M1 Macs Reviewed

[NotTheFirstDaniel]

3 hours ago, ColeWorld said:

It looks like Linus' early impression and skepticism to the announcement video wasn't echoed by other reviewers because they had the M1 already in hand being tested. Dave Lee sorta hinted at that in the WAN show last week too. This rift between LMG and Apple just puts LMG a bit behind on reviews for Apple products. 

You could hear it Dave's voice. Also I think he literally said "I'm on NDA so I can't say anything" on the WAN show.

 

I'm really not surprised at the performance. I mean it's incredible don't get me wrong but I expected it. I didn't hype myself up too much and I didn't hold myself back too much either.

 

What's funny to think is that this M1 chip will represent the absolute worst Apple Silicon chip Apple will ever produce. The worst one. Let that sink in. They've hit the ground running and fast too.

[Curufinwe_wins]

10 minutes ago, NotTheFirstDaniel said:

You could hear it Dave's voice. Also I think he literally said "I'm on NDA so I can't say anything" on the WAN show.

 

I'm really not surprised at the performance. I mean it's incredible don't get me wrong but I expected it. I didn't hype myself up too much and I didn't hold myself back too much either.

 

What's funny to think is that this M1 chip will represent the absolute worst Apple Silicon chip Apple will ever produce. The worst one. Let that sink in. They've hit the ground running and fast too.

I mean yes, but also A14 only managed a 5% ipc improvement over A13. And A13 was less of an improvement over A12 than A12 was over A11. I don't want to pretend like I'm expecting Apple to flounder, but they have been slowing down their CPU growth recently. GPUs and other dedicated silicon? Now that is flying ahead. But it's worth noting that it seems like Apple is also hitting some diminishing return situations. 

 

This may not actually be the worst relative position Apple has compared to incumbent x86, but that all depends on how much iterative performance from here out each side can deliver. And we've been waiting for quite a while still for Intel to pick back up. Obviously AMD has been punching yoy gains far and above anyone else atm.

 

 

 

Quote

For the rest of the workloads, the A14 generally looks like a relatively linear progression from the A13 in terms of progression, accounting for the clock frequency increase from 2.66GHz to 3GHz. The overall IPC gains for the suite look to be around 5% which is a bit less than Apple’s prior generations, though with a larger than usual clock speed increase.

 

 

Also also! This literally validates Linus's position IMO. M1 is only 10-15% faster than A14 despite being almost 4x the power (comparing peak to peak with iphone vs mac mini). I do hope that Apple can broaden their performance curves a bit more moving forward.

 

It is an iPad. Now turns out, iPads are pretty darn powerful though.

[NotTheFirstDaniel]

1 minute ago, Curufinwe_wins said:

I mean yes, but also A14 only managed a 5% ipc improvement over A13. And A13 was less of an improvement over A12 than A12 was over A11. I don't want to pretend like I'm expecting Apple to flounder, but they have been slowing down their CPU growth recently. GPUs and other dedicated silicon? Now that is flying ahead. But it's worth noting that it seems like Apple is also hitting some diminishing return situations. 

 

This may not actually be the worst relative position Apple has compared to incumbent x86, but that all depends on how much iterative performance from here out each side can deliver.

What's the IPC difference between the X series? It seems like they update that every die shrink. Those are the chips that are actually tuned for performance, the regular A series is tuned much more for efficiency. So that would be A9X v A10X v A12X v M1?

[Curufinwe_wins]

Just now, NotTheFirstDaniel said:

What's the IPC difference between the X series? It seems like they update that every die shrink. Those are the chips that are actually tuned for performance, the regular A series is tuned much more for efficiency. So that would be A9X v A10X v A12X v M1?

I don't have that info now, but that is multigenerational gaps in between. 2 each time for the last two. Not sure I buy using that as the comparison basis honestly.

[RedRound2]

This is quite amazing for their first Mac chip. And the reviewers are surprised as well. Not only it’s a huge gain in performance, compatibility seems to be a non issue under Rosetta 2 but it also has great battery life to boot.

 

Dave2D’s review is great. There’s more data in his review and just look at how excited he is after using for a week

 

 

Wonder what happened to the folks online who really wanted this to fail?

[NotTheFirstDaniel]

5 minutes ago, Curufinwe_wins said:

I don't have that info now, but that is multigenerational gaps in between. 2 each time for the last two. Not sure I buy using that as the comparison basis honestly.

Well the A12X trounced over the A13, and is about on par in single core but higher in multi-core with the A14. Of course, the A12X uses much more power than the A14, but it says a lot when it will take 3 generations of low power CPUs to finally surpass what the A12X did. And of course the M1 (and the upcoming A14X) are much more powerful than the A14.

 

And with the A14 in particular Apple focused on efficiency first because of 5G in the iPhone 12. They barely mentioned performance and focused on power efficiency instead.

[NotTheFirstDaniel]

1 minute ago, RedRound2 said:

Wonder what happened to the folks online who really wanted this to fail?

You have your usual "Apple Bad!!!!" minorities.

 

But I think a lot of it was skepticism, like with Zen1. Until we got it, we really didn't know how good it would be, but now that we have it, it's clear that the demos weren't BS-ing.

 

I wonder if we may get a rebuttal to the I think year or two old video called "Macs are SLOWER than PCs. Here's why." with "Macs are FASTER than PCs. Here's why." at the end of this transition when the desktop chips come out (of course they'd be up against Zen4 on a die shrink, so GL to Apple).  I don't think they'll actually do that though, look at the like to dislike ratio on "10 ways iPhones are just better". Jesus Christ, his subscribers did not like that!

[thechinchinsong]

Really exciting stuff. Now the "Apple-tax" covers not only great software, but also decent hardware. Will definitely consider getting the 2nd generation of these Apple silicon Macs.

[A51UK]

If the benchmark right, the M1 did ok but not as good as the AMD Zen 3  like the AMD Ryzen 5 5600X . So ARM not going take over and Apple Mac are not as powerful as the new computers that come out soon but it did far better them I did think it would and that good news. 

 

I did watch a review today that did high light some problem and poor performance of the Rosetta 2.  

 

 

 

[Bombastinator]

21 minutes ago, RedRound2 said:

This is quite amazing for their first Mac chip. And the reviewers are surprised as well. Not only it’s a huge gain in performance, compatibility seems to be a non issue under Rosetta 2 but it also has great battery life to boot.

 

I recommend watching Dave2D review. There’s more data in his review and he seems genuinely blown away by this chip

 

 

Wonder what happened to the folks online who really wanted this to fail?

Watched it.  It works apparently a LOT better than expected for a lot of things.   OK.  All the comparisons made were between it and other MBs though.  No comparisons between the m1s and non MacBook stuff.  With 2 exceptions:

 

 1 was a guy that built a superhardcore hackintosh using a 5950.  It got beat(!) so cpu performance is apparently wow!

2 was a GPU test which is not as strong as the CPU stuff.  The guy played a non optimized PC aaa game at full 1440p (30xx x 1440) with high settings and got 17fps.  There was a top end MBP with the biggest discrete graphics card available on Mac (which generally mean medium to medium high for everything else) that waxed it with 56fps.  Nothing else did though.  Other stuff was lower.  So the way I see this, is the M1 machines are up there with literally everything in cpu, and literally everything in iGPU and even low end discrete gpu.  There are things that can beat them and even beat them bad, but they need a big discrete gpu to do it, (so beats all thin-and-lights, most non thin and light laptops and some desktops)  and only in situations where a big gpu provides an advantage like gaming.   It’s sounding to me like if they can get some sort of gpu expandability into these things, they could meet or crush anything current at all. The problem is that while their iGPU is awesome as far as iGPUs go, it’s still an iGPU and there is no way to attach anything else currently. 

 

[leadeater]

3 hours ago, LAwLz said:

Also, the SPEC2006 and SPEC2017 results are more or less the same. Can we please stop hating on SPEC2006 because "hurr durr it's old so therefore it's bad" now?

Changing from SPEC2006 to SPEC2017 changes things a little, but not to the point where one chip might look really good and then in the newer/older test it looks bad all of a sudden.

Ah no, because (and I haven't gone in to look at each test result specifically yet) it shows precisely what I was pointing out. If you don't think 10% difference in INT and 5% difference in FP matters for performance analysis then I have literally no idea what to say, that's more than some production generation improvements we've gotten in the past so to write that off as "close enough" is dumb.

 

Run the new thing, not the old. SPEC2006 belongs in the bin at this point. We have a replacement for it, SPEC20017 so run that.

 

And you still seem to be stuck in a mindset around my comments on it that I never said, just drop it. At this point you're arguing against something you've made up in your head and not what I actually said at all.

[Curufinwe_wins]

46 minutes ago, NotTheFirstDaniel said:

Well the A12X trounced over the A13, and is about on par in single core but higher in multi-core with the A14. Of course, the A12X uses much more power than the A14, but it says a lot when it will take 3 generations of low power CPUs to finally surpass what the A12X did. And of course the M1 (and the upcoming A14X) are much more powerful than the A14.

 

And with the A14 in particular Apple focused on efficiency first because of 5G in the iPhone 12. They barely mentioned performance and focused on power efficiency instead.

Except the M1 isn't that much more powerful than the A14. (Peak to peak esp single core). That's the key. Anandtech's dive into the Mac Mini showed peak clocks of 3.2 GHz and sustained of 3 GHz which actually is the same as the peak of the A14 high power cores. The benchmarks seem to indicate only a 10-15% perf between the peak of the A14 (consuming around 6W in the process) and sustained loading of the M1 (~20-25W in the mac mini). Now it is worth noting that that does include some additional separate hardware but that is true for all laptops and phones as well.

 

The main perf gains of M1 relative to A14 are actually not frequency (near identical) or uarch (shared), but increased cache sizes and memory access gains. Now being 4 firestorm vs 2 firestorm is big on MT of course, but MT the 16/32 core desktop chips obviously trash these (not a fair fight).

 

Real talk time... that the M1 is indeed competitive with the best of mobile x86 is pure and total validation on the argument that for the hyper majority of users, A14 and even A12/S865 are way more powerful than users have use for, and buying new phones at current prices is a mistake if 'speed/performance' is what you are looking for. Future generations hopefully offering efficiency gains at iso perf is what we really want/need right now.

[EarlTheAdopter]

21 minutes ago, thechinchinsong said:

Really exciting stuff. Now the "Apple-tax" covers not only great software, but also decent hardware. Will definitely consider getting the 2nd generation of these Apple silicon Macs.

Unfortunately the apple-tax might end up being higher on that second gen, especially if it really should introduce new chassis / models. 

[Bombastinator]

18 minutes ago, A51UK said:

If the benchmark right, the M1 did ok but not as good as the AMD Zen 3  like the AMD Ryzen 5 5600X . So ARM not going take over and Apple Mac are not as powerful as the new computers that come out soon but it did far better them I did think it would and that good news. 

 

I did watch a review today that did high light some problem and poor performance of the Rosetta 2.  

 

 

 

Video won’t play for me

[Bombastinator]

6 minutes ago, Curufinwe_wins said:

Except the M1 isn't that much more powerful than the A14. (Peak to peak esp single core). That's the key. Anandtech's dive into the Mac Mini showed peak clocks of 3.2 GHz and sustained of 3 GHz which actually is the same as the peak of the A14 high power cores. The benchmarks seem to indicate only a 10-15% perf between the peak of the A14 (consuming around 6W in the process) and sustained loading of the M1 (~20-25W in the mac mini). Now it is worth noting that that does include some additional separate hardware but that is true for all laptops and phones as well.

 

The main perf gains of M1 relative to A14 are actually not frequency (near identical) or uarch (shared), but increased cache sizes and memory access gains. Now being 4 firestorm vs 2 firestorm is big on MT of course, but MT the 16/32 core desktop chips obviously trash these (not a fair fight).

That doesn’t say the m1 is slow though.  It just says the a14 is ridiculously fast.

[Bombastinator]

11 minutes ago, leadeater said:

Ah no, because (and I haven't gone in to look at each test result specifically yet) it shows precisely what I was pointing out. If you don't think 10% difference in INT and 5% difference in FP matters for performance analysis then I have literally no idea what to say, that's more than some production generation improvements we've gotten in the past so to write that off as "close enough" is dumb.

 

Run the new thing, not the old. SPEC2006 belongs in the bin at this point. We have a replacement for it, SPEC20017 so run that.

 

And you still seem to be stuck in a mindset around my comments on it that I never said, just drop it. At this point you're arguing against something you've made up in your head and not what I actually said at all.

This apparently was done.  The results were near identical. 

[BuckGup]

So when are we going to see a Mac Pro with discrete apple graphics and an equivalent xeon based M1 chip?

[Curufinwe_wins]

3 minutes ago, Bombastinator said:

That doesn’t say the m1 is slow though.  It just says the a14 is ridiculously fast.

100%, in fact that just goes to the only concern or rather question I have which was on pg 1 about performance/power scaling.

 

Also can see edit on the previous comment there.

[leadeater]

2 minutes ago, Bombastinator said:

This apparently was done.  The results were near identical. 

No they are not, literally read my post. The difference is outside just dismissing it as "the same".

 

Edit:

When is the last time in a performance review we looked at two different CPUs and one was 10% faster and we all went "yea basically the same the difference does not matter", yet now somehow it's fine and no problem when it's a benchmark tool. Has everyone lost the plot?

[Bombastinator]

1 minute ago, BuckGup said:

So when are we going to see a Mac Pro with discrete apple graphics and an equivalent xeon based M1 chip?

The “xenon based M1 chip” thing leaves me confused. Xenon is intel. Do you mean socket compatible with current xenon?  I don’t think that will ever happen. 

[Curufinwe_wins]

Just now, Bombastinator said:

The “xenon based M1 chip” thing leaves me confused. Xenon is intel. Do you mean socket compatible with current xenon?  I don’t think that will ever happen. 

No, he means a scaled up system that can compete with the dGPU workstation boys and a cpu beefed up in cores and similar to compete with Epyc/Xeon. (Or even TR3 / 5950x MT)

[Bombastinator]

2 minutes ago, leadeater said:

No they are not, literally read my post. The difference is outside just dismissing it as "the same".

You’re saying the 2006 and 2017 versions of the benchmark were not similar? Or that the 2017 test was not run?  Someone on a previous page said it was run, and that they were similar. 

[leadeater]

14 minutes ago, Curufinwe_wins said:

That's the key. Anandtech's dive into the Mac Mini showed peak clocks of 3.2 GHz and sustained of 3 GHz which actually is the same as the peak of the A14 high power cores

 

7 minutes ago, Bombastinator said:

sustained loading of the M1 (~20-25W in the mac mini). Now it is worth noting that that does include some additional separate hardware but that is true for all laptops and phones as well.

 

So I've just gotten out of bed so I have not read any of the review or really the thread but reading this post I'd like to just double check, so the M1 doesn't run at 3GHz rather 3.2GHz for many SC test? Also the power draw of the M1 (in this Mac Mini) is greater than 10W? Both these two bits of information indicate to me that the M1 is a lot less of the halo some have been trying to make out it is. Don't get me wrong it's still very good, but not the performance per watt metric people were using.

 

Did I understand what you said wrong?

[Bombastinator]

2 minutes ago, Curufinwe_wins said:

No, he means a scaled up system that can compete with the dGPU workstation boys and a cpu beefed up in cores and similar to compete with Epyc/Xeon. (Or even TR3 / 5950x MT)

Ah.  The impression I got from the result of the guy that ran a 5950 hackintosh is that it already does compete cpu wise. (Though I may be confused there) It’s the dGPU that is the issue. 

[Maticks]

Wait for the m2 early adoption is never rewarded.

This is good for Devs to iron things out.

Not sure if these benchmarks are right.

Benchmarks are one thing. Actual programs and experience is another.