Leaked MacBook Air GB5 benchmark shows score higher than 16-inch MacBook Pro; SC higher than 5950X

[NotTheFirstDaniel]

Summary

A leaked benchmark of the MacBook Air 10,1 has appeared on Geekbench database showing a single-core score of 1687 and a multi-core score of 7433. Notably, this is score higher than any other Apple Silicon, and the 16-inch MacBook Pro with an Intel Core i9-9980HK. What's also to note is that the single-core score is higher than any Mac ever released, even the $59,999 Mac Pro, and in fact is one of the best single-core scores in the industry, only slightly edging out the just released Ryzen 9 5950X...

 

on a laptop without a fan.

 

The Mac mini result appears to be on the database as well, but shows a similar score.

 

Quotes

Quote

"The M1 chip, which belongs to a ‌MacBook Air‌ with 8GB RAM, features a single-core score of 1687 and a multi-core score of 7433. According to the benchmark, the M1 has a 3.2GHz base frequency."

 

"In comparison to Macs, the single-core performance is better than any other available Mac, and the multi-core performance beats out all of the 2019 16-inch MacBook Pro models, including the 10th-generation high-end 2.4GHz Intel Core i9 model. That high-end 16-inch MacBook Pro earned a single-core score of 1096 and a multi-core score of 6870."

 

My thoughts

Five words. Moore's law is not dead.

 

It could be possible that the M1 breaks Geekbench 5 in a way CPUs broke Geekbench 4 and a theoretical Geekbench 6 may be better in line to compare these chips with x86 chips, but at face value holy shit. On a LAPTOP?! Imagine what they can do on a fucking Mac Pro... 

 

What's next? The M1 GPU can outperform a 1660Ti with a fraction of the power?

 

I assume the difference between the MBP/Mac mini and the MBA is that the former can run sustained without throttling. I refuse to believe Apple broke the laws of physics. But then that actually bothers me. They must've limited the chip to the power constraints of the MBA. Why not expand it for the Mac mini? It has an actual cooling system. There's much more thermal headroom for the Mac mini for it to run faster. I just assume the M1X and M1Z must be too power-hungry to fit in the Mac mini so they just put the M1 in it for now.

 

I personally believe these results to be valid, I mean remember this isn't x86 code that's going to be running on this. macOS itself is processor independent, so the system isn't running any emulated components, and all the apps around it and built for ASi will use ARM instructions. Also, if an x86 instruction requires 3 ARM equivalent instructions, is an ARM CPU with 3.1x IPC than the x86 CPU not definitively faster? Food for thought.

 

I apologize for getting a little excited, now bring me back to reality! It just seems too good to be true. So good that it's probably true...

Sources

https://www.macrumors.com/2020/11/11/m1-macbook-air-first-benchmark/

https://browser.geekbench.com/v5/cpu/4648107

https://browser.geekbench.com/v5/cpu/4648680

[WindirBear]

I thought macbooks thermal throttled all the time before even utilizing it's CPUs full potential. I'm not too familiar with apple computers I may be wrong

[Rooked]

4 minutes ago, NotTheFirstDaniel said:

Summary

A leaked benchmark of the MacBook Air 10,1 has appeared on Geekbench database showing a single-core score of 1687 and a multi-core score of 7433. Notably, this is score higher than any other Apple Silicon, and the 16-inch MacBook Pro with an Intel Core i9-9980HK. What's also to note is that the single-core score is higher than any Mac ever released, even the $59,999 Mac Pro, and in fact is one of the best single-core scores in the industry, only slightly edging out the just released Ryzen 9 5950X...

 

on a laptop without a fan.

 

The Mac mini result appears to be on the database as well, but shows a similar score.

 

Quotes

 

My thoughts

Five words. Moore's law is not dead.

 

It could be possible that the M1 breaks Geekbench 5 in a way CPUs broke Geekbench 4 and a theoretical Geekbench 6 may be better in line to compare these chips with x86 chips, but at face value holy shit. On a LAPTOP?! Imagine what they can do on a fucking Mac Pro... 

 

What's next? The M1 GPU can outperform a 1660Ti with a fraction of the power?

 

I assume the difference between the MBP/Mac mini and the MBA is that the former can run sustained without throttling. I refuse to believe Apple broke the laws of physics.

 

I personally believe these results to be valid, I mean remember this isn't x86 code that's going to be running on this. macOS itself is processor independent, so the system isn't running any emulated components, and all the apps around it and built for ASi will use ARM instructions. Also, if an x86 instruction requires 3 ARM equivalent instructions, is an ARM CPU with 3.1x IPC than the x86 CPU not definitively faster? Food for thought.

 

I apologize for getting a little excited, now bring me back to reality!

Sources

https://www.macrumors.com/2020/11/11/m1-macbook-air-first-benchmark/

https://browser.geekbench.com/v5/cpu/4648107

https://browser.geekbench.com/v5/cpu/4648680

Lower CPU count makes it easier to get a better single core. Ryzen 9 5950x has double the multicore score. But I think it's still pretty cool, sorry for bothering.

[NotTheFirstDaniel]

Just now, WindirBear said:

I thought macbooks thermal throttled all the time before even utilizing it's CPUs full potential. I'm not too familiar with apple computers I may be wrong

With that definition, almost every laptop thermal throttles.

 

For me, thermal throttling is when a CPU drops below base clock to regain thermals. The only MacBook not to do that in a long time is the 2019 15-inch MBP and the 16-inch MBP, and those inch near that mark.

[NotTheFirstDaniel]

2 minutes ago, Rooked said:

Lower CPU count makes it easier to get a better single core.

I assume that's why Intel kept making quad cores even after the Zen destruction?

 

I'm not questioning legitimacy, I just genuinely want to know.

[Belgarathian]

I'm so excited for reviews to start coming out about the new range of products, and for Anandtech's deep-dive into the M1.

[Rooked]

1 minute ago, NotTheFirstDaniel said:

I assume that's why Intel kept making quad cores even after the Zen destruction?

 

I'm not questioning legitimacy, I just genuinely want to know.

Yeah. They have the 14nm architecture, they can't really get more cores. Also they're advertised as the best gaming CPU manufacturer (not anymore, because of the god Lisa Su and her engineers. You can call me an AMD fanboy all you want, I have only built AMD PC's.)

[Blademaster91]

14 minutes ago, WindirBear said:

I thought macbooks thermal throttled all the time before even utilizing it's CPUs full potential. I'm not too familiar with apple computers I may be wrong

The intel macbooks throttle quite badly, if they can't sustain anything but base clocks and still reach over 90-100C then thats throttling, IMO.

I think Apple has been purposely holding back their Intel laptop performance with crap cooling on purpose for a while, then apple could say their ARM laptops are significantly faster.

For example the 2020 macbook air, it has a useless cooling fan with no heatpipes connected to it.

[NotTheFirstDaniel]

2 minutes ago, Blademaster91 said:

The intel macbooks throttle quite badly, if they can't sustain anything but base clocks and still reach over 90C then thats throttling, IMO.

I think Apple has been purposely holding back their Intel laptop performance with crap cooling on purpose for a while, then apple could say their ARM laptops are significantly faster.

For example the 2020 macbook air, it has a useless cooling fan with no heatpipes connected to it.

Let me try to explain what I think happened.

 

When the 2016 MacBook Pro came out, I think Apple expected Intel's 10nm die shrink known as "Kaby Lake" to be released alongside the MacBooks (It wouldn't have been the first time new 3rd part hardware premiered with a Mac, Vega is one of them). They designed the thinner MacBook Pro with the 10nm chips in mind. But by 2016 it was apparent that Intel wasn't going to have its 10nm out by then. I think this was the first time they pushed back the 10nm die shrink. By this time, the design was already in place, mass production started on parts, it was too late to call it back and make adjustments. So they went ahead with it assuming next year they'd get their shit together. Fast forward till today and they haven't.

 

Also, around 2016/2017 was the time when Apple allegedly stared looking into using their own processors for their computers. We saw this with the T2 chip, taking more control away from the CPU and onto Apple Silicon. Apple probably figured to wait out the four years, because while Intel CPUs were hot, they weren't that hot, right?

 

That's where Zen comes in. Zen forced the core-count to increase on all Intel CPUs, negatively affecting Apple. The design that could barely hold together a quad-core CPU was now expected to hold a hexa-core and octa-core CPU. This would not do. Apple had to cut their ego and made a laptop slightly thicker than the one before it. I think that was the last straw, Apple admitting defeat to Intel. We all know that if there's one thing Apple hates doing it's admitting fault.

 

And yeah, I don't think anyone in the world can explain the MacBook Air fan fiasco. I mean you can argue the 8th-gen chips were fanless, and Apple didn't expect the 10th-gen chips to jump up in power usage, but meh. Its a fan and a heatsink, get it together...

[Slottr]

When are reviewers releasing their stuff? Do we know? 

[NotTheFirstDaniel]

Just now, Slottr said:

When are reviewers releasing their stuff? Do we know? 

The new Macs ship as early as next week. It's possible Apple has an embargo lift for them, I'd look out for that either tomorrow (That's when Big Sur comes out) or Monday/Tuesday. I think Tuesday/Wednesday is when we get independent first impressions.

[Loote]

It's probably going to be really hard to compare, GB scores I think promote ARM processors a lot, we'll probably get Macs working great with some apps and horrendously slow with others. Honestly I am conflicted about it because ARM might be great for CPUs, but having Apple as the top option in any field is just terrible.

[Spindel]

PCMR people:

 

”BWAH! A test of a CPUs general purpose computing favors a CPU good at general purpose computing. 
 

Not fair!”

[panzersharkcat]

3 hours ago, Blademaster91 said:

For example the 2020 macbook air, it has a useless cooling fan with no heatpipes connected to it.

The heat sink doesn't even make contact with the CPU. 

[saltycaramel]

So the claim of best single core on earth is confirmed. Ooops.

 

I’ll corroborate my “utterly destroy” comment from the other thread here, now that we have actual benchmarks, instead of Anandtech’s thought experiment. 

 

This M1 is the bottom of the barrel of AS Mac CPUs. It’s the lowest end, for the 13” thin and light, the entry level SFF desktop and the low end 2-port 13” Macbook Pro (whereas the “powerful” 4-port 13” MBP is yet to be superseded, they still sell the Intel one). It’s estimated to have a ~15W TDP.

 

And yet, in SC it beats fully fledged, time-fold wattage, fully-cooled desktop CPUs. 

 

And in MC it easily beats the i9-9980HK 45W mobile 8-core CPU (and keep in mind the Intel part has 8 actual regular cores, whereas the M1 has 4 performance cores and 4 low power cores for light background tasks). 

 

Now, this is what we’re probably gonna see in the next few months, based on rumors:

- 14” MBP with 12 cores (8 performance + 4 low power)

- 16” MBP with 12 cores (8 performance + 4 low power) and maybe a discrete GPU by Apple

- 24” miniLED iMac with 16 cores (12 performance + 4 low power) + discrete Apple GPU + user upgradable RAM on the back probably

 

Now I ask, if these are the benchmarks we are seeing today with just 4 performance cores, what benchmarks are we gonna see in 3-4 months on the 8 performance cores and 12 performance cores versions of the M1? (or whatever they’re gonna be called, like M1X, M1Z, etc.)

 

By the look of it, they could be pretty impressive to say the least.

 

So, skeptics, enjoy the next 3-4 months comparing Apple’s 15W ultrabook CPU to Intel/AMD desktop CPUs while it lasts, I guess

[Brooksie359]

8 minutes ago, saltycaramel said:

So the claim of best single core on earth is confirmed. Ooops.

 

I’ll corroborate my “utterly destroy” comment from the other thread here, now that we have actual benchmarks, instead of Anandtech’s thought experiment. 

 

This M1 is the bottom of the barrel of AS Mac CPUs. It’s the lowest end, for the 13” thin and light, the entry level SFF desktop and the low end 2-port 13” Macbook Pro (whereas the “powerful” 4-port 13” MBP is yet to be superseded, they still sell the Intel one). It’s estimated to have a ~15W TDP.

 

And yet, in SC it beats fully fledged, time-fold wattage, fully-cooled desktop CPUs. 

 

And in MC it easily beats the i9-9980HK 45W mobile 8-core CPU (and keep in mind the Intel part has 8 actual regular cores, whereas the M1 has 4 performance cores and 4 low power cores for light background tasks). 

 

Now, this is what we’re probably gonna see in the next few months, based on rumors:

- 14” MBP with 12 cores (8 performance + 4 low power)

- 16” MBP with 12 cores (8 performance + 4 low power) and maybe a discrete GPU by Apple

- 24” miniLED iMac with 16 cores (12 performance + 4 low power) + discrete Apple GPU + user upgradable RAM on the back probably

 

Now I ask, if these are the benchmarks were seeing today with just 4 performance cores, what benchmarks are we gonna see in 3-4 months on the 8 performance cores and 12 performance cores versions of the M1? (or whatever they’re gonna be called, like M1X, M1Z, etc.)

 

By the look of it, they could be pretty impressive to say the least.

 

So, skeptics, enjoy the next 3-4 months comparing a 15W ultrabook Apple CPU to Intel/AMD desktop CPUs while it lasts, I guess

I would prefer the single core benchmark on cinebench r20 tbh but hopefully we will see that when official reviews are out. 

[Blademaster91]

20 minutes ago, panzersharkcat said:

The heat sink doesn't even make contact with the CPU. 

Yeah that is the video I was referring to, it seems like Apple designed that heatsink for their ARM chip, not an Intel CPU.

More on topic though I'd really like to see Cinebench scores, and multicore performance, because geekbench isn't the most realistic benchmark.

[panzersharkcat]

1 minute ago, Blademaster91 said:

Yeah that is the video I was referring to, it seems like Apple designed that heatsink for their ARM chip, not an Intel CPU.

More on topic though I'd really like to see Cinebench scores, and multicore performance, because geekbench isn't the most realistic benchmark.

Yeah, I was surprised by that. I knew about the fan not being connected to the heat sink because of Louis Rossmann. I though it'd at least make contact, though.

[Bombastinator]

4 hours ago, WindirBear said:

I thought macbooks thermal throttled all the time before even utilizing it's CPUs full potential. I'm not too familiar with apple computers I may be wrong

Many did.  Iirc the 16” was an exception, at least for some models.  Afaik it was the only macbook model in years that had decent cooling.   I’m in the “benchmark got broken” camp myself.  We’ll know soon enough one way or the other.

Edited November 12, 2020 by Bombastinator

[LAwLz]

2 hours ago, Loote said:

It's probably going to be really hard to compare, GB scores I think promote ARM processors a lot, we'll probably get Macs working great with some apps and horrendously slow with others. Honestly I am conflicted about it because ARM might be great for CPUs, but having Apple as the top option in any field is just terrible.

The reason why GeekBench "promote ARM processors a lot" is because the tests are very, VERY CPU intense but barely puts any stress on the memory.

Basically, the entire test can be loaded into the cache and then run from there, while more realistic workloads requires a lot of access to main memory, where traditionally x86 have had an advantage compared to ARM devices.

 

If you want a CPU benchmark that also puts stress on memory then SPEC is essentially GeekBench but with much larger data sets which puts a lot of stress on the memory.

 

 

As it turns out, Apple's A14 (and thus the M1) have no problems with memory either and the results from Geekbench does not seem to show any favoritism for Apple's ARM cores.

Geekbench can show skewed results because a powerful CPU with terrible memory can still get roughly the same scores as a powerful CPU with really fast memory access. Apple's chips are in the "REALLY powerful CPU with great memory" category though.

 

 

What I find really interesting is that Apple seem to be able to achieve these fantastic results while using ARMv8 instead of ARMv9 with SVE(2) support. That means that for the next generation of chips they can easily get a MASSIVE performance boost by just including support for SVE2.

[UditDey]

3 hours ago, LAwLz said:

The reason why GeekBench "promote ARM processors a lot" is because the tests are very, VERY CPU intense but barely puts any stress on the memory.

Basically, the entire test can be loaded into the cache and then run from there, while more realistic workloads requires a lot of access to main memory, where traditionally x86 have had an advantage compared to ARM devices.

 

If you want a CPU benchmark that also puts stress on memory then SPEC is essentially GeekBench but with much larger data sets which puts a lot of stress on the memory.

 

 

As it turns out, Apple's A14 (and thus the M1) have no problems with memory either and the results from Geekbench does not seem to show any favoritism for Apple's ARM cores.

Geekbench can show skewed results because a powerful CPU with terrible memory can still get roughly the same scores as a powerful CPU with really fast memory access. Apple's chips are in the "REALLY powerful CPU with great memory" category though.

 

 

What I find really interesting is that Apple seem to be able to achieve these fantastic results while using ARMv8 instead of ARMv9 with SVE(2) support. That means that for the next generation of chips they can easily get a MASSIVE performance boost by just including support for SVE2.

How does x86 have an advantage compared to ARM for memory access speeds? Never knew there was something inherent to the instruction set that could make it better for memory

[LAwLz]

25 minutes ago, UditDey said:

How does x86 have an advantage compared to ARM for memory access speeds? Never knew there was something inherent to the instruction set that could make it better for memory

Well the ways they read and write memory are different (since they are different instruction sets) and they have different memory models but I don't know enough about that to answer your question.

I am not sure how it translates to real world performance though. From what I remember reading ~10 years ago, once you start increasing reliance on memory the gap between x86 and ARM in terms of performance has increased. That might just have been because ARM chips traditionally didn't have good and high speed memory though. It wasn't until ~2011 that we started seeing dual channel memory support in ARM chips for example, and even then it was like 600MHz DDR2 at the time 1600MHz DDR3 was standard for x86 processors.

 

ARM chips have caught up with x86 in terms of memory support so the whole "x86 has an advantage when it comes to memory performance" might be completely untrue these days. I don't know. All I know is that at one point, when Torvalds was mad at Geekbench for "favoring ARM", x86 had way faster memory than what you could find in ARM devices.

[leadeater]

8 hours ago, NotTheFirstDaniel said:

Also, if an x86 instruction requires 3 ARM equivalent instructions, is an ARM CPU with 3.1x IPC than the x86 CPU not definitively faster? Food for thought.

That's not really how it works, for example GB5 single thread benchmark takes all that in to account as it is. Secondly it can require more instructions to do the same thing as on x86 so it might take 3 instructions to do the same thing as x86 can do in 1. Comparing IPC across ISA's is fraught with problems, I wouldn't bother. Just have a look at data from benchmarks as that'll tell you what you need to know and you won't have to worry about how many instructions are required for each and how those differ.

 

As for the M1 in the MBA, MBP and Mac Mini I don't think Apple is power capping peak performance in any of the 3 devices, more just that they are controlling the sustained power in the MBA and adjusting with current thermal temperatures. For lighter or shorter workloads this should mean all 3 perform very similarly which is great, makes the MBA way way more appealing compared to the previous generations.

 

So really  the way I see it the best thing about this is you are not performance constraining yourself based of the desired platform you choose. If you want ultra portable you are still retaining largely equivalent performance to that of larger products. You can't really get that on the PC side, except maybe Ryzen Mobile U series which are also exceptionally good performance wise and power (probably why Apple choose not to use that for their comparisons, other than no product of theirs ever using them too). Apple M1 is faster, 4800U maybe not but there isn't enough data to compare these, but the 4800U would only win out on well threaded applications and that only if it actually does.

 

Spoiler

4800U is nearly the same performance as the 3600 for example. Have a look at this if you are interested in the AMD Ryzen U lower power CPUs (15W, configurable to 25W).
https://www.techspot.com/review/2121-amd-ryzen-4700u-4500u/

 

[UditDey]

4 minutes ago, LAwLz said:

Well the ways they read and write memory are different (since they are different instruction sets) and they have different memory models but I don't know enough about that to answer your question.

I am not sure how it translates to real world performance though. From what I remember reading ~10 years ago, once you start increasing reliance on memory the gap between x86 and ARM in terms of performance has increased. That might just have been because ARM chips traditionally didn't have good and high speed memory though. It wasn't until ~2011 that we started seeing dual channel memory support in ARM chips for example, and even then it was like 600MHz DDR2 at the time 1600MHz DDR3 was standard for x86 processors.

 

ARM chips have caught up with x86 in terms of memory support so the whole "x86 has an advantage when it comes to memory performance" might be completely untrue these days. I don't know. All I know is that at one point, when Torvalds was mad at Geekbench for "favoring ARM", x86 had way faster memory than what you could find in ARM devices.

I know x86 decently well and yeah maybe their different addressing modes could affect memory speeds (don't quote me on that), but I'm pretty sure at least nowadays there aren't any major differences with memory speeds. Whatever Torvalds says should be taken with a grain of salt anyways. And if geekbench can run off the cache as you suggested earlier, it should be able to run off the cache on both ARM and x86. The M1 vs 5950x scores seem to just be genuinely good engineering from apple here. I think its x86's instruction decode overhead and lack of a big.LITTLE-esque system thats holding current x86 cpus back.

[leadeater]

8 hours ago, Rooked said:

Lower CPU count makes it easier to get a better single core. Ryzen 9 5950x has double the multicore score. But I think it's still pretty cool, sorry for bothering.

It's actually a little bit more than double, most results are in the mid 17K for that CPU.