Apple Silicon M1: benchmarks and review for Mac Mini, MacBook Air and MBP 13 M1 and comparison with other brands

[dboris]

CS Go benchmarks someone?

 

[Sanpete]

As for the reviews, it appears Renoir (especially in its 35w-45w variants) remains the highest performing CPU in multithreaded performance, and the M1 has a big edge in single thread. The 25w version of 4800u still seems to be faster in multithreaded performance. Of course Renoir has a higher power draw, but the M1 is on 5nm and on the MacBook Pro/Mac Mini is using seem to use much more than 10w. ?

That could be, except when throttling is considered. @nikidimi says above that the 4800U loses 25% performance when it throttles, putting it on a par for multi-core with the actively cooled M1. That's a big advantage of a lower TDP. (One reason I asked about a laptop with the chip rather than just the chip earlier is that in a particular laptop it has to be effectively cooled, etc.)

 

[echodriver]

That could be, except when throttling is considered. @nikidimi says above that the 4800U loses 25% performance when it throttles, putting it on a par for multi-core with the actively cooled M1. That's a big advantage of a lower TDP. (One reason I asked about a laptop with the chip rather than just the chip earlier is that in a particular laptop it has to be effectively cooled, etc.)

The 4800U can be set at different TDP's.

If it throttles, it's likely because the higher TDP setting isn't matched by the cooler in the laptop, which is definitely a problem with a lot of OEM's out there.

Anywho, now look at this: https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested/2

The 15W 4800U beats it in MT at 9,286 against the 7,833 of the Mac mini at Native in Cinebench R23.

And that's with a 15W locked 4800U, which is a much more apples-to-apples comparison against the M1 which is probably drawing at 15W too (the Cinebench R23 power draw #'s people are saying adds up to ~15W as well).

And this is Cinebench R23, which runs a lot longer and thus thermal throttling can affect things more if cooling can't keep up.

Either way, it's clear that Intel is toast. AMD already took the mobile performance crown this past year, and just took the desktop performance crown away, and now Apple is adding fuel to the overheating Intel CPU fire.

 

[jeanlain]

And that's with a 15W locked 4800U, which is a much more apples-to-apples comparison against the M1 which is probably drawing at 15W too (the Cinebench R23 power draw #'s people are saying adds up to ~15W as well).

is the 4800U actually locked at 15W? Usually, those chips can go higher than the TDP numbers for some amount of time.

 

[Sanpete]

The 4800U can be set at different TDP's.

If it throttles, it's likely because the higher TDP setting isn't matched by the cooler in the laptop, which is definitely a problem with a lot of OEM's out there.

Anywho, now look at this: https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested/2

The 15W 4800U beats it in MT at 9,286 against the 7,833 of the Mac mini at Native in Cinebench R23.

And that's with a 15W locked 4800U, which is a much more apples-to-apples comparison against the M1 which is probably drawing at 15W too (the Cinebench R23 power draw #'s people are saying adds up to ~15W as well).

And this is Cinebench R23, which runs a lot longer and thus thermal throttling can affect things more if cooling can't keep up.

Either way, it's clear that Intel is toast. AMD already took the mobile performance crown this past year, and just took the desktop performance crown away, and now Apple is adding fuel to the overheating Intel CPU fire.

Right, but even the 15W version actually runs a lot hotter than that, according to reports here (as does the M1 in the actively cooled housings), so throttling may remain a problem. In comparing the M1 Air to the M1 MBP, the throttling only showed up when R23 was on a loop, so it might not show for the 4800U either in the results you link to.

 

[Icelus]

And, most impressively, the M1 MacBook Pro used extremely little power to do so. Just 17% of the battery to output an 81GB 8k render. The 13” MacBook Pro could not even finish this render on one battery charge.

Yeah, Apple’s M1 MacBook Pro is powerful, but it’s the battery life that will blow you away | TechCrunch

Survival and strategy games are often played in stages. You have the early game where you’re learning the ropes, understanding systems. Then you have

techcrunch.com

 

[wiesson]

CS Go benchmarks someone?

Looks like CS Go uses OpenGL? This guy tried to start it

 

[donth8]

is the 4800U actually locked at 15W? Usually, those chips can go higher than the TDP numbers for some amount of time.

It is not, see this chart from another Anandtech article:

 

[echodriver]

It is not, see this chart from another Anandtech article:
View attachment 1671246

Two things: one, it depends on if they lock it or not. 'Normal' operation allows it to boost above 15W, yes, but AMD is pretty liberal allowing people to set target power consumption. For instance, you can set the 65W 5600X (which can draw a max of 88W stock) to "just" 65W - you lose 2-5% in performance but consume 33% less power. You can even set it down to 45W or lower - it's actually a great illustration that power and performance don't scale linearly.

Other part is that it also comes down to how long CB R23 runs then, and if they ran it multiple times.

Of note, Andrei of Anandtech, on Twitter, said he saw 15W of power draw on the M1 - but he wasn't sure if it boosts above that or not (and thus behaves like the CPUs above). He did see 21W in SPEC, so there is clearly headroom there.

is the 4800U actually locked at 15W? Usually, those chips can go higher than the TDP numbers for some amount of time.

It is set by the OEM (or by the user in the UEFI for those that have it unlocked).

I know TDP is confusing, especially since everyone has their own definition - Intel straight ignores it, AMD adheres mostly to it but lets you go above it (like the 65W TDP 5600X can go up to 88W of draw), whereas Apple is unknown (Anandtech showed 26W active power draw in compute MT thread, 21W in SPEC, and 15W in CB R23) so it's obvious Apple isn't strictly adhering to a 10W or 15W TDP either

Unfortunately, reviewers have to be VERY specific about it or else it's not easy to compare - which is why we're seeing so many different takes today, it seems.

Right, but even the 15W version actually runs a lot hotter than that,

I believe people are talking about it being able to boost higher than 15W

according to reports here (as does the M1 in the actively cooled housings), so throttling may remain a problem. In comparing the M1 Air to the M1 MBP, the throttling only showed up when R23 was on a loop, so it might not show for the 4800U either in the results you link to.

Yeah, the problem is that the few reviews we have today are either pretty amateur-level ones from tech tubers or only initial looks like from Anandtech where they are few on details.

I'm waiting for the notebookcheck one since they run these things in loops and do actual apples-to-apples comparisons

Either way, from the initial power numbers, Andrei@Anandtech saw SPEC was able to pull 21W from the CPU cores only, so the CPU is clearly not just a 10W or 15W chip.

 

[donth8]

Two things: one, it depends on if they lock it or not. 'Normal' operation allows it to boost above 15W, yes, but AMD is pretty liberal allowing people to set target power consumption. For instance, you can set the 65W 5600X (which can draw a max of 88W stock) to "just" 65W - you lose 2-5% in performance but consume 33% less power. You can even set it down to 45W or lower - it's actually a great illustration that power and performance don't scale linearly.

Other part is that it also comes down to how long CB R23 runs then, and if they ran it multiple times.

Of note, Andrei of Anandtech, on Twitter, said he saw 15W of power draw on the M1 - but he wasn't sure if it boosts above that or not (and thus behaves like the CPUs above). He did see 21W in SPEC, so there is clearly headroom there.


It is set by the OEM (or by the user in the UEFI for those that have it unlocked).

I know TDP is confusing, especially since everyone has their own definition - Intel straight ignores it, AMD adheres mostly to it but lets you go above it (like the 65W TDP 5600X can go up to 88W of draw), whereas Apple is unknown (Anandtech showed 26W active power draw in compute MT thread, 21W in SPEC, and 15W in CB R23) so it's obvious Apple isn't strictly adhering to a 10W or 15W TDP either

Unfortunately, reviewers have to be VERY specific about it or else it's not easy to compare - which is why we're seeing so many different takes today, it seems.


I believe people are talking about it being able to boost higher than 15W



Yeah, the problem is that the few reviews we have today are either pretty amateur-level ones from tech tubers or only initial looks like from Anandtech where they are few on details.

I'm waiting for the notebookcheck one since they run these things in loops and do actual apples-to-apples comparisons

Either way, from the initial power numbers, Andrei@Anandtech saw SPEC was able to pull 21W from the CPU cores only, so the CPU is clearly not just a 10W or 15W chip.

Do you have a source for the 2-5% less performance for a 33% power savings on the 5600x?

 

[Gnattu]

I'm waiting for the notebookcheck one since they run these things in loops and do actual apples-to-apples comparisons

apples-to-apples comparison is hard to do with Macs. The mostly argued offerings from AMD are not officially available under macOS, and M1 Macs cannot install Windows. Comparing with Intel is boring as M1 is better than everything Intel has to offer in the same form factor. By the way I personally do not like the idea of installing Windows and doing benchmarks for comparison as the numbers are only useful when using windows on a Mac.

 

[rayyu882]

Pretty satisfied so far.

MacBook Air M1 16GB / 512 GB.

 

[echodriver]

Do you have a source for the 2-5% less performance for a 33% power savings on the 5600x?

5600X was my own benchmark on my SFF with a power meter

Look up Eco mode for Ryzen processors. You can actually see this with the older Zen 2 CPUs too. Here is the 3900 (OEM non-X variant) down-rated to 65W:

AMD Ryzen 9 3900 Review: a Taste of Eco Mode

Pre-built systems just took a big step forward.

www.tomshardware.com

It draws a whopping 55W less than the 3900X (90 vs 145W) with only small single digit % drops in performance

Here's a video where Optimum Tech benches the processors in Eco Mode and shows difference in temps and performance:

Long story short: CPU performance doesn't scale linearly with power consumption. You can get massive reductions in power for minimal decreases in performance with CPUs - desktop CPUs are very inefficient because they're all about maximizing every shred of performance (likewise with GPUs... a RTX 3080 draws over 300W but isn't 15-30x faster than the M1 in gaming... but if you need 144 fps minimum on a AAA title on a 1440p screen, raw performance is the only answer).

I'll be curious to see how well M1 scales up for future ASi chips. Anandtech mentioned that DDR4 seemed saturated on this design, so I'm curious how they'll handle it with more cores and higher power usage

apples-to-apples comparison is hard to do with Macs. The mostly argued offerings from AMD are not officially available under macOS, and M1 Macs cannot install Windows. Comparing with Intel is boring as M1 is better than everything Intel has to offer in the same form factor. By the way I personally do not like the idea of installing Windows and doing benchmarks for comparison as the numbers are only useful when using windows on a Mac.

Well then a lot of that makes the point of saying "M1 is 3x faster than Intel offerings" pretty moot. And there are Ryzen hackintoshes, so it's not as hard to compare as you'd think.

Moreover, the point was methodology. If a single Cinebench run is all that is used, we won't get the same answer as if we did a 30 minute loop for every CPU tested. A 30 minute run is likely more representative of someone doing rendering or music production or code compiling than say a benchmark that runs for 20 seconds

 

[Gnattu]

5600X was my own benchmark on my SFF with a power meter

Look up Eco mode for Ryzen processors. You can actually see this with the older Zen 2 CPUs too. Here is the 3900 (OEM non-X variant) down-rated to 65W:

AMD Ryzen 9 3900 Review: a Taste of Eco Mode

Pre-built systems just took a big step forward.

www.tomshardware.com

It draws a whopping 55W less than the 3900X (90 vs 145W) with only small single digit % drops in performance

Here's a video where Optimum Tech benches the processors in Eco Mode and shows difference in temps and performance:

The echo mode on 65w parts decrease the TDP to 35w level if I remembered it correctly.

I'm an owner of 3900x and i can say that the Cinebench on eco mode is not only single digit percentage drop, it's more than 10% at around 13% level.

AMD is pushing the core clock near the limit of the node for Ryzen Desktop parts, that's why it has a such deep frequency vs power curve. On server EPYC parts, the curve is not as deep because of lower clocks.

 

[Gnattu]

Well then a lot of that makes the point of saying "M1 is 3x faster than Intel offerings" pretty moot. And there are Ryzen hackintoshes, so it's not as hard to compare as you'd think.

Moreover, the point was methodology. If a single Cinebench run is all that is used, we won't get the same answer as if we did a 30 minute loop for every CPU tested. A 30 minute run is likely more representative of someone doing rendering or music production or code compiling than say a benchmark that runs for 20 seconds

Apple has comments on their site saying at what circumstance M1 is 3x faster. Like Xcode compiling comparing to an i3-8100B, M1 is 3 times faster(It is easy to verify, people having 10-core iMac Pro claim M1 is faster, people having 12-core iMac Pro claim M1 is on par). You can still argue that compiling runs only for a couple of seconds, but I don't personally think Mac Mini has Thermal problems for M1, this is more likely a problem for MacBook Air.

For Cinebench at least, 30 minute loop does not affect the results of MBP and Mac Mini, the Air ends up with ~17% slower. I would say that is impressive given the Air is a fan-less device.

Comparing hackintosh adds extra problems on the table like how do you know you AMD cpu performance is not impacted(negatively)? BTW I don't know if there is any APU powered laptops been hackintosh-able, if you comparing it to desktop then there will be again "apples-oranges" problems.

 

[onfire23]

Two things: one, it depends on if they lock it or not. 'Normal' operation allows it to boost above 15W, yes, but AMD is pretty liberal allowing people to set target power consumption. For instance, you can set the 65W 5600X (which can draw a max of 88W stock) to "just" 65W - you lose 2-5% in performance but consume 33% less power. You can even set it down to 45W or lower - it's actually a great illustration that power and performance don't scale linearly.

Other part is that it also comes down to how long CB R23 runs then, and if they ran it multiple times.

Of note, Andrei of Anandtech, on Twitter, said he saw 15W of power draw on the M1 - but he wasn't sure if it boosts above that or not (and thus behaves like the CPUs above). He did see 21W in SPEC, so there is clearly headroom there.


It is set by the OEM (or by the user in the UEFI for those that have it unlocked).

I know TDP is confusing, especially since everyone has their own definition - Intel straight ignores it, AMD adheres mostly to it but lets you go above it (like the 65W TDP 5600X can go up to 88W of draw), whereas Apple is unknown (Anandtech showed 26W active power draw in compute MT thread, 21W in SPEC, and 15W in CB R23) so it's obvious Apple isn't strictly adhering to a 10W or 15W TDP either

Unfortunately, reviewers have to be VERY specific about it or else it's not easy to compare - which is why we're seeing so many different takes today, it seems.


I believe people are talking about it being able to boost higher than 15W



Yeah, the problem is that the few reviews we have today are either pretty amateur-level ones from tech tubers or only initial looks like from Anandtech where they are few on details.

I'm waiting for the notebookcheck one since they run these things in loops and do actual apples-to-apples comparisons

Either way, from the initial power numbers, Andrei@Anandtech saw SPEC was able to pull 21W from the CPU cores only, so the CPU is clearly not just a 10W or 15W chip.

My laptop's 4700u can boost to 30+watts package power all day long. It is definitely more than a few minutes.

In the battery saver it is locked to 15 watts and I get 4800 MC in CB23 compared to 7000 MC in performance mode.

 

[Sanpete]

My laptop's 4700u can boost to 30+watts package power all day long. It is definitely more than a few minutes.

In the battery saver it is locked to 15 watts and I get 4800 MC in CB23 compared to 7000 MC in performance mode.

Getting past 7000 may be where the trouble is ...

There will probably be some more comparisons that would invite throttling, so we can see how it works.

 

[M1 Processor]

That could be, except when throttling is considered. @nikidimi says above that the 4800U loses 25% performance when it throttles, putting it on a par for multi-core with the actively cooled M1. That's a big advantage of a lower TDP. (One reason I asked about a laptop with the chip rather than just the chip earlier is that in a particular laptop it has to be effectively cooled, etc.)

All chips throttle, I think the point of the discussion between you and I was I was saying these ”Latest PC Laptop chips” Apple was referring to are not from AMD. The M1 at 10w isn’t outperforming a 40w Renoir.

 

[nikidimi]

Unfortunately, reviewers have to be VERY specific about it or else it's not easy to compare - which is why we're seeing so many different takes today, it seems.

Yes, we are making a lot of assumptions here. I think that 4800U at 25W is capable of boosting to ~35W and that's when it reaches 10000. If it's set at 15W, it's going to boost to only around ~25W and then it's reaching 7500 (or if set to 25W and it's throttling down). The M1 reports around 15W as @Gnattu posted. So the M1 is the fastest CPU below like ~20W and it's capable of achieving the same performance with 25/15W = 40% less energy, which is quite impressive.

 

[Sanpete]

All chips throttle, I think the point of the discussion between you and I was I was saying these ”Latest PC Laptop chips” Apple was referring to are not from AMD. The M1 at 10w isn’t outperforming a 40w Renoir.

Apple doesn't claim to outperform the peak performance of the mystery chip, only match it. If the AMD throttles during testing, that may be why the cooler M1 matches it, as explained above. The figures do fit that scenario.

 

[Nikolaosth]

It is not, see this chart from another Anandtech article:
View attachment 1671246

andrei from anandtech says the m1 goes up to 32 watts when loaded so not much different than both tiger lake and mobile zen

32watt tdp andrei on Twitter

 

[Kung gu]

andrei from anandtech says the m1 goes up to 32 watts when loaded so not much different than both tiger lake and mobile zen

32watt tdp andrei on Twitter

keep in mind that andrei has mac mini m1, all m1 have different TDPs

 

[Gnattu]

andrei from anandtech says the m1 goes up to 32 watts when loaded so not much different than both tiger lake and mobile zen

32watt tdp andrei on Twitter

It is CPU + GPU combined, and you cannot compare it to numbers with cpu alone

 

[Zackmd1]

Not really a benchmark but my MBA 8/512 config can run Fortnite at native resolution (70% render resolution) with a mixture of medium/high settings and achieve a locked 60FPS.

That is damn impressive considering other iGPUs struggle to hit 30fps in this game at the lowest possible resolution and lowest settings. And this is running through Rosetta 2...

 

[M1 Processor]

Apple doesn't claim to outperform the peak performance of the mystery chip, only match it. If the AMD throttles during testing, that may be why the cooler M1 matches it, as explained above. The figures do fit that scenario.

M1 delivers significantly higher performance at every power level when compared with the very latest PC laptop chip. At just 10 watts (the thermal envelope of a MacBook Air), M1 delivers up to 2x the CPU performance of the PC chip. And M1 can match the peak performance of the PC chip while using just a quarter of the power.

When they claim to match the peak performance, the were talking at less than 10w per their chart.