learning about new technologies, but in practical terms, I personally don't give two craps about what the PC companies are offering because I'm never going to switch to their products. I'm a macOS user, I like the Mac experience. Nothing in these posts showing grave concern over Apple's future (while getting a case of the vapors) is going to get me to switch, full stop. Perhaps there are folks around the margins, but they probably aren't visiting an enthusiast forum, this is MacRumors, and most of us have hardened opinions on these subjects.

Besides, considering that Mac revenue increased by 25% last quarter, and around half of Mac purchases were by customers new to the platform, I'd say Apple is doing just fine at attracting people away from Windows PCs. I'm sure that won't stop the PC partisans from plying their trade around these parts, in fact, it will probably just increase in intensity as Apple continues to gain marketshare.

macOS has been very buggy. Vuntura being the worst offender. I am NOT having a great user experience in Ventura.

 

[Colstan]

I'm like you. My preference for MacOS over Windows is so strong (and has such an influence on my ease of use and productivity) that I'd probably choose a Mac even if it ran on vacuum tubes.

I like that metaphor, I might steal it some day.

The twist I'd add is that, even though I personally wouldn't switch to Windows, it is important (at least to me) how Macs perform compared to PC's:

1) Better Mac performance relative to PC's increases Mac conversion, and thus the size of the Mac user base, which ensures the continued robustness of the Mac ecosystem (including the percent of programs that run on Macs, as opposed to being Windows-only).

2) As PC's have gotten speedier, software overhead has increased--that's why it's tough to run new software on older machines, even if they are technically compatible. If Macs don't keep up with that evolution it will become increasingly difficult to run modern software on the Mac, since most software is probably designed with PC capability in mind.

3) I can't deny the psychological element of knowing that, by choosing the Mac, you're not giving up performance (and may in fact be gaining it).

I agree with everything you say here. I also very much want Apple to outperform the other semiconductor manufactures with the M-series. I was more addressing the constant doomsayers who come in here with the latest version of "Apple is doomed" that we've heard for decades.

I understand the legitimate concern that Apple could fall behind, I don't dismiss that, unlike a lot of the background noise that we are often subjected to. However, I haven't seen any reason to raise the alarm, as of yet. I obviously have no inside sources, but it's easy to forget that the Mac is still in the beginning stages of the transition, and we're waiting on Apple to show us what they are really capable of. Right now, we've basically got version 1.5 silicon with the handful of base M2 models. The next Mac Pro is waiting inside Apple's skunkworks and we've still got an Intel Mac mini that needs replacing.

I'm sure various global factors, which we are all aware of, have had an impact on the M-series roadmap, but that goes for everyone else in the industry. I think a lot of the frustration with Apple and the Mac is a result of their secrecy. The PC guys leak all over the place, very messy, but we know their plans a year or more ahead of time. Apple remains an inscrutable black box until they decide to launch. That results in many months of no news to feed us tech nerds who want all the details right now, otherwise we start plucking our feathers out in frustration. I'm just as guilty of that as everyone else here.

The next Mac event isn't likely to be until around March, which means speculation and rumor take hold. It's entirely possible that the Apple Silicon Mac Pro won't make an appearance until WWDC. Since Apple doesn't pre-announce chip plans like the PC companies, we're left in a vacuum, and the leakers simply don't have the reliable sources they once did. So, all we can do is speculate, which explains some of the angst and tumult around here from people who genuinely care about the Mac platform. If Apple is still in this gully by the second half of next year, then there is reason for concern, but I'm not betting against Johny Srouji and his team inside Apple.

 

[DeepIn2U]

Most on this thread realize this, but it bears repeating:

When Apple designed their M-series chips they considered these two key points:

1) Laptop sales are much larger than desktop sales, and their most popular Mac, overall, is a small fanless laptop—the Air.

2) Most programs are single-threaded, and people rarely run more than a few programs at a time. Thus, for the overwhelming majority of use cases, when it comes to CPU performance, what matters is SC rather than MC speed.

Thus they designed a chip that allowed their laptops generally, and their most popular and least expensive laptop in particular, to have a CPU that effectively outperformed anything found in any laptop PC, for the overwhelming majority of applications...while being quiet and offering extraordinary battery life.

This was brilliant. The two areas where they fall short are top-end GPU performance, and top-end SC desktop speed. Their SC performance is close enough to the fastest x86 machines that it's probably not worth giving up their massive efficiency advantage to equalize that. But with the GPU there's a bigger gap, and it will be interesting to see what they do with the 3 nm M2's in Spring 2023.

This makes no sense.

1. There was only 1 model MBA that was fabless when Apple was designing their Apple Silicon chips - the Intel M3 chip and it was NOT that popular.

2. Single threaded apps and the chips they run on have NOT increased in performance since beyond 10 yes ago regardless of cpu design or speed; single threaded performance has been at a snails pace ever since.

From Geekbench score version to the next despite the cpu single core at speed compared rarely ever jumps. Including real world use - after initial launch those single core office apps and the like haven’t really gotten any faster and these are the soul sucking time sucking apps!

 

[theorist9]

This makes no sense.

1. There was only 1 model MBA that was fabless when Apple was designing their Apple Silicon chips - the Intel M3 chip and it was NOT that popular.

2. Single threaded apps and the chips they run on have NOT increased in performance since beyond 10 yes ago regardless of cpu design or speed; single threaded performance has been at a snails pace ever since.

From Geekbench score version to the next despite the cpu single core at speed compared rarely ever jumps. Including real world use - after initial launch those single core office apps and the like haven’t really gotten any faster and these are the soul sucking time sucking apps!

Dude, this is such incoherent word salad that I'm not going to waste my time trying to reply.

 

[DeepIn2U]

Dude, this is such incoherent word salad that I'm not going to waste my time trying to reply.

But you read it and understand it. No debate shows I’m most likely right in terms of single threaded apps not increasing in performance regardless of os nor the chip used.

 

[bcortens]

But you read it and understand it. No debate shows I’m most likely right in terms of single threaded apps not increasing in performance regardless of os nor the chip used.

Single threaded performance in real world use cases has increased, many things are snappier with M1 than on the Intel Macs they replaced and I don't even know where you get the idea that 10 years ago single threaded performance was on par with today.

For example: https://www.anandtech.com/bench/product/2659?vs=2908
While many of the benchmarks listed can take advantage of more than 1 core where you do look at single threaded tasks it is usually more than double the performance. And that is a 4770K not even sandy bridge!

It's so out of touch with reality to claim that ST performance hasn't changed...

 

[DeepIn2U]

Single threaded performance in real world use cases has increased, many things are snappier with M1 than on the Intel Macs they replaced and I don't even know where you get the idea that 10 years ago single threaded performance was on par with today.

For example: https://www.anandtech.com/bench/product/2659?vs=2908
While many of the benchmarks listed can take advantage of more than 1 core where you do look at single threaded tasks it is usually more than double the performance. And that is a 4770K not even sandy bridge!

It's so out of touch with reality to claim that ST performance hasn't changed...

So an Excel workbook with a few add-Ons and multiple cells linking to other sheets on a shared drive, or Outlook moving folders into others or rules or navigation really that much faster?! That’s the “in touch” today vs 10ths ago I’m talking about. Not numbers on a spec sheet.

 

[bcortens]

So an Excel workbook with a few add-Ons and multiple cells linking to other sheets on a shared drive, or Outlook moving folders into others or rules or navigation really that much faster?! That’s the “in touch” today vs 10ths ago I’m talking about. Not numbers on a spec sheet.

Depends on how long the task takes - some of the tasks you mentioned are so trivial as to be instantaneous 10 years ago vs today, people don't complain about their computers feeling slow at certain tasks if those tasks are instantaneous.

The single threaded performance most people 'feel' on a day to day basis is the web, JS performance has improved not just on spec sheets but in real terms. Lightroom is not a spec sheet task, my wife uses it regularly and her newer Mac has lead to radically improved responsiveness in her editing tasks...

 

[leman]

But you read it and understand it. No debate shows I’m most likely right in terms of single threaded apps not increasing in performance regardless of os nor the chip used.

This does t make any sence. My M1 machine runs my single-threaded R scripts approximately 30-50% faster than my previous laptop with top Intel CPU. The single-threaded performance almost tripled in last 10 years. And critical user-facing software components are larger single-threaded.

Also, what the hell is a “fabless M3 chip”?