Hyperbole and irrational fanboy flexing aside, Apple’s M1 is indeed a powerful chip. But to get a better feel against its contemporaries we sat down and retested a pile of current Windows 10 laptops to get an idea where the M1 MacBook Air 13 lands in the pecking order of laptops.
While we don’t have direct access to an MacBook Pro M1, our sister publication, Macworld does, so we cribbed from its glowing review of the 13-inch MacBook Pro M1. Macworld also graciously ran a couple of additional tests for our analysis. For two other results, we relied on published results from Puget Systems, a bespoke system builder that manufactures workstations and rolls its own productivity-focused benchmarks.
Our tests show the M1 MacBook Air 13 performing very well compared to equivalent PC laptops. You can jump straight to our conclusion for the final analysis, but if you want to dig deeper to see exactly where and how the M1 does well, read on…
The laptops we tested
Apple’s MacBook Pro M1 features its spiffy new M1 Arm-based SoC/CPU, a 512GB custom SSD, 16GB of LPDDR4X/4267 memory, a 13.3-inch 2560×1600 screen, and lap weight of 3.1 lbs.
For our PC comparisons, we decided to pick from laptops that are similar in size and weight:
MSI’s older Prestige 14 is equipped with a six-core, 10th-gen Core i7-10710U, GeForce GTX 1650 Max-Q graphics, 16GB of LPDDR3/2133 memory, a 1TB PCIe 3.0 SSD, and a 14-inch 4K screen. It weighs 2.8 lbs.
Lenovo’s Yoga Slim 7 has an eight-core Ryzen 4800U with Radeon graphics, 16GB of LPDDR4X/4267 memory, a 512GB PCIe 3.0 SSD, a 14-inch FHD, and lap weight of 3.1 lbs.
The Asus ROG Zephyrus G14 includes an eight-core Ryzen 9 4800HS, GeForce RTX 2060 Max-Q graphics, 16GB of DDR4/3200 memory, a 1TB PCIe 3.0 SSD, a 14-inch FHD screen, and a weight of 3.6 lbs.
Before you object to having the Zephyrus G14 laptop here, pay close attention to the weight and size of the laptop. At 3.6 lbs., it’s actually very close to the weight of the Lenovo Yoga Slim 7 and the Apple MacBook Pro 13, which are each 3.1 lbs. We think there are indeed some people who might consider the Zephyrus G14 to get the extra power its GeForce RTX 2060 Max-Q provides. Yes, the 180 watt power brick adds even more weight to the G14, but it may just be worth it to those folks.
We think discrete graphics are one of the overlooked features of the older MSI Prestige 14 too—a laptop that weighs less than the Lenovo Slim 7 and MacBook Pro M1, but features a GeForce GTX 1650 Max-Q GPU inside. There are indeed several compromises in the older Prestige 14 to get that GPU, which we detailed in our original review, but weight is typically the great equalizer.
That’s why the final laptop we included is really there for a raw performance comparison, since no one would consider it remotely in the class of the laptops above:
The Acer Predator Triton 500 with a six-core 10th-gen Core i7-10750H CPU, GeForce RTX 2080 Super graphics, 32GB of DDR4/3200 memory, a 1TB PCIe 3.0 SSD, and a 15.6-inch 300Hz FHD screen. It weighs 4.6 lbs.
We wanted this laptop particularly for its Core i7-10750H CPU, which is very similar to Intel’s 8th-gen and 9th-gen “H” class CPUs found larger laptops such as Apple’s MacBook Pro 16. Most of the laptops here feature lower-power CPUs, so we wanted to see just how well the M1 and other chips in smaller laptops compared against a chip that sucks down many more watts.
Of all of the chips here, Apple’s M1 uses TSMC’s most advanced 5nm process, with the two Ryzen laptops using TSMC’s 7nm. Intel’s 11th-gen “Tiger Lake” Core i7-1185G7 is on Intel’s newest 10nm process, while the Core i7-10710U is like the Core i7-10750H and built on Intel’s—how can we say this politely?—wise 14nm process.
Next page: Performance benchmarks
Apple M1 vs. Intel 11th-gen Core vs. AMD Ryzen 4000 performance benchmarks
Cinebench R20 performance
We’ll kick off our results with Maxon’s older Cinebench R20. It’s a 3D modelling benchmark built on the company’s in-house engine used in its commercial Cinema4D product. Cinebench R20 had versions for x86 on Windows and x86 on MacOS. Maxon’s newer Cinema R23 offers native support for Apple’s M1 chip, but the older R20 version must use Apple’s Rosetta 2, a technology that handles just-in-time translation of x86 instructions to Arm from non-native code.
Having to pay a real-time translation penalty typically blows chunks so we expected the M1 to cough up furballs, but it’s well-known now that Apple’s unlimited funding and hard work has paid off handsomely. Yes, you can look at the black bar in the chart and see that the red Ryzen 4000 chipsstomps the M1 into the ground. And yes, the quad-core Core i7-1185G7 is faster too, despite the Mac have eight physical cores, but remember the translation penalty the MacBook is paying and how much it saps performance. This is an impressive showing by Apple.
Cinebench allows you to measure a single-threaded performance as well. For the x86 side, where Ryzen previously beat Core, it’s now flipped as we see Intel’s new 11th-gen Core i7-1185G7 leading the way. The Apple MacBook Pro M1? It’s really not bad again when you consider that Apple is paying a hefty translation penalty. Some believe it to be as high as 30 percent, which coincidentally is matches the fee Apple takes from App Store purchases, sparking its war with Epic. We don’t know if the two are related, but we’d guess the M1 picks up a lot more mileage with native code.
Cinebench R23 Performance
Maxon released a new version of Cinebench R23 with native M1 support, but there are some other key changes which should be mentioned as well. With Cinebench R20, the benchmark would run render a single scene and produce a score based on its completion. With Cinebench R23, the render scene is the same, but Maxon has interestingly changed it to render the same scene over and over for 10 minutes. The benchmark will even run beyond 10 minutes if the scene is still rendering when the timer hits zero.
On a six-core Intel H-class chip, Cinebench R15 takes 34 seconds to complete while Cinebench R20 takes about 108 seconds. With Cinebench R23 it now takes a minimum of 600 seconds to run under the new method. On a desktop or workstation with far more cooling it’s not an issue, but on laptops an all-core test that runs for at least 10 minutes can be far harsher—especially on CPUs that either make more heat, or laptops with more limited cooling. It’s actually called a “throttle test” which is a different method than before for Cinebench.
None of this bothers Apple’s M1 much though. Based on TSMC’s most advanced 5nm process, it’s a stone cold killer, with Macworld reporting no fan noise at all during the run. That can’t be said of the x86 laptops, which all vary from fairly quiet to a little rackety.
The performance is impressive though, with the eight-core M1 Mac now ahead of the four-core 11th-gen Tiger Lake as well as the older six-core Core i7-10710U. But if you give Intel’s older six-core more thermal head room it’s almost dead even with the silent M1. An eight-core variant of an Intel H-chip would be even faster obviously.
AMD, however, just shrugs at all this and yawns as its eight-core Ryzen 4000 chips easily beat all comers when set to their performance modes.
Single-threaded performance probably matters more for what most people do and the M1 doesn’t disappoint there either. Neither does Intel’s Core i7-1185G7, to be fair. Its single-threaded prowess has pushed aside Ryzen 4000 and you see that here where both the M1 and the Core i7-1185G7 are basically tied in Cinebench R23. There’s some conjecture that “we’re just not testing x86 right” as Usman Pirzada argues at WCCFtech. However, Joel Hruska at Extremetech (who worked on the results with Usman) argues the ball is really back in Intel and AMD’s court for how to address the limitations.
The main take away is the single-threaded performance on the M1 as well as Intel’s Tiger Lake is nothing to dismiss.
But as we said, there is indeed a cost to be had for when you run every CPU core that hard on a laptop. Cinebench R23 allows you to actually turn off the “Throttle Test,” so we did that to record scores of Cinebench R23’s multi-core benchmark in a more traditional single render scene. Laptops that can benefit from only running full tilt for three minutes versus more than 10 minutes get a decent 6 to 8 percent boost. The Ryzen 4000 laptops actually open up their performance gap against The M1 and Intel chips even further when the run time is reduced.
This just goes to show you the boosty nature of laptops improves from lowering the thermal load, putting in more fans to dissipate that thermal load, or simply making your CPU more efficient.
You can see that in the next chart, where we took our Cinebench R23 single-run result and compared it to the result of Cinebench R23 in its default throttle test. The laptops that face more of a thermal limit all show 6 to 8 percent improvement when the load is cut by two-thirds, except for two: The Apple MacBook Pro M1 and the Acer Predator Triton 500.
In the Predator Triton 500’s case, its larger chassis, additional heat pipes and additional venting means the CPU is just never generating enough heat to hit that thermal wall, even with a fairly high-wattage Intel chip inside. Its six cores of 14nm oldness still can’t outpace the other chips, but the thermals clearly have no issues. That shouldn’t surprise, as the Triton 500’s cooling was designed to keep the GeForce RTX 2080 Super Max-Q running smooth as well, so running a pure CPU load means the cooling headroom is far higher.
The MacBook Pro M1 is equally impressive for a different reason. Despite being a thin and light laptop, it just doesn’t seem to generate enough heat to hurt it. You can run Cinebench R23 for 3 minutes or 10 minutes—it just doesn’t care. That’s something that should be appreciated.
And yes, the older Prestige 14 with its 14nm Comet Lake U chip on its default setting basically doesn’t move much because on default, it’s probably so lackadaisical it doesn’t matter anyway. It also benefits from cutting the render time down on its performance setting though.
V-Ray Next Performance
Up next is Chaos Groups’ V-Ray Next benchmark, another 3D rendering test. Like Cinebench R20, it’s non-native to the new MacBook and loves more cores. With the M1’s translation tax, V-Ray Next loses some performance, but it’s still enough to to just about equal Intel’s four-core Tiger Lake part. AMD and its eight-core Ryzen CPUs again gives all others a good Nelson Muntz laugh.
V-Ray allows you to also run a render on the GPU instead, which is increasingly popular among advanced engines. Of the chips with onboard graphics, the Ryzen 7’s Radeon GPU cores put AMD in front, Apple’s M1 falls in the middle, and Intel lags a bit behind.
GPU rendering performance, obviously, gets better with more powerful GPUs. The GeForce RTX 2080 Super Max-Q in the Triton 500 thus wipes the floor with everyone else, while the GeForce RTX 2060 Max-Q doesn’t do too shabby either. This, frankly, is one of the reasons we think the Asus G14 is still so compelling. Its 0.5-lbs heavier than the laptops with onboard graphics but that GeForce RTX 2060 Max-Q makes a yuge difference in performance.
Rise of the Tomb Raider Performance
We used just one game to look at the graphics performance of the M1 compared to Windows laptops. We ran Rise of the Tomb Raider set to 1920×1080 on the Windows laptops on high and compared it to the result Macworld got on the MacBook Pro M1 running at 1920×1200 on High.
While the integrated graphics in Intel 11th-gen laptop is in front of AMD and Apple’s onboard GPUs, let’s just call it a tie and agree all three would be capable of moderate gaming at lower resolutions. Less intense games would be easier of course. None of them would compare to a laptop with a discrete GPU though. As you can see, the GeForce GTX 1650 Max-Q walks away from the integrated graphics laptops, and we’ll remind you again the Prestige 14 with the GTX 1650 inside weighs less than two of the other laptops.
Of course, we can’t ignore that Asus Zephyrus G14 nor the Predator Triton 500 either. If you really intend to do anything graphically intense, get a laptop with discrete graphics in it.
Still, the three integrated graphics laptops do fairly well. We wouldn’t even bother trying to compare to laptops with Intel’s UHD graphics. Focusing on the MacBook M1, it again offers very impressive performance. But yes, if someone is going to squawk at you about its graphics performance, pretty much any Windows laptop with half-decent discrete graphics will eat its lunch, and for less money. We’re pretty certain that a budget Black Friday gaming laptop sold for $700 is likely to whoop the MacBook Pro M1 in games thanks to its discrete graphics chip. And no, we can’t run Counter Strike: Global Operations or Red Dead Redemption II on the new MacBook Pro M1. You know why so don’t ask.
PugetBench Photoshop Performance
All this talk of 3D rendering, single-threaded performance and blah, blah tests probably sound like what your dog hears when you scold him for pooping on the new carpet. In real life, people use these things called applications. Among the most popular is Adobe’s Photoshop.
To test its performance, we used Puget Systems’ PugetBench, which is where we also sourced the M1 performance results. Puget Systems looked at the M1’s performance to help people have a clearer understanding of where the new MacBook Pro M1 stands in relation to desktop performance (which will always be faster). Despite the flexing from Mac super fans, even an old desktop PC is faster than the MacBook Pro M1, as Puget Systems found.
Comparing it against contemporary laptops, though, shows the M1 does quite well with performance, surpassing the Lenovo Yoga Slim 7 and its Ryzen 7 4800U chip. It’s about tied with the Ryzen 9 4900HS in the Zephyrus 14 too. Photoshop has traditionally been very single-threaded, so we’re not too surprised to best honest. Single-threaded CPU performance never really was Ryzen 4000’s strength. Intel’s 11th-gen Tiger Lake chip is in front of the MacBook Pro M1 by a small amount.
Also of interest are the results from two older 14nm-based Intel laptops. The Prestige 14 with its GeForce GTX 1650 Max-Q and very boost-limited high clocks doesn’t quite pay off the dividends you’d expect, so we really think the test makes far sparser use of the discrete GPU than you might think.
What doesn’t disappoint though is that relatively large Acer Predator Triton 500. It’s the overall winner. We attribute that to its far better cooling, which enables higher-clock speeds; possibly its GeForce RTX 2080 Super Max-Q GPU; and, crucially, its 32GB of base memory. Intense Photoshop users should always opt for as much memory as you can pack into a laptop.
PugetBench Premiere Performance
Moving on to something far stiffer, we also used PugetBench Premiere on the laptops. Since the test takes over an hour and requires as much performance as you can get, we’ve decided to limit it only to the laptops in their respective performance modes. Against laptops with integrated graphics—the Ryzen 4000 Slim 7 and the 11th-gen Prestige 14 Evo—it’s a win for the MacBook Pro M1. More so against the Slim 7, which falls behind further than expected.
Laptops with discrete graphics fare better, with the GeForce GTX 1650 Max-Q in the older MSI Prestige 14 pulling even with Apple’s M1. The GeForce RTX 2080 Max-Q in the Acer Predator Triton 500 has no problems leaving the smaller laptops far behind though. The compact Asus ROG Zephyrus G14 with its 8-core Ryzen 9 and GeForce RTX 2060 Max-Q which dusts off all comers.
This probably indicates that GPU encoding capabilities matter for how PugetBench sets its test up. When balanced with a multi-core chip, it’s a winning combination overall. Still, it’s a good showing for the M1 Mac, but realistically, it’s probably going to be slower than a budget gaming laptop in many Premiere tasks.
The big question is how much of a boost the M1 gets once Adobe offers native support in its apps. Since Adobe’s products are so vast and what people do are so varied, it’s pretty hard to say if the native performance boost will be as large as we see in Cinebench. There are just so many things being touched. We can say there will indeed a performance bump—but nobody knows how big it will be.
Topaz Gigapixel AI Performance
One more chore we wanted to throw at the laptops uses advanced AI features. We find that in Topaz Lab’s Gigapixel AI application, which uses machine learning models to upsample images in a method that can be far more effective than traditional enlargement techniques.
For our test, we took an image of a US Air Force F-22 Raptor shot in 2010 on an 8.2MP Canon EOS 1D Mk IIn DSLR. We feed the image into Gigapixel AI and task it with increasing the resolution by 6x. The application uses Intel’s OpenVINO framework and is quite impressive—on Intel hardware. Since Macs up until now have been based on Intel, that’s been a plus for the many Topaz Labs customers.
Macworld ran the same build of the app on the MacBook Pro M1 with the same settings that we used on our Windows 10 machines. If you read our 11th-gen Tiger Lake preview, you know that Intel’s new-look CPU rocks the Ryzen 4000 chip. It’s fair to say that it’s a slaughter for the Slim 7 laptop with its Ryzen 7 4800U. The 11th-gen Tiger Lake chip also easily runs its 14nm sibling in older Prestige 14 off the field.
And yes, the MacBook Pro M1 performs the worst of all of the laptops here. We’re talking about 1 minute for the 11th-gen Intel chip in the MSI Prestige 14 Evo versus just under 5 minutes for the MacBook Pro M1. That’s for just a single image, too.
Gigapixel AI also supports discrete graphics, which actually helps the Zephyrus 14 and Predator Triton 500 stay in the game. But even with their much faster discrete GPUs they’re still significantly slower than the 11th Tiger Lake chip.
In the past, AMD would throw it back in our face by saying that such specialty apps are used by far, far fewer people than even 3D modelling or a video encoder. We suspect Apple fans would say the same thing—and imply that once Gigapixel AI is updated to support the MacBook M1 and its own inference engines, the M1 will far better.
In a perfect world, yes, but we really don’t know if or how long that will take. We should also point out this is a real application being used by an untold number of photographers and digital artists.
Apple M1 Rosetta 2 cost
There have been plenty measurements of the translation cost needed to make x86 apps run on the MacBook M1, but we wanted to take yet another stab it. For that we decided to use Cinebench R23 and Cinebench R20. As we said earlier, R20 is native to Intel-based Macs, while Cinebench R23 is not. Cinebench R20 bases its score on a single render, while Cinebench R32 bases it on multiple renders by default. To see if they compare, we timed how long it takes to run a single render in both versions. On the Windows version, it’s within the margin of stop watch error on the MSI Prestige 14 Evo. That’s not the case with the M1.
Keep in mind when you look at the chart below that the shorter bar indicates higher performance as we’re looking at how long it takes to render the scene.
Since both versions render what appears to be exact same frame, we can see the rendering time for both multi-threaded and single-threaded performance is about 28 percent less going from the translated version to the native version of Cinebench on Apple’s MacBook M1. That’s really not bad, and in line with what others have seen the cost of the Rosetta 2 to be.
What’s a little strange is there’s another way to run Cinebench R23 that forces the M1 to run it translated: Order the universal binary to install the Intel version of it, rather than the M1-native version. Macworld again graciously agreed to run it this way and the result was significantly worse.
Here’s the default 10-minute run of Cinebench R23 on the MacBook Pro M1 using the native-binary and the Intel-binary. We’re looking at closer to a 50 percent hit from using the Rosetta 2 translator to make the x86-code run on the Arm-based M1 chip. That’s not pretty.
The big takeaway? Software that gets converted to native code on the M1 should almost always pick up very large gains in performance. And for applications that never get converted to native code, you’ll always be paying a very hefty performance tax.
That’s bad, but you need to keep it in context. Overall, Rosetta 2 and the M1 is still largely a success for those who had performance concerns. And if you’re wondering why people had concerns over performance, you can see that below, where we take Cinebench R20 (non-native on the MacBook Pro’s M1) and dial in a score generated on Microsofts Surface Pro X tablet.
The Surface Pro X tablet is unique among PCs because it’s based on Qualcomm’s custom SQ1 chip built for Microsoft, running on Arm architecture rather than traditional x86 cores. This week Microsoft released its beta 64-bit support and let’s just say it doesn’t perform nearly as well as Rosetta 2 does on Apple’s M1 chip. Our Surface Pro X versus MacBook M1 comparison contains much deeper information, but Arm on Windows has never looked worse.
In today’s technology circles, many people have retreated into their corners, with irrational fans choosing to always cheer for their team and give no quarter to competitors—even if the other team won. That shouldn’t be any rational person’s perspective because ultimately, these are just companies looking to take your money. It’s also not classy.
With that said, we think Apple should be applauded for what it has pulled off with ithe new M1 chip. It offers truly amazing performance that looks like it can carry the Mac down the fork of the road away from the world of x86. You can arguably say Macs built on the new M1 will be worthy of the heritage of the Macintosh name, which hasn’t been the case over the last few years. With the M1 and Apple’s future iterations, heads can be held high again.
Yes, there will be faster chips from Apple. And we can tell you there will be faster chips from AMD, Intel, and Nvidia too. There might even be faster chips from Qualcomm.