M1 Macs Reviewed

[LAwLz]

59 minutes ago, Generalcrow said:

http://support.primatelabs.com/kb/geekbench/interpreting-geekbench-5-scores Oddly my data is not up to date on geek bench 5 but how you scale an i5 to an arm processor is a very hard question.  To your other analogy it is like trying to compare a truck to a bicycle as a form of moving building materials, you can do it but is it the best use of the gear.

That does not say Geekbench uses a different baseline for each architecture, which is how I interpreted your post.

 

And how to scale an i5 to an ARM processor is not hard. The tests Geekbench does are things like "how long did it take to compile this piece of code" or "how long did it take to compress this file using 7-zip".

Like I said, just because one car as a diesel engine and another car has a petrol engine does not mean you can't compare their lap times. Geekbench measures lap times. How long it takes to finish a certain set of tasks.

 

By your logic, we can't compare AMD to Intel either because both companies use different internal instruction sets. It doesn't matter how a processor accomplishes something. What matters is how long it takes for a processor to complete a certain task, and that's what Geekbench measures.

[Spindel]

Just got my M1 Mini and I’m currently restoring my TM backup to it. 
 

 

I won’t do any benchmarks but I’ll report my experiences in my general usage later on. 

[Generalcrow]

2 hours ago, LAwLz said:

 

By your logic, we can't compare AMD to Intel either because both companies use different internal instruction sets. It doesn't matter how a processor accomplishes something. What matters is how long it takes for a processor to complete a certain task, and that's what Geekbench measures.

Citation needed. Intel and AMD do use the same instruction set, you might be thinking of IA-x64 which was Intel's failed x64 instruction set (they use AMDs now), it was not x86 compatible.

[Generalcrow]

Geekbench relies on a score weighting given to jobs that do not run the same on each platform.  The real way to test is to get a real world test (Blender, photoshop, Lightroom and Mathlab) push a 2 hour job through each and see which is faster where.  I bet the M1 is good but I bet is no where near the weighting the Geekbench score have given it.  Or are you in the camp of people that think this is faster than a PS5/XboxX because that is what Geekbench is saying and that makes no sense.

[Bombastinator]

4 minutes ago, Generalcrow said:

Geekbench relies on a score weighting given to jobs that do not run the same on each platform.  The real way to test is to get a real world test (Blender, photoshop, Lightroom and Mathlab) push a 2 hour job through each and see which is faster where.  I bet the M1 is good but I bet is no where near the weighting the Geekbench score have given it.  Or are you in the camp of people that think this is faster than a PS5/XboxX because that is what Geekbench is saying and that makes no sense.

It might. Ps5/XboxX use what I have been given to understand to be sort of a hybrid between ryzen+ and ryzen2 for cpu, and navi10 and Navi 21 for gpu and both were heavily optimized for running games.  They could be very bad at non games stuff perhaps.  Only perhaps. 
 

In any case I think it likely that over the next few months various head to head app tests will be conducted in various places and will get collected together. Synthetic benchmark apps are famous for their limited accuracy in general.  The truth will out.  It will support or refute the geek bench numbers. 

[LAwLz]

21 minutes ago, Generalcrow said:

Citation needed. Intel and AMD do use the same instruction set, you might be thinking of IA-x64 which was Intel's failed x64 instruction set (they use AMDs now), it was not x86 compatible.

I think you missed the word "internal".

Neither AMD nor Intel actually execute x86 instructions anymore (in a lot of cases at least). They accept x86 instructions, but translates them to their own proprietary instructions.

That's what I referred to when I said "internal instruction set".

 

33 minutes ago, Generalcrow said:

Geekbench relies on a score weighting given to jobs that do not run the same on each platform.

What exactly do you mean by "do not run the same on each platform"?

 

 

35 minutes ago, Generalcrow said:

The real way to test is to get a real world test (Blender, photoshop, Lightroom and Mathlab) push a 2 hour job through each and see which is faster where. 

According to whom? You?

I don't think most reviewers run a 2 hour test on each program. Would you say the bencharmsk from Gamers Nexus and Anandtech aren't valid either because they are too short?

 

36 minutes ago, Generalcrow said:

I bet the M1 is good but I bet is no where near the weighting the Geekbench score have given it.

You do know we have other tests as well, such as Cinebench, and it gives similar scores?

 

36 minutes ago, Generalcrow said:

Or are you in the camp of people that think this is faster than a PS5/XboxX because that is what Geekbench is saying and that makes no sense.

I don't think it is faster than the PS5 and Xbox series X.

The GPU isn't faster. The CPU probably is on a per core basis, but since both consoles has twice the number of cores they should come out on top in the end anyway.

 

Can you show me Geekbench saying the M1 is faster than the PS5 and Xbox series X?

[Spindel]

This thing is incredibly fast.

It's hard to describe. 

[Generalcrow]

15 hours ago, LAwLz said:

I think you missed the word "internal".

Neither AMD nor Intel actually execute x86 instructions anymore (in a lot of cases at least). They accept x86 instructions, but translates them to their own proprietary instructions.

That's what I referred to when I said "internal instruction set".

 

What exactly do you mean by "do not run the same on each platform"?

 

 

According to whom? You?

I don't think most reviewers run a 2 hour test on each program. Would you say the bencharmsk from Gamers Nexus and Anandtech aren't valid either because they are too short?

 

You do know we have other tests as well, such as Cinebench, and it gives similar scores?

 

I don't think it is faster than the PS5 and Xbox series X.

The GPU isn't faster. The CPU probably is on a per core basis, but since both consoles has twice the number of cores they should come out on top in the end anyway.

 

Can you show me Geekbench saying the M1 is faster than the PS5 and Xbox series X?

Whats an internal instruction set, you mean the physical layout of the logic gates, don't invent new terms just google the right one?  I am confused by your confusion.

 

"do not run the same on each platform" is English it means what it says, the jobs for Arm chips are smaller on Geekbench and compiled for Arm the ones on X86 are compiled for AMD64 and run on that platform.  The tests do not run the same.

 

This is how you benchmark IRL not in cosy app you downloaded from appstore, this is exactly (using different applications) how Gamers Nexus and LTT will test this chip.

 

Cinebench rates this chip as a i5 8400 NOT a 5950X single core speed (ignoring graphics and AI) remember this only has 4 performance core and 4 low power cores. For it to be given a 1700 rating is dumb dumb dumb dumb (south park reference).

 

https://www.extremetech.com/computing/317304-benchmark-results-show-apple-m1-beating-every-intel-powered-macbook-pro

https://browser.geekbench.com/processors/amd-ryzen-9-5950x

 

Look I am not replying to your trolling anymore.  

 

[LAwLz]

53 minutes ago, Generalcrow said:

Whats an internal instruction set, you mean the physical layout of the logic gates, don't invent new terms just google the right one?  I am confused by your confusion.

I am not confused.

In a lot of cases, modern AMD and Intel CPUs do not directly execute x86 instructions. They fetch x86 instructions and then they transform those instructions into their own proprietary ISA which gets executed on the CPU. If you want the correct name for it it's microcode. I thought I'd make it a bit simpler for you by calling it "internal ISA".

 

In any case, saying that you can't compare an ARM CPU to an x86 because they do not use the same instruction set is like saying you can't compare an AMD CPU to an Intel CPU because they do not use the same microcode set.

Of course you can compare them, because the instruction set (or microcode) is just the way the different processors handles the tasks, but that's not what we are interested in. What we are interested in is how quickly they complete the tasks. That can be measured regardless of how different their approaches are to processing the data.

 

 

53 minutes ago, Generalcrow said:

"do not run the same on each platform" is English it means what it says, the jobs for Arm chips are smaller on Geekbench and compiled for Arm the ones on X86 are compiled for AMD64 and run on that platform.  The tests do not run the same.

[Citation Needed]

According to the whitepaper for Geekbench the "jobs" are exactly the same. Regardless of which ISA you are testing Geekbench on, the tests are the same.

The processor is handed the same task, with the same data, and are expected to arrive at the same conclusion, and then the time it took is measured. For example in the Clang compile test both an x86 and ARM processor are both going to compile the same source code, targeting the same architecture, using the same compiler.

 

53 minutes ago, Generalcrow said:

This is how you benchmark IRL not in cosy app you downloaded from appstore, this is exactly (using different applications) how Gamers Nexus and LTT will test this chip.

What are you talking about. What does it being from an appstore have to do with anything? You can download Cinebench from the app store too, does that mean it isn't a real benchmark?

Also, Geekbench isn't really a single benchmark. It is a benchmark suite. Geekbench is a program that contains multiple programs such as the LZMA SDK, libpng, libjpeg-turbo, Gumbo HTML5, the Duktape JS engine, SQLite, PDFium, FreeType, Clang and many more.

 

A benchmark isn't better just because you download it from a website rather than an app store. Nor is a benchmark necessarily better just because it takes longer to run.

 

 

53 minutes ago, Generalcrow said:

Cinebench rates this chip as a i5 8400 NOT a 5950X single core speed (ignoring graphics and AI) remember this only has 4 performance core and 4 low power cores. For it to be given a 1700 rating is dumb dumb dumb dumb (south park reference).

Numbers from Anandtech:

Cinebench R23 gives the 5950X a single core score of 1647.

Cinebench R23 gives the M1 a single core score of 1522.

 

 

Geekbench 5 gives the 5950X a single core score of 1655.

Geekbench 5 gives the M1 a single core score of 1745.

 

In Cinebench the 5950X is slightly ahead, and in Geekbench 5 the M1 is slightly ahead.

Don't you think that is reasonable? I mean, when you compare Cinebench to other "real benchmarks" the scores can fluctuate a bit too. It is not uncommon for two processors to trade blows in different benchmarks. Sometimes processor X wins, and sometimes processor Y wins.

 

Also, why are you bringing up the core count when we are talking about the single core scores? The fact that the 5950X has far more cores than the M1 doesn't matter when we are only running the benchmark on a single core.

 

 

Pretty much nothing you have said has been true. You're just making stuff up.

[saltycaramel]

A leaker just leaked the codename of the M2: Jade.

 

https://appleterm.com/2020/11/25/m2-only-for-imac-and-mac-pro/

 

And he said it will only go in the iMac and MacPro.

Also, Kuo said we won’t see redesigned MBPs in the H1 2021. (I’m guessing early H2)

 

So this is my updated AS Macs timeline based on these new leaks:

1) March 2021: iMac 5K 24” miniLed with M1T 16-core (12p+4e) and discrete Apple GPU (“Lifuka”)

2) July 2021: 14” MBP, 16” MBP and dark MacMini with M1X 12-core (8p+4e), plus a “Lifuka” dGPU on the 16” MBP but at reduced TDP compared to the one on the iMac

3) November 2021: iMac 6K 32” miniLED with M2T (5nm+ node) and “SFF” MacPro with M2T, with the big non-SFF Intel MacPro still being sold alongside it for another year

 

No new laptops until October/November 2022 with the M3. 

An iPad Pro kinda schedule. 1.5 to 2.5yrs.

[randomhkkid]

1 minute ago, saltycaramel said:

A leaker just leaked the codename of the M2: Jade.

 

https://appleterm.com/2020/11/25/m2-only-for-imac-and-mac-pro/

 

And he said it will only go in the iMac and MacPro.

Also, Kuo said we won’t see redesigned MBPs in the H1 2021. (I’m guessing early H2)

 

So this is my updated AS Macs timeline based on this new leaks:

1) March 2021: iMac 5K 24” miniLed with M1T 16-core (12p+4e) and discrete Apple GPU (“Lifuka”)

2) July 2021: 14” MBP, 16” MBP and dark MacMini with M1X 12-core (8p+4e), plus a “Lifuka” dGPU on the 16” MBP but at reduced TDP compared to the one on the iMac

3) November 2021: iMac 6K 32” miniLED with M2T (5nm+ node) and “SFF” MacPro with M2T, with the big non-SFF Intel MacPro still being sold alongside it for another year

That 14" laptop would be really interesting to me. Am hoping that they are able to leverage the smaller board size of the M1 to fit a 100WHr battery into a 14" frame, one can hope for a 24Hr battery.

[saltycaramel]

The 14” will be even more interesting for me if they put FaceID in it.

A press of the spacebar and you’re logged in.

Like on the iPad Pro + Magic Keyboard. 

[Laborant]

Am I allowed to point out, that fair and meaningful benchmarking between platforms is extremely hard? Modern CPUs - and especially the M1 - are highly heterogenous systems, constisting of more than just some ALUs with Registers and a bit of cache. Some applications might be accelerated by a hardware IP (like encryption and decryption or certain videocodecs), others depend on fast memory access and have maybe lots of cache-misses and some could actually be accelerated by custom-IPs like a DSP or an NPU.

 

Boiling the performance of such a system down to one single number is stupid. That's why good tech-channels show different bar graphs for several use cases. Different benchmarks cover different use-cases - most important are the ones actually covering the use-cases of your target audience. 

Thinking that the M1 is still a Mobile CPU (even though it's also present in the Mac Mini), even bragging about it being slower than a 105W Desktop CPU shows how impressive the 1st generation Chip from Apple really is.
Give the Apple Engineers some time to tailor the M1 for a pure desktop, loosen the power-budget a bit and these things will take the crown. 

[randomhkkid]

1 minute ago, saltycaramel said:

The 14” will be even more interesting for me if they put FaceID in it.

A press of the spacebar and you’re logged in.

Like on the iPad Pro + Magic Keyboard. 

Yeah for sure, this is what I miss most using a MBP 16 for work compared to my Surface Pro 4

[Bombastinator]

There seems to be a lot of disagreement about the quality and accuracy of comparison testing generally discussing the accuracy of geekbench as a benchmark.  Both sides seem to agree that actual testing in a suite of apps will show the speed of the device though.  Has this been done?

[thechinchinsong]

4 hours ago, Laborant said:

Am I allowed to point out, that fair and meaningful benchmarking between platforms is extremely hard? Modern CPUs - and especially the M1 - are highly heterogenous systems, constisting of more than just some ALUs with Registers and a bit of cache. Some applications might be accelerated by a hardware IP (like encryption and decryption or certain videocodecs), others depend on fast memory access and have maybe lots of cache-misses and some could actually be accelerated by custom-IPs like a DSP or an NPU.

 

Boiling the performance of such a system down to one single number is stupid. That's why good tech-channels show different bar graphs for several use cases. Different benchmarks cover different use-cases - most important are the ones actually covering the use-cases of your target audience. 

Thinking that the M1 is still a Mobile CPU (even though it's also present in the Mac Mini), even bragging about it being slower than a 105W Desktop CPU shows how impressive the 1st generation Chip from Apple really is.
Give the Apple Engineers some time to tailor the M1 for a pure desktop, loosen the power-budget a bit and these things will take the crown. 

Therein lies the rub. Seeing how Apple is a mobile first and mobile focused company, when will they (if they will ever) optimize that heterogenous architecture for desktop. It took them years to update the mac pro and that's without redesigning their own silicon. I'm not saying that's the wrong thing to do as a company, seeing as their company is built on mobile devices, but I'm not holding my breath for any desktop optimized Apple silicon just yet. 

[Spindel]

19 minutes ago, thechinchinsong said:

... when will they (if they will ever) optimize that heterogenous architecture for desktop...

It will be a slaughter. 
 

I’ve been working all day on my M1 Mini. Mainly AutoCAD and excel both under Rosetta and it trades blows with high end desktops (given both apps are first and foremost singelthreaded). 

[thechinchinsong]

1 minute ago, Spindel said:

It will be a slaughter. 
 

I’ve been working all day on my M1 Mini. Mainly AutoCAD and excel both under Rosetta and it trades blows with high end desktops (given both apps are first and foremost singelthreaded). 

I tell myself to not get my hopes up anytime soon, but I still can't help looking forward to it .

 

[Commodus]

6 minutes ago, Spindel said:

It will be a slaughter. 
 

I’ve been working all day on my M1 Mini. Mainly AutoCAD and excel both under Rosetta and it trades blows with high end desktops (given both apps are first and foremost singelthreaded). 

It seems like the M1 chip changed opinions of the Mac mini almost overnight. Before, it was mainly a niche computer you got if you wanted a Mac desktop but refused to get an iMac and were willing to make tradeoffs. Now, it's a pretty appealing machine for just about anyone who wants a desktop and isn't particularly attached to gaming or Windows apps.

 

I'd have a very hard time recommending the current 21.5-inch iMac over this, and even the 27-inch model is an "only if you need X" suggestion.

[RedRound2]

5 hours ago, Laborant said:

Some applications might be accelerated by a hardware IP (like encryption and decryption or certain videocodecs), others depend on fast memory access and have maybe lots of cache-misses and some could actually be accelerated by custom-IPs like a DSP or an NPU.

Even x86 has dedicated accelerator units for encryption/decryption and video codecs like H.265 decoder/encoder. 
 

The only extra thing the M1 has is the NPU as far as I know (I don’t remember if Intel implemented some form of it)

[RedRound2]

20 hours ago, Spindel said:

This thing is incredibly fast.

It's hard to describe. 

Yeah many people have echoed the same. They say it feels like an iPad in launching apps and going through UI. Having the 2020 MacBook Air and using an iPad I can definitely imagine what you’re talking about.

13 minutes ago, Spindel said:

It will be a slaughter. 
 

I’ve been working all day on my M1 Mini. Mainly AutoCAD and excel both under Rosetta and it trades blows with high end desktops (given both apps are first and foremost singelthreaded). 

Damn. AutoCad works well? I bought the new MacBook Air like 4 months back after knowing about Apple Silicon. But I assumed it would take a couple of years for software’s to mature and it to be on par with x86 and I needed a new laptop. Everything working this well right out of the gate kinda caught the whole industry off guard

[Spindel]

44 minutes ago, RedRound2 said:

Yeah many people have echoed the same. They say it feels like an iPad in launching apps and going through UI. Having the 2020 MacBook Air and using an iPad I can definitely imagine what you’re talking about.

Damn. AutoCad works well? I bought the new MacBook Air like 4 months back after knowing about Apple Silicon. But I assumed it would take a couple of years for software’s to mature and it to be on par with x86 and I needed a new laptop. Everything working this well right out of the gate kinda caught the whole industry off guard

Initially autodesks license check freaked out, a reboot fixed that. 
 

Only app I need that don’t work is OneNote, it  seems to work for other people so I’ll do some troubleshooting tomorrow. 

[Spindel]

I have to add that the icestorm cores seem to off load the system really well so the firestorm cores don’t have to waste cpu cycles on system services.

 

This actually works unlike HT/SMT, I hate it that OSs shows HT/SMT as logical cores. IT DOES NOT WORK THAT WAY! But that’s a different topic.

[saltycaramel]

That’s the other thing: even if Intel manages to pump out a GREAT heterogeneous CPU in the coming years, would Windows’ scheduler be as good as Apple’s in managing big and small cores? Apple being the same company doing the CPU+OS and with years of experience on iPhone/iPad at this...

[leadeater]

14 hours ago, RedRound2 said:

The only extra thing the M1 has is the NPU as far as I know (I don’t remember if Intel implemented some form of it)

VNNI, but it's not really widely available across all consumer CPUs yet, just select ones.