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

[leadeater]

7 minutes ago, LAwLz said:

At this point I don't think people would even be impressed if it cured cancer and generated electricity rather than consumed it. 

 

It's Apple so it must be bad, right? 

Well Apple did create cancer 

[LAwLz]

Just now, gabrielcarvfer said:

2014 arch + 14nm vs 2017 arch + 5nm (assuming 3 years of development for each uarch)

See my edit. That's compared to zen2 which is only 1 year old. 

 

Quote

For comparison, the 3600 at 4.2GHz gets a cinebench R23 score of around 8500.

 

So a desktop CPU with 50% more cores, 40% higher clock and 850% higher power consumption is only ~13% faster. 

 

And people aren't impressed by that? Holy crap... 

 

[captain_to_fire]

15 hours ago, FloRolf said:

This makes me wonder though, why are their higher priced models still on Intel CPU's when the M1 is apparently much faster? 

 

https://www.apple.com/shop/buy-mac/macbook-pro/13-inch

Because Intel charges a lot of money for their chips, that’s why the newer M1 MacBook Pro is cheaper than the Intel ones. The i5-1038NG7 (4c/8t/28W TDP) inside the 2020 13” MacBook Pro costs $320. 

[exhaustedbrit]

1 minute ago, like_ooh_ahh said:

Because Intel charges a lot of money for their chips, that’s why the newer M1 MacBook Pro is cheaper than the Intel ones. The i5-1038NG7 (4c/8t/28W TDP) inside the 2020 13” MacBook Pro costs $320. 

It's also worth mentioning, the distinction between the two port MacBook Pro and the higher end MacBook Pro with four ports carries through here. The two-port models have always had one (albeit slightly bigger) fan whereas the four-port models have an additional heatpipe and two fans, my wife's 2018 pro has 4 ports and can handle just above 32 watts at full tilt in my testing. Despite the two-port model previously having a 15W TDP, Apple didn't run it at this and would still manage a hair above 25 watts, which makes me wonder if they're running the M1 chip cooler, or if the package-on-package design prevents it from really going much further thermally. Because 25W is a lot more than 10W I'd expect more than a few percent in terms of bench results.

[Master Delta Chief]

26 minutes ago, like_ooh_ahh said:

Because Intel charges a lot of money for their chips, that’s why the newer M1 MacBook Pro is cheaper than the Intel ones. The i5-1038NG7 (4c/8t/28W TDP) inside the 2020 13” MacBook Pro costs $320. 

Yet the M1 Macs consumer prices stayed relatively the same in comparison to their predecessor models which didn't include a self-made processor. It is funny that there were some outlets that claimed the Macs were getting cheaper.

[lostcattears]

If only If only  Intel just stop using 14nm and use a competitors's 7nm/5nm I am sure they can CRUSH amd...

[NotTheFirstDaniel]

Someone got their M1 MacBook Pro a day early and ran Cinebench R23 (Native/Universal).

https://twitter.com/mnloona48_/status/1328473223193784321?s=20

 

SC - 1493

MC - 7566

 

It beats all chips on this leaderboard anyways until you get to the i7-1165G7, and then Zen3. The i7 is also a 28W part though so I don’t know if it’s comparable, and then of course Zen3 is Zen3. Multi core lacks a lot because it’s only a quad core CPU, but the SC pretty good for a 10-18W part.

 

I can’t wait to see what they do with the 16” MBP and the Mac Pro.

[captain_to_fire]

1 hour ago, Master Delta Chief said:

Yet the M1 Macs consumer prices stayed relatively the same in comparison to their predecessor models which didn't include a self-made processor. It is funny that there were some outlets that claimed the Macs were getting cheaper.

The M1 MacBook Pro is definitely cheaper than the older one with 10th gen i5

 

 

[NotTheFirstDaniel]

29 minutes ago, like_ooh_ahh said:

The M1 MacBook Pro is definitely cheaper than the older one with 10th gen i5

It's a bit confusing, but consider those 13" MacBook Pros to be like a MacBook Pro Pro or a MacBook Pro Max. Those were formerly called the Touchbar MBP 13", but now that every MacBook Pro has a touchbar, they're just called the expensive ones no one buys.

 

The MacBook Pro 13" started at $1299 for a 1.4 GHz i5 (with TB of course, just a low base clock) and 8/256 if I remember. The only difference between them are 2 extra thunderbolt ports. But it's to be said that there's some limitation with the M1 chip which means you can't have more than two thunderbolt ports or hook up more than one monitor. So they couldn't upgrade the 4 Thunderbolt model with the current M1 and the more performant M1-class SOCs are still probably in testing.

 

If I had a crystal ball (also based off of previous leaks), this $1799 model will most likely be replaced in March with a 14" MacBook Pro alongside a 16" MacBook Pro with an M1X or M2, depending on if they're done with this core already (I don't see this though they'll probably keep the Macs in line with the A series for now, so most likely an M1X until an A15-based M2 comes out late next year). That would give you your extra thunderbolt ports and more performance.

[Dabombinable]

27 minutes ago, like_ooh_ahh said:

The M1 MacBook Pro is definitely cheaper than the older one with 10th gen i5

 

 

That's a $320 USD CPU...that price discrepancy is far more than just because its an i5.
https://ark.intel.com/content/www/us/en/ark/products/196594/intel-core-i5-1038ng7-processor-6m-cache-up-to-3-80-ghz.html

Apple has been notorious for overpricing any of their Macbooks with a Core ix CPU compared to their competitors.
Also, $200 to go from 256GB to 512GM with the M1 based Macbook, and $200 to go from 512GB  to 1TB in the i5 Macbook.

[captain_to_fire]

2 minutes ago, gabrielcarvfer said:

Half RAM and half storage. Meh.

Well even if someone opted for the 1TB SSD/16GB RAM upgrade, the M1 MacBook Pro is still cheaper than the Intel MBP...by $100

 

[Blademaster91]

51 minutes ago, like_ooh_ahh said:

Well even if someone opted for the 1TB SSD/16GB RAM upgrade, the M1 MacBook Pro is still cheaper than the Intel MBP...by $100

 

No 32GB option, and 2 less thunderbolt ports.

Edit- Or an option for a 4TB SSD.

Edited November 17, 2020 by Blademaster91

[NotTheFirstDaniel]

3 minutes ago, Dabombinable said:

That's a $320 USD CPU...that price discrepancy is far more than just because its an i5.
https://ark.intel.com/content/www/us/en/ark/products/196594/intel-core-i5-1038ng7-processor-6m-cache-up-to-3-80-ghz.html

Apple has been notorious for overpricing any of their Macbooks with a Core ix CPU compared to their competitors.
Also, $200 to go from 256GB to 512GM with the M1 based Macbook, and $200 to go from 512GB  to 1TB in the i5 Macbook.

Storage and RAM I agree.

 

But with CPUs they're actually in line with other brands. Like the Dell XPS line and MacBook Air lines followed similar pricing structures for the CPU for the i3 and i5... and then the i7 falls apart because of Apple's insane pricing for storage.

[Bombastinator]

1 hour ago, Master Delta Chief said:

Yet the M1 Macs consumer prices stayed relatively the same in comparison to their predecessor models which didn't include a self-made processor. It is funny that there were some outlets that claimed the Macs were getting cheaper.

So 10% was not a big enough price drop for you then. 

[Master Delta Chief]

1 hour ago, Bombastinator said:

So 10% was not a big enough price drop for you then. 

I should've been more specific, but prices in Belgium in my case have stayed relatively the same with only a minor decrease. Thing is as mentioned by others in the thread, they also get less RAM and same goes for storage. 

[Bombastinator]

1 minute ago, Master Delta Chief said:

I should've been more specific, but prices in Belgium in my case have stayed relatively the same with only a minor decrease. Thing is as mentioned by others in the thread, they also get less RAM and same goes for storage. 

Well 10% could be considered a minor decrease.  I’m not sure how the EU pricing is working, and Belgium may have other factors at play.  Probably not as extreme as India’s I suspect.  Someone in the thread mentioned that even with apple’s highly expensive factory upgrades, the thing was still slightly cheaper after even this was done.   Pricing of everything varies by location.  It’s still all supposition in any case.  The device will release, things will be tested, prices will harden. 

[hishnash]

18 hours ago, leadeater said:

 

 

Apple for sure is working on higher power SoC designs that will have better I/O options and will target higher end SKUs of product lines, I don't know when but there will be Mac Mini's with 10Gb again.

The fact apple did not replace the intel macMini (they are still selling it unlike the AIR and 2port MBP 13") means they do plan on replacing it and i'm sure we will get the 10Gb option there again. 

 

 

13 hours ago, LAwLz said:

Once they start replacing their 16" Macbooks for example then they will probably have chips with more connectivity.

They will for sure, if they were not planning on having more ports they would have just put this ship in at least the 4-port MBP 13 it is already faster in GPU and CPU than the best intel configuration but they clearly do not thing they have replaced it it as it is still on sale unlike the 2port 13" were there is not intel option left on sale

 

13 hours ago, LAwLz said:

Microsoft's x86 emulation layer in Windows for ARM seems to suck ass compared to Rosetta Stone 2. 

So i think there are 2 reasons for this: 1) apple have much more experience writing very well optimised compilers that target both x86 and ARM64 (LLVM) 2) To be able to do the tricks they are doing apple are depending on dedicated features they have added to the memory controler in silicon on the M1 that let them make the memory opeations (when running rosseta2 translated apps) behave just the same as x86 this means they just need to translate the x86 assebly instructions to arm versions (on disk) before the app runs then run the app as an arm app without any emulation/man in the middle opeartions that MS need to do to ensure the memory access behaves the same as the apps expect.

 

[RoseLuck462]

Amazing how powerful this 1st gen tech is @_@

[Bombastinator]

4 minutes ago, RoseLuck462 said:

Amazing how powerful this 1st gen tech is @_@

I don’t know if it’s actually 1st gen tech or not.  That would depend on how similar m1 is to the other CPUs Apple has done is.

[CarlBar]

@LAwLz First my point is it's not the "OMG Intel killer" levels that where being bandied about. It's still an impressive piece of work, but it's not the unbelievably better chip that it was being made out as.

 

Second @leadeater covered a couple of pages back how IO and memory support adds a lot to the power usage of a chip. And when your down in the ultra low power usage of a chip like this that really starts to matter as every watt taken by IO is a watt the cores aren't using.

 

Third Silicon is normally binned. For things like Intel and AMD mainstream CPU's this means the desktop chips are by far the least efficient silicon. As an example even withing the desktop skews look at the Zen 2 64 core threadripper vs the 16 core 3950x. 4 times the cores but only twice the TDP and with a much bigger IO die soaking up some of the extra wattage. The server chips will have even better silicon. start factoring that in with @leadeater 's graph on the 5950X and the per core power consumption is getting down to somwhere likely around 2W per core.

 

We don;t have a clear idea how apple is binning, but there's little reason they couldn't bin tightly. AMD and Intel can;t do it on desktop normally because the demand for high quality silicon from the server side makes it impractical, (they'd need to manufacture a lot more chips and both are using all the capacity they've got), and server gets priority because it's higher margins. Apple doesn't have that. they're free to treat their low power Mainstream offerings like AMD and Intel do their Server offerings. And they'll be free to treat their mainstream mid to high power offerings like Intel and AMD do their Mobile chips, (which generally get the next best silicon after server side).

 

Fourth, Apple is manufacturing this on 5nm, according to TSMC that equates to a 20-30% drop in power use if i remember right. Remember that 2W per core figure from earlier, well now where looking at even less. Using the 5600x as an example it has roughly the same per core Multi-core performance as the M1 and using server grade silicon on a 5nm variant you probably could get a version with minimal IO into a 10W TDP. It's right on the limit of what i think AMD could do and Intel is definitely behind the curve right now. But it's not somthing that x86 can't match if someone really wanted to.

 

 

Like i said thats not to say it's not impressive. Apple's gone and put together somthing thats right on par with the bleeding edge of what plausible with x86 , once you account for things like minimal IO, process node, binning, e.t.c. But it's not the x86 demolisher that the hype was trying to sell people.

[Spindel]

FCPX performance on M1 mac compared to iMac Pro:

 

Exporting H.264 Sony 10 bit 422 footage with just one rec709 lut takes:

 

- 11 minutes and 30 seconds on iMac Pro with Vega 56 and 128gigs of ram.
- 10 minutes and 20 seconds on M1 MacBook Pro with 8gigs of ram.

 

30 seconds long H.265 canon 10-bit 422 footage at 100fps takes 

- 80 seconds on iMac Pro 
- 45 seconds on new MacBook Pro

 

 

https://forums.macrumors.com/threads/fcpx-performance-on-m1-chip.2268919/

[05032-Mendicant-Bias]

6 hours ago, NotTheFirstDaniel said:

Someone got their M1 MacBook Pro a day early and ran Cinebench R23 (Native/Universal).

https://twitter.com/mnloona48_/status/1328473223193784321?s=20

 

SC - 1493

MC - 7566

 

It beats all chips on this leaderboard anyways until you get to the i7-1165G7, and then Zen3. The i7 is also a 28W part though so I don’t know if it’s comparable, and then of course Zen3 is Zen3. Multi core lacks a lot because it’s only a quad core CPU, but the SC pretty good for a 10-18W part.

 

I can’t wait to see what they do with the 16” MBP and the Mac Pro.

Now, this is at least a believable score, and still very impressive considering TDP.

Half performance of a 3600x at a fraction of the power.

Power efficiency comes from the better process and the weaker simpler core architecture.

[LAwLz]

7 hours ago, gabrielcarvfer said:

The screenshot shows the 1700x, that could be kept at 60 watts by reducing clocks to 3GHz.
Assuming each node shrink saves 50% of power while keeping performance the same, we would have 60w->30w->15w. Additional savings from uarch optimizations and boom, doesn't sound that impressive.

50% power save? You are insane? 

That is way more than we typically get. The move from 7 to 5 for example gave us 30% reduction. 

5nm to 3nm is quoted by TSMC as 25-30%.

 

You're also talking about overvolting some kind of golden sample chip by 30%, and we still have that part about this being a quad core vs a 6 core Zen2 CPU. 

 

Just admit that this is really impressive. All your mental gymnastics are embarrassing. 

[Spindel]

43 minutes ago, Spindel said:

FCPX performance on M1 mac compared to iMac Pro:

 

Exporting H.264 Sony 10 bit 422 footage with just one rec709 lut takes:

 

- 11 minutes and 30 seconds on iMac Pro with Vega 56 and 128gigs of ram.
- 10 minutes and 20 seconds on M1 MacBook Pro with 8gigs of ram.

 

30 seconds long H.265 canon 10-bit 422 footage at 100fps takes 

- 80 seconds on iMac Pro 
- 45 seconds on new MacBook Pro

 

 

https://forums.macrumors.com/threads/fcpx-performance-on-m1-chip.2268919/

And further.

 

10bit 422 canon h265 8k ipb footage for 30 seconds with one rec709lut.

- 2 min and 47 sec on iMac Pro vs
- 62 seconds on M1 MacBook Pro.

 

[LAwLz]

5 hours ago, Blademaster91 said:

No 32GB option, and 2 less thunderbolt ports.

Edit- Or an option for a 4TB SSD.

Stop moving the goal post. Here is the entire conversation.

 

Someone: The prices stayed relatively the same. Some outlets claimed that Macbooks were getting cheaper.

Someone else: It did get cheaper compared to the i5.

A third person: Yeah but with half the RAM and storage.

Person 2 again: Even if you upgrade to them to have the same RAM and storage it's still 100 dollars cheaper.

You: But you can't upgrade it to 32GB or a 4TB SSD!!!!

 

The conversation wasn't about which laptop could be upgraded the most. It was which was cheaper and the argument that "it's only cheaper because it has less RAM and less storage" was debunked.

 

 

1 hour ago, Bombastinator said:

I don’t know if it’s actually 1st gen tech or not.  That would depend on how similar m1 is to the other CPUs Apple has done is.

The A14 is very similar in some aspects, and very different in other aspects. I think it's pretty safe to say that the core design is not "first gen" though. They have been making custom cores for ages.

 

 

45 minutes ago, CarlBar said:

@LAwLz First my point is it's not the "OMG Intel killer" levels that where being bandied about. It's still an impressive piece of work, but it's not the unbelievably better chip that it was being made out as.

Like I said in my previous post. You honestly don't think it is impressive that a 10 watt quad core can keep up with a Zen2 CPU that:

1) Have 50% more cores.

2) Have 40% higher clocks.

3) Have 850% higher power consumption.

4) Have desktop cooling.

 

Are you for real? What would you have thought was impressive then? A single core outperforming a 6 core or something? You have to take into account the fewer cores, lower clock speed, less cooling and way lower power usage for it to be impressive. If all you care about is some maximum performance figure then you should have read my post several pages ago where I said something along the lines of "AMD and Intel will still beat it in multicore performance, but only once you start looking at their 6 cores".

 

 

49 minutes ago, CarlBar said:

Second @leadeater covered a couple of pages back how IO and memory support adds a lot to the power usage of a chip. And when your down in the ultra low power usage of a chip like this that really starts to matter as every watt taken by IO is a watt the cores aren't using.

We don't know if increasing the IO will balloon the power usage like leadeater claims it will. I get that Apple haters will clutch to that idea for as long as possible now that you have been disproven on the whole "all the benchmarks that have been released are just bad and doesn't count!"

Also, @leadeateris deliberately trying to make AMD look as good as possible by showing the power consumption on the 5950X which according to Anandtech had some rather odd behavior when it came to power consumption. If we instead look at the 5900X, the 5800X or the 5600X we see a much more expected and linear power consumption increase as the core count increases.

A single Zen3 core uses more power than the entire M1 chip uses. This is not something AMD can just optimize away with less IO and a node change.

https://www.anandtech.com/show/16214/amd-zen-3-ryzen-deep-dive-review-5950x-5900x-5800x-and-5700x-tested/8

Spoiler

 

 

 

Even if we completely subtract all the power of the non-CPU part of these Zen3 chips, we still end up with a single Zen3 core using more power than the entire M1 has allocated to it (with 4 big cores).

 

Like I said earlier, according to TSMC the move from 7nm to 5nm only gives 20% power reduction. This is not just the difference between 7nm and 5nm and a bigger IO die. Stop being in denial about this.