The M1 Benchmarks Continue - Emulated performance appears to *still* outperform any intel-based Mac

[Spindel]

5 minutes ago, KuroNanashi said:

M1 MacBook Air CB23 - Multi 7179, Single 1476

Wait a minute, that's not far off the result from the MBP with active cooling. 

 

CB23 does not stress the CPU enough, so it's a crap benchmark for sustained loads. 

 

Let's wait for more real benchmarks. 

[leadeater]

8 minutes ago, LAwLz said:

In a best case scenario for the 4700U, that 6874 score is at the 10 watt limit. In a worst case scenario, it's at the 25 watt limit. We don't know that yet but even in the base case scenario the M1 is between 10% to 26% faster while using the same or maybe less than half the power.

 

 

It is mind blowing that people aren't impressed by this.

No I like the M1, I'm just not going to praise it to the heavens especially lacking important information on it from proper testing.

 

Also you can take that 10% to 26% performance lead, take off a majority of it when the mobile parts release using Zen 3.

 

Far as single core goes, obviously a complex single core heavy design is going to score higher, why am I supposed to be so surprised at a specific design intent utilizing a superior process node?

 

P.S. the 4800U also exists.

[Spindel]

2 minutes ago, KuroNanashi said:

Absolutely, I could start a thread. Any tests in particular you'd like? I re-ran CB23 again after things have settled down, I think I had iCloud stuff syncing and scored 7573 points this time on multi.

OK that's the same as the actively cooled MBP. 

 

CB23 a crap benchmark of performance confirmed!

[RedRound2]

11 minutes ago, KuroNanashi said:

Absolutely, I could start a thread. Any tests in particular you'd like? I re-ran CB23 again after things have settled down, I think I had iCloud stuff syncing and scored 7573 points this time on multi.

@LAwLz?

I guess GPU is what is still a question mark, aprt from one article that claims it has 1050 level performance.

Also, you can check rendering times with iMovie and monitor temperature while you're at it

9 minutes ago, Spindel said:

Wait a minute, that's not far off the result from the MBP with active cooling. 

 

CB23 does not stress the CPU enough, so it's a crap benchmark for sustained loads. 

 

Let's wait for more real benchmarks. 

He has the new MacBook Air. We can ask him to run more realistic loads

[Spindel]

1 minute ago, RedRound2 said:

@LAwLz?

I guess GPU is what is still a question mark, aprt from one article that claims it has 1050 level performance.

Also, you can check rendering times with iMovie

He has the new MacBook Air. We can ask him to run more realestic loads

Still stands CB23 sucks as a benchmark if it can't make an MBA throttle  

[exhaustedbrit]

34 minutes ago, RedRound2 said:

@LAwLz?

I guess GPU is what is still a question mark, aprt from one article that claims it has 1050 level performance.

Also, you can check rendering times with iMovie

He has the new MacBook Air. We can ask him to run more realestic loads

For reference I have access to the following machines for comparisons:

• 2018 MacBook Pro 13" (Core i5-8269U, 8GB, 256GB, 4-port model with 2 fans)
• 2019 MacBook Pro 15" (Core i9-9880H, 32GB, 1TB, Vega 20)
• 2020 MacBook Air 13" (Core M1 8-core GPU, 8GB, 512GB)

I also use two Razer Core X Chroma enclosures with Vega 56s which sadly does not work with the M1 macbook.
 

If you guys have benchmarks you care about just let me know and I'll include them in a thread later and can run them whilst I'm working. In a conference at the moment so just waiting for Shadow of the Tomb Raider to finish downloading.

 

The new thread is over here!

Edited November 17, 2020 by KuroNanashi
Added link to aforementioned thread.

[CarlBar]

5 minutes ago, KuroNanashi said:

This is just me hypothesising, but I actually have a suspicion that the bulk of the thermal budget on the M1 chip is not reserved for the CPU. I've run stuff pretty intensely on my M1 MacBook Air and it's barely getting warm, I haven't launched any GPU intensive tasks yet, I'm installing Shadow of the Tomb Raider and some other games. But given there's a supposed 2.6 TFLOP measure of GPU silicon in there I'd expect that's where this magical 10W envelope actually gets hit.

I had one of the Iris Plus MacBook Airs when they were released earlier this year (and swiftly returned it) and remember how quickly the temperatures shot up and barely sustained 3GHz without any sort of GPU load, when these equivalent mobile chips from Intel or Ryzen are performing in tasks which also use the GPU they have to fight for thermal budget. What that also meant that actually playing games was a toss up between whether I wanted that dual core 3GHz or whether I wanted some semblance of GPU performance in that MacBook Air and the result was an inconsistent game with framerates all over the place. My guess is that we aren't even hitting that 10W power budget with a multicore CB23 run because it only stresses the CPU, Apple releases products not hardware (for now anyway) and that's why these comparisons are difficult.

I'm not suggesting it's meaningless to compare the CB23 results with Tiger Lake and Ryzen mobile chips - but I do think the way in which power budget and thermals are approached with this Apple chip is not directly comparable, and a focus on balancing the overall constituent functions (CPU, GPU, ISP etc) for consistent performance has been the goal here.

Best to wait for the reviews - but for now here are my numbers.

M1 MacBook Air CB23 - Multi 7179, Single 1476

 

Unless Apple screwed up big time the fact that cinebench isn't loading the GPU especially much should mean virtually all of the 10w budget is going to the CPU.

 

19 minutes ago, leadeater said:

Careful with that though, Ryzen Mobile U is 15W, with TDP up 25W mode and TDP down 10W mode. I suspect the 10K points is on a laptop set to 25W.

 

Fair, i'd got it into my head for some reason that the 25w limit part was a seperate skew.  As is probably obvious i only pay very loose attention to the laptop side of things. I know enough not to be clueless but i'm not exactly boned up.

 

1 minute ago, leadeater said:

Far as single core goes, obviously a complex single core heavy design is going to score higher, why am I supposed to be so surprised at a specific design intent utilizing a superior process node?

 

This @LAwLz  The M1 is has a whole bunch of features and a process node advantage behind it aimed at getting the power to performance tuned really good. It performing slightly better per watt than somthing a generation behind it architecturally is the minimum bar it needs to meet. Anything worse and it's a failure.

 

1 minute ago, Spindel said:

Still stands CB23 sucks as a benchmark if it can't make an MBA throttle  

 

Or a 10W power budget just doesn't need much cooling. Hard to say without a long list of benchmarks run on both. Not that i'd use R23 as the be all and end all, but it was all we had a few pages back that was a third party run.

 

10 minutes ago, KuroNanashi said:

Absolutely, I could start a thread. Any tests in particular you'd like? I re-ran CB23 again after things have settled down, I think I had iCloud stuff syncing and scored 7573 points this time on multi.

 

I'd like to see Shadow of the Tomb raider when it's installed. 3D Mark if you own it too. Any other games with a benchmark would be cool too. Not sure what would make a good production style benchmark that we have comparable data for.

[CarlBar]

3 minutes ago, KuroNanashi said:

For reference I have access to the following machines for comparisons:

• 2018 MacBook Pro 13" (Core i5-8269U, 8GB, 256GB, 4-port model with 2 fans)
• 2019 MacBook Pro 15" (Core i9-9880H, 32GB, 1TB, Vega 20)
• 2020 MacBook Air 13" (Core M1 8-core GPU, 8GB, 512GB)

I also use two Razer Core X Chroma enclosures with Vega 56s which sadly does not work with the M1 macbook.
 

If you guys have benchmarks you care about just let me know and I'll include them in a thread later and can run them whilst I'm working. In a conference at the moment so just waiting for Shadow of the Tomb Raider to finish downloading.

 

Data, sweet, juicy, adorably fluffy, data. 

[A51UK]

44 minutes ago, RedRound2 said:

Overpriced? The iPhones, iPads, entry Macs, AW, homePod have very well been documented on how it's actually reasonably priced, especailly for the unparalleled support, ecosystem and in house OS and software they provide. Acessories and upgrades are what I would categorize as overpriced and that definitely exists to position Apple as premium brand (even though iPhones are one fo the most common devices found)

 

Right to repair - I agree

 

Customer right problem? Same point as above?

 

You seem to have a wrong understanding of T1 chip, if you actually think it's a problem and we've been through this in the forum

 

No care for developers? Where does this stem from? Need more sources for this. The 30% thing is an industry standard, but of course only Apple exists in negative light

 

Keep removing ports is a tradeoff they take to move forward with industry. It's becasue of Apple we have cheap and decent bluetooth headsets today, and that USB C is more common in newer devices today

The point being? Is there a law that states that if AMD can't figure it out, Apple also can't

Windows on ARM is a mess for multitude of reasons. None of the issues which Apple have

Good for you? Again point?

 

Yes not all thing are over price (e.g. IPod) but all mac and IPhone are overprice 

 

What wrong with my point on T2 chip?  It over kill security and stop any form of data recovery if something go wrong with it? 99% people do not need this type of security.  We should be able to choice if what the data or SSD to be encrypted or not. It security vs data protection.  This one problem for Customer right.  Too much security is a bad thing. You need a balance between useable and security, too much security less useable it become. 

 

Try to stop customer from upgrade, stop people from using a different market place for app, trying to stop other OS from loading (e.g. IPhone, m1 chips Mac). All are Customer right. problems. 

 

You have to use Apple tool and pay apple to build apps and many other problem when developers app for apple. I agree that 30% is ok and fair but should allow for other market place as well. 

 

Apple not been forward with industry in port at all, remove thing that useful is not move forward just to sell more new things. 

 

The point of AMD  and Windows on ARM is to show that we not bias on x84-64, 

 

I do think you are vary bias on apple, try to look out side of the box and say we need to wait for the reviews. 

 

 

 

 

[saltycaramel]

Real wold usage is starting to emerge, the end of the road for fine benchmark connoisseurs.

In the following example, the low end M1 (soon to be beaten next Easter by his big brother M1X with double the perfomance cores) compiles in 10% less time than a 10-core 120W Xeon W (Skylake).

 

 

 

[Spindel]

22 minutes ago, CarlBar said:

Or a 10W power budget just doesn't need much cooling. Hard to say without a long list of benchmarks run on both. Not that i'd use R23 as the be all and end all, but it was all we had a few pages back that was a third party run.

So let me get this straight.

 

Geekbench, that tries several real world tasks is bad because it's bursty and which means most CPUs run it in full turbo mode. 

While Cinebench that does exactly one task is good because it show sustained loads but it can't even get a passively cooled MBA to throttle.

 

Also apparently Affinity benchmark wasn't representative of real world perofrmance. And by the lack of responses I guess the FCPX comparisons aren't a good real world test. 

 

Check!

[LAwLz]

8 minutes ago, Spindel said:

So let me get this straight.

 

Geekbench, that tries several real world tasks is bad because it's bursty and which means most CPUs run it in full turbo mode. 

While Cinebench that does exactly one task is good because it show sustained loads but it can't even get a passively cooled MBA to throttle.

 

Also apparently Affinity benchmark wasn't representative of real world perofrmance. And by the lack of responses I guess the FCPX comparisons aren't a good real world test. 

 

Check!

You just can't win. No matter what evidence gets thrown at people they brush it off and come up with a reason for why isn't valid.

I had someone in another thread say you can't compare the Cinebench scores because one was running MacOS and the other one was running Windows. That was of course a comment made after that same person had said "let's wait for Cinebench" when the Geekbench and SPEC results were posted...

 

It's like the Olympic games of mental gymnastics right now.

[Spindel]

2 minutes ago, LAwLz said:

You just can't win. No matter what evidence gets thrown at people they brush it off and come up with a reason for why isn't valid.

I had someone in another thread say you can't compare the Cinebench scores because one was running MacOS and the other one was running Windows. That was of course a comment made after that same person had said "let's wait for Cinebench" when the Geekbench and SPEC results were posted...

 

It's like the Olympic games of mental gymnastics right now.

Lite roligt är det

 

Translation: It's fun  

[RedRound2]

1 hour ago, A51UK said:

Yes not all thing are over price (e.g. IPod) but all mac and IPhone are overprice 

And how exactly are Macs and iPhones overpriced? iPhones go all the way from $399 SE with flagship chip to $729 and $829 which is what most people should buy. The Pro offers extremely few things that only 10 people care about, while in Androidland, phones with comparable specs come over a thousand dollars - with the ones equivalient to the 12 series have usually having one or the other compromises.

 

Also, who pays for software? The price of iPhones also factor in iOS development costs and the support it comes with. On android, the OS is essentially free and there is basically no support, maybe 1-2 updates if you're lucky

 

Comparable laptops and Macs are usually on par with each other. Even on desktops too, except the Mac Pro (which is irrelevant). Even Linus has admitted this.

1 hour ago, A51UK said:

What wrong with my point on T2 chip?  It over kill security and stop any form of data recovery if something go wrong with it? 99% people do not need this type of security.  We should be able to choice if what the data or SSD to be encrypted or not. It security vs data protection.  This one problem for Customer right.  Too much security is a bad thing. You need a balance between useable and security, too much security less useable it become. 

Try to stop customer from upgrade, stop people from using a different market place for app, trying to stop other OS from loading (e.g. IPhone, m1 chips Mac). All are Customer right. problems. 

You have to use Apple tool and pay apple to build apps and many other problem when developers app for apple. I agree that 30% is ok and fair but should allow for other market place as well. 

Apple not been forward with industry in port at all, remove thing that useful is not move forward just to sell more new things. 

I genuinly dont have the time to get into this rn,I've discussed about most of this long time ago in so many threads and I really dont want to derail this topic. We've covered T2 in other threads (Ill post a link i find it). In short, t@ does a lot more than you think - it offloads various processes from CPU. Why do customers want to load other OSs onto their iPhones? Parallels still work on ARM macs. Xcode is a free to use software. What is many other problems? Developing on iOS and macOS is far easeir than on Android or Windows. That's why devs have a much easier time porting their apps to ARM. You pay $99 a year for Apple developer account (really nobody complains about this). What does Apple sell more becasue they removed a port? Please tell me. They released AirPods, but people would've bough it regardless the iPhone having headphone jack or not. Nobody was forced to buy anything here. 

 

1 hour ago, A51UK said:

The point of AMD  and Windows on ARM is to show that we not bias on x84-64, 

See, if this was Microsoft or Qualcomm's effort, EVERYBODY here would be celebrating and talking about how CPUs havebecome exciting again. But it's Apple, so all benchmarks and proof of their in house chip is irrelvant and useless

1 hour ago, A51UK said:

I do think you are vary bias on apple, try to look out side of the box and say we need to wait for the reviews. 

I did not say anything about ther performance of M1 chip other than react to other's peoples comments here. We have preliminary results on the database, and people going out of their way to justify why it doesnt matter is the issue here. Because if it was any other company, we all would be celebrating, but most people are still in denial here. Give me a break

[NotTheFirstDaniel]

The review embargo just lifted for all M1 Macs.

[Blade of Grass]

23 hours ago, Taf the Ghost said:

The Apple ecosystem will stretch from mobile into the two places in Desktop that mattered for 20 years to Apple: Compute Illiterates and Graphic Designers.

22 hours ago, Blademaster91 said:

While saying macs are for computer illiterates and graphic designers is a generalization,  I think it isn't far too far from being true, the enthusiasts that use macs for Unix compatibility are definitely not the majority of users.

Y'all really have never graced the halls of Silicon Valley or the computer science department at a university eh? 

[CarlBar]

7 hours ago, Spindel said:

So let me get this straight.

 

Geekbench, that tries several real world tasks is bad because it's bursty and which means most CPUs run it in full turbo mode. 

While Cinebench that does exactly one task is good because it show sustained loads but it can't even get a passively cooled MBA to throttle.

 

Also apparently Affinity benchmark wasn't representative of real world perofrmance. And by the lack of responses I guess the FCPX comparisons aren't a good real world test. 

 

Check!

 

I didn't say any of that. personally since it's not heavily used or emphasised by any of the tech press that i follow i tend not to give geekbench much credence, if it was a great benchmark i'd expect them to use it. Also the geekbench numbers where a result of a special early review sample to a single reviewer, that makes them suspect just as if apple themselves had run them.

 

It's only when you get units in the hands of people with no strings attached from apple that the benchmarks numbers become trustworthy. And even then you should be looking long term at multiple different benchmarks. We didn't have that though a few pages back.

[LAwLz]

3 hours ago, CarlBar said:

I didn't say any of that. personally since it's not heavily used or emphasised by any of the tech press that i follow i tend not to give geekbench much credence, if it was a great benchmark i'd expect them to use it. Also the geekbench numbers where a result of a special early review sample to a single reviewer, that makes them suspect just as if apple themselves had run them.

 

It's only when you get units in the hands of people with no strings attached from apple that the benchmarks numbers become trustworthy. And even then you should be looking long term at multiple different benchmarks. We didn't have that though a few pages back.

Are you seriously trying to say that benchmarks posted now can't be trusted because we need long term reviews from people who have bought the products themselves rather than trusted reviewers like Anandtech? 

 

The goalpost keep moving... 

[CarlBar]

1 hour ago, LAwLz said:

Are you seriously trying to say that benchmarks posted now can't be trusted because we need long term reviews from people who have bought the products themselves rather than trusted reviewers like Anandtech? 

 

The goalpost keep moving... 

 

I'm saying receiving special early access isn't the same as a regular review sample. In exactly the same way as when the RTX 3k series launched and one tech news site got an early sample everyone said to wait till they where confirmed by people with just regular do whatever you please review samples.

[Belgarathian]

12 hours ago, CarlBar said:

 

Unless Apple screwed up big time the fact that cinebench isn't loading the GPU especially much should mean virtually all of the 10w budget is going to the CPU.

 

 

Fair, i'd got it into my head for some reason that the 25w limit part was a seperate skew.  As is probably obvious i only pay very loose attention to the laptop side of things. I know enough not to be clueless but i'm not exactly boned up.

 

 

This @LAwLz  The M1 is has a whole bunch of features and a process node advantage behind it aimed at getting the power to performance tuned really good. It performing slightly better per watt than somthing a generation behind it architecturally is the minimum bar it needs to meet. Anything worse and it's a failure.

 

 

Or a 10W power budget just doesn't need much cooling. Hard to say without a long list of benchmarks run on both. Not that i'd use R23 as the be all and end all, but it was all we had a few pages back that was a third party run.

 

 

I'd like to see Shadow of the Tomb raider when it's installed. 3D Mark if you own it too. Any other games with a benchmark would be cool too. Not sure what would make a good production style benchmark that we have comparable data for.

The Verge said that the MBA didn't throttle until the 9min mark in their testing, so you'd need to run CB23 for multiple passes and take the average.

 

As a side note, they also mentioned in their review that Spotlight is noticeably better and there is now a dedicated dictation button in the function row which is very good (might be taking advantage of the built-in neural engine?).

[Bombastinator]

The test I would like to see is a wide selection of games at 1080p low or medium, and see what kind of fps is gotten. There’s probably several ways to install such things which could make a difference. 
 

Another would be the various major productivity apps that people buy macs for.  
 

The question is what do these things compare to?  The impression I get is that they are very likely to beat the pants off other thin-and-lights, but do they beat anything bigger? Also with the games 1080p medium/low is like 90% of making a game go.  If you get a usable frame rate at that it might not be the highest fidelity gaming experience, but it would be a functional one.