If I am reading the GeekBench numbers right, the M1 MacBook Pro is faster than both the Mac Pro and iMac Pro on single core tasks! You mean to tell me that the M1 MacBook Pro is faster than my $7500 Mac Pro? 😲
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Silicon M1 Emulating x86 is Still Faster Than Every Other Mac in Single Core Benchmark
- Thread starter MacRumors
- Start date
- Sort by reaction score
If I am reading the GeekBench numbers right, the M1 MacBook Pro is faster than both the Mac Pro and iMac Pro on single core tasks! You mean to tell me that the M1 MacBook Pro is faster than my $7500 Mac Pro? 😲
Yet, it is absolutely a very hard problem, but no.
Modern CPUs devote a lot of engineering and silicon to extract as much parallelism as possible from single-threaded code. Given its performance, the M1 probably devotes even more to it than other CPUs. However, something like Rosetta 2 potentially has even more information to use in making decisions about dependencies, out of order and speculative execution, etc. It would not be surprising if they are trying to extract even more parallelism at that layer.
Something interesting to me is that all the intel processors score about evenly regardless of generation or clockspeed. A 27" iMac with an i9 @ 3.6ghz performs about the same as a macbook pro with an i7 @ 2.3 ghz. There might be a software limitation with the different silicon. I'd like to see performance comparisons with Adobe products or something to really understand what's happening.
The ultrasparc V I worked on was actually sold as ultrasparc iv. Code name was millennium. And both Athlon 64 and opteron also had deep reorder buffers.
And there is nothing unusual about apple’s SoC. There are LOTS of competing x86 processors with integrated gpus, encryption units, and secure cores. Of course they are all different, but there is nothing about M1 that is “outside the mainstream” of CPU architecture which is what the claim was.
And as for your accusation that I may be faking who I am, I’ve been here for a VERY long time, and my user name corresponds to this guy, so either I’m him or this is a very long con I’m playing:
Former AMD engineer talks about Bulldozer
X-bit Labs reports an ex-AMD engineer spoke out about some of the reasons why Bulldozer disappoints: The reason why performance of the long-awaited Bulldozer was below expectations is not only because...
www.guru3d.com
Ex-AMD Engineer explains Bulldozer fiasco
Ex-AMD Engineer Explains Bulldozer Fiasco: Lack of Fine Tuning Automated Design = 20% Bigger, 20% Slower
forums.anandtech.com
A 533-MHz BiCMOS superscalar RISC microprocessor
This 533-MHz BiCMOS very large scale integration (VLSI) implementation of the PowerPC architecture contains three pipelines and a large on-chip secondary cache to achieve a peak performance of 1600 MIPS. The 15 mm/spl times/10 mm die contains 2.7 M transistors (2M CMOS and 0.7 M bipolar) and...
ieeexplore.ieee.org
etc. Etc.
Yes, in single threaded performance, that is using one core only, it beats any Intel system that uses only one core. Since it has four fast cores, you would expect it to beat any Intel system with four cores in multi-threaded performance, using all cores, and it does. It doesn't beat 6 core Intel system in multi-threaded performance (but the native benchmark does) and doesn't come close to systems with 8-28 cores (native beats older 8 core systems but not the latest ones, but it _is_ the lowest configuration with only four cores.
It will use as many threads as possible, and will run faster than an Intel quad core CPU.
But for many video applications Apple has heavily optimised libraries. So a video application using those libraries would automatically use Apple's ARM code, and not translate Apple's Intel code.
Last edited:
This is ridiculous. There is no way that Rosetta is taking a single thread of x86 code and splitting it into multiple parallel Arm threads. Stop it.
Why is it "NOT GOOD. NOT ENOUGH"? Not enough for what? 8GB is plenty for the majority of common IT tasks, and 16GB is still the maximum RAM in a lot of consumer laptops.
You can't say whether 16GB is sufficient or not unless you specify your usage. For the vast majority of people who would buy these computers, it will be enough. Maybe not for you, but you didn't say what you need to do.
What "problems and defects" are that need to be worked out? So far, we don't know of any. There may well be problems, but until they are shown to exist, they do not need to be fixed.
I have to say that I am mightily sick of the modern tendency for people to quote opinion and subjectivity as though it were fact. It's rife within society, and has been greatly amplified by Internet and social media use.
You need to prefix every single statement with "I believe", "in my opinion", or "I think", and then PUT FORWARD YOUR ARGUMENT FOR YOUR BELIEF.
Is this basic mental discipline no longer taught in schools????
I asked the guys at my company writing our server software, and at the moment i'm told everything we use except one application could be run on an ARM VM. Not that it matters, because iOS and Android developers don't run those VMs on their development machines.
Yep, i was thinking that as well. Particularly if some recalcitrant developer lags in providing a native binary, Apple may be motivated to look into how to make Rosetta handle that app better. But, of course, they don’t want to make Rosetta TOO good, because that just discourages developers from providing native code.
Rosetta2 is not emulation its more like IBM TIMI which does a one time conversion of the application binary into native ARM binary. Its an impressive capability.
Why do you buy Macs at all then? Buying Macs to run Windows (or based on spending a lot of time in Windows on it) is just getting the worst of both worlds imho.
As I see it... unless you have a very specific niche they're very appealing - they run MS office well enough, they run cloud apps as well or better than any other platform, they run native macOS applications which are generally better than Windows apps at what they do.
Windows? Personally I don't want Windows. I don't need Windows. Given I don't need/want Windows, the realistic choices right now are Linux or macOS and macOS is nicer. Couple that with 2x battery life and 3x processing speed in a portable... its a no brainer imho. Not running Windows is a feature.
Even Microsoft is supporting other platforms better than Windows these days - iPad Office 365 is actually touch friendly - runs way better than anything you can get for the MS Surface.
Hopefully these speed improvements will get more developers to jump ship. Even MS are more of an application/cloud company now - don't be surprised to see them pushing "better than PC" apps on the Mac if these chips are as good as they appear. It's just another application platform for MS now - one they don't even have to maintain.
Last edited:
It wasn’t an accusation so check your statement please. It was a statement of fact that people online may not be who they say they are. Posting a link doesn’t matter. There are people in the world who spend their lives pretending to be someone they are not. Therefore your comments fall on deaf ears regarding who you are because, as stated, there is insufficient proof other than that you may have convinced others you are who you say you are. You seem to have taken offence in this. Don’t. It’s a fact of the online world. I don’t believe what most say online unless it can be independently verified by a trusted third party, and expect others to view anything I say in the same light.The ultrasparc V I worked on was actually sold as ultrasparc iv. Code name was millennium. And both Athlon 64 and opteron also had deep reorder buffers.
And there is nothing unusual about apple’s SoC. There are LOTS of competing x86 processors with integrated gpus, encryption units, and secure cores. Of course they are all different, but there is nothing about M1 that is “outside the mainstream” of CPU architecture which is what the claim was.
And as for your accusation that I may be faking who I am, I’ve been here for a VERY long time, and my user name corresponds to this guy, so either I’m him or this is a very long con I’m playing:
Former AMD engineer talks about Bulldozer
X-bit Labs reports an ex-AMD engineer spoke out about some of the reasons why Bulldozer disappoints: The reason why performance of the long-awaited Bulldozer was below expectations is not only because...Ex-AMD Engineer explains Bulldozer fiasco
Ex-AMD Engineer Explains Bulldozer Fiasco: Lack of Fine Tuning Automated Design = 20% Bigger, 20% SlowerA 533-MHz BiCMOS superscalar RISC microprocessor
This 533-MHz BiCMOS very large scale integration (VLSI) implementation of the PowerPC architecture contains three pipelines and a large on-chip secondary cache to achieve a peak performance of 1600 MIPS. The 15 mm/spl times/10 mm die contains 2.7 M transistors (2M CMOS and 0.7 M bipolar) and...ieeexplore.ieee.org
etc. Etc.
I will note on the count of refuting arguments other than vague hand waving you haven’t really identified how other processors are all the same and following the same design path. You did state knowledge 1year ago of what Apple was going to do, but chip design is far more long term than that, so surely you knew years ago? It’s also of note that other cpu designers are stating some surprise at the design and apparent performance of the m1. This doesn’t really indicate that it’s the normal path of chip design.
The integrated GPU in the M1 is pretty impressive as well.
MacBookPro17,1 - Geekbench
Benchmark results for a MacBookPro17,1 with an Apple Silicon processor.
browser.geekbench.com
Last edited:
On one hand it’s pretty amazing how well dialed in MacOS is on AS. On the other hand it’s painfully obvious how poorly it runs on Intel. I’m sure AS brings some impressive gains, but it’s not like Intel instruction set was fully optimized on MacOS. It’s great Apple is finally spending a lot of resources on Macs again.
Intel has the anchor of backwards compatibility with code back to 1978 8086 cpu. Then decades of cruft on top of an operating system called windows which has backwards compatability with 16bit windows 3.1 code. All this adds up to massive efficiency issues.
Until Microsoft and Intel partners together to work on a new from scratch 21st hw/software architecture you will see little progress.
Nah, the haters will continue pounding on the 720p webcam and two ports.
Well, that's just ridiculous. Not exactly surprising given what the last generation of Rosetta was capable of and the general performance numbers coming out of the M1, but still ridiculous.
Nitpick with the headline "Apple Silicon M1 Emulating x86 is Still Faster Than Every Other Mac in Single Core Benchmark", though:
Rosetta is not an emulator, at least in the standard technical use of the term. This kind of performance out of an emulator would be mind-boggling. Rosetta is a dynamic binary translator, which is a big reason the performance hit is so minor, but also why it only works on certain things.
That didn't necessarily carry through the entire PPC era, and I won't argue that the PPC's best performance areas weren't necessarily consumer-oriented, nor did they carry through toward the end of the era. But I've seen nothing in the past several years to indicate that the extreme difference between A-series benchmarks and Android CPU benchmarks are inaccurate in real-world use, so I don't see why the case would be any different with M1 versus x86.
I mean, I can go to Browserbench.org and run the exact same (single-core) benchmark on the exact same browser on my phone and my 15" i9 laptop with active cooling, hear the fan ramp up on my laptop, watch both do the same thing, and watch my phone outperform my laptop by a factor of 1.5 to 2. The I9-8950HK does outperform the A14 in a couple of specific tests, but the edge goes to the A-series in most cases, sometimes by a huge margin.
More importantly, and very differently from the PPC, Apple has been designing their chips for a single, consumer-use platform for a decade now, so if anything I'd expect them to perform better in some real world use cases than more general-purpose Intel chips. The chip design literally has one customer, and that customer only makes a handful of products, most of which have a similar target market.
Nitpick with the headline "Apple Silicon M1 Emulating x86 is Still Faster Than Every Other Mac in Single Core Benchmark", though:
Rosetta is not an emulator, at least in the standard technical use of the term. This kind of performance out of an emulator would be mind-boggling. Rosetta is a dynamic binary translator, which is a big reason the performance hit is so minor, but also why it only works on certain things.
I remember the G4/G5 days quite well--before things started to stagnate later in the PPC era, I remember running a paralleled scientific data-processing application I built in a cross-platform environment on one of our x86 Windows machines and one of our shiny new PPC Macs, and the Mac completed the data analysis so much faster I at first thought I had done something wrong.
That didn't necessarily carry through the entire PPC era, and I won't argue that the PPC's best performance areas weren't necessarily consumer-oriented, nor did they carry through toward the end of the era. But I've seen nothing in the past several years to indicate that the extreme difference between A-series benchmarks and Android CPU benchmarks are inaccurate in real-world use, so I don't see why the case would be any different with M1 versus x86.
I mean, I can go to Browserbench.org and run the exact same (single-core) benchmark on the exact same browser on my phone and my 15" i9 laptop with active cooling, hear the fan ramp up on my laptop, watch both do the same thing, and watch my phone outperform my laptop by a factor of 1.5 to 2. The I9-8950HK does outperform the A14 in a couple of specific tests, but the edge goes to the A-series in most cases, sometimes by a huge margin.
More importantly, and very differently from the PPC, Apple has been designing their chips for a single, consumer-use platform for a decade now, so if anything I'd expect them to perform better in some real world use cases than more general-purpose Intel chips. The chip design literally has one customer, and that customer only makes a handful of products, most of which have a similar target market.
I'm still running a mid-2013 MacBook Air and am keen to update but I've held off knowing that the transition to Apple silicon was coming. That appears to have been a wise choice as these numbers are looking great, and that's coupled with the improved battery life. However, as great as these numbers are, I'm going to wait until the rumoured 14" MacBook Pro drops. I think Apple has just used the 2020 Air and 13" Pro, in which the Apple SoC is the only meaningful change, as a platform to demonstrate the performance gains as a result of the M1 versus the same setup on Intel. As such, I'll wait for the more exciting refresh, which will hopefully drop before mid-2021. After all, I've already persevered with my 2013 MBA for a long time, so what's another 6 months or so wait for a device that expect to hold onto for another 6 - 7 years.
Yeah, and it’s tough to drop any sort of backward compatibility if you are the majority. I mean we can read how many people still unwilling to upgrade from Windows 7. Even now intel is forcing some people to not use an old OS by requiring Windows 10 on some newer CPUs.
But intel’s entrance into discrete GOU might at least give a glimpse that they finally care about GPU. Bit late of course.
You are giving me a headache. With out of order execution, everything runs on _one_ core, and that core guarantees that the results are the same. The hardware can pick the next instructions that can be executed within 330 or so picosends out of a few hundred instructions that are waiting. With multiple threads, that's absolutely impossible.
swingerofbirch
macrumors 68040
The state of Apple for me: Good chips, otherwise middling hardware, software is now consistently bad.
I can't imagine in the year 2000 with the dawn of OS X and all the promise it held that I would today be considering a slower Windows computer because Apple's software had become so unreliable for me.
I can't imagine in the year 2000 with the dawn of OS X and all the promise it held that I would today be considering a slower Windows computer because Apple's software had become so unreliable for me.
Just like the 16GB limit on the Macs these replaced.
The M1 is optimized for one thing: running apps. X86 processors are optimized for many, many different things, some of which conflict with each other.
The larger instruction caches on the m1 suggest to me that Apple discovered that speculative execution wins and code cache misses are more problematic than memory misses. That's something you can only figure out with profiling. In this multiple task/process world that might be more true than it used to be.
The larger instruction caches on the m1 suggest to me that Apple discovered that speculative execution wins and code cache misses are more problematic than memory misses. That's something you can only figure out with profiling. In this multiple task/process world that might be more true than it used to be.