All the benchmarks to outperform Microsoft’s MacBook Air

Microsoft is on a mission to take down the MacBook Air with its new range of Copilot Plus computers. She’s so confident that she’s finally got Windows right that she spent an entire day pitting her new Surface laptop against a MacBook Air at its headquarters in Redmond, Wash., last month. the edge Multiple benchmarks and real-world simulation tests were demonstrated to prove that the new Qualcomm-powered Surface Laptop outperforms Apple’s best-selling laptop.

Although I’ve covered Microsoft’s confidence in beating the Apple M3 processor previously, I thought it would be helpful to examine all the benchmark claims and battery life estimates in detail. Microsoft touched on some of these during its Surface and Windows AI event last week, but the claims on stage weren’t always as detailed as Microsoft employees showed me last month.

I wasn’t able to run the benchmarks myself, but the results should serve as an important data point as we get closer to the launch of these Copilot Plus PCs on June 18. It’s also important to note that, unlike Apple’s MacBook Air, Microsoft’s new Surface Laptop has no fan, allowing it to get more performance. Microsoft only compared the Surface Laptop to the MacBook Air M3 — not the MacBook Pro, which comes with fans.

Either way, benchmarks aren’t everything, and we’ll get a better idea of ​​real-world performance and battery life when we review the Surface Laptop next month.

Microsoft kicked off the benchmarks day by first measuring raw performance between the Surface Laptop and the MacBook Air M3. It showed two benchmark claims, featuring sustained performance using the Cinebench 2024 multithreaded workload and peak performance using the Geekbench 6 multithreaded test.

The Surface Laptop achieved a score of 980 on the Cinebench 2024 multithreaded benchmark and a score of 14,000 on the Geekbench 6 multithreaded benchmark. Microsoft avoided highlighting individual results for both benchmarks, perhaps because the MacBook Air would fare slightly better here.

Either way, Microsoft claims the new Surface Laptop will outperform the MacBook Air M3 in multi-threaded workloads in Cinebench by 50 percent. In Geekbench 6, the Surface Laptop is only 16 percent better. On stage last week, Microsoft also claimed that its range of Copilot Plus PCs would be “58 percent faster than the MacBook Air M3.”

Next, Microsoft covered what it called “real performance.” The main test here was the HandBrake ToS test, which measures how long it takes to encode a 4K video file. The Surface Laptop with the Snapdragon

More importantly, this was twice as fast as the Surface Laptop 5 using a 12th Gen Intel Alder Lake CPU, which took 10 minutes and 30 seconds to complete the task. The Surface Laptop 4 took longer, at 13 minutes and 32 seconds.

Microsoft’s comparisons to the MacBook Air M3 also extend to battery life. During testing, I saw Microsoft simulating battery life across web browsing and video playback. Microsoft uses a script to simulate web browsing. On the 2022 Intel-based Surface Laptop 5, it took eight hours and 38 minutes to fully drain the battery; The new Surface Laptop lasted twice as long, to the tune of 16 hours and 56 minutes. This beats the same test on the 15-inch MacBook Air M3, which lasted 15 hours and 25 minutes.

Microsoft ran a similar video playback test, with the Surface Laptop lasting more than 20 hours, while the MacBook Air M3 hit 17 hours and 45 minutes. That’s also about eight hours more than the Surface Laptop 5, which lasted 12 hours and 30 minutes.

Microsoft claimed on stage last week that new Copilot Plus computers with Qualcomm’s Snapdragon

The final benchmarks Microsoft showed me were related to NPU performance. Microsoft claims that the NPU inside the Snapdragon

The Surface Laptop scored 1,745 on the Procyon AI Score, while the MacBook Air scored 889. Qualcomm’s Snapdragon

Microsoft also showed that the Surface Laptop achieves 4.5x the inference efficiency of the Phi Silica model for fast processing over the M3, along with 24 TOPS/watt of peak inference efficiency.