|
Projects Publications Resume Contact About Youtube LinkedIn |
|
I was curious about Apple M-series ARM processors after using my 2019 Intel iMac for over 5 years and after trying a Snapdragon X Elite ARM Windows laptop. I picked up a couple Mac mini M4's and I compared the performance of Speedometer 3.0 on various machines I had access to. I also compared virtualization performance for a couple scenarios and the results shocked me. This testing favors single-core performance and the benefits of multiple cores are not evident here. This testing just shows how snappy and responsive the various machines feel in basic lightweight tasks like web browsing.
| year | make | model | cpu | arch | host/hypervisor | cores | ram (GB) | os | browser | speedometer 3.0 |
| 2019 | Apple | iMac 27" | Intel Core i9 (9th Gen) | x64 | host | 8 | 128 | macOS | Safari | 21.2 |
| 2019 | Apple | iMac 27" | Intel Core i9 (9th Gen) | x64 | host | 8 | 128 | macOS | Firefox | 17.2 |
| 2024 | Apple | Mac mini | M4 | arm64 | host | 10 | 24 | macOS | Safari | 46.0 |
| 2024 | Apple | Mac mini | M4 Pro | arm64 | host | 12 | 24 | macOS | Safari | 45.0 |
| 2024 | Apple | Mac mini | M4 Pro | arm64 | host | 12 | 24 | macOS | Firefox | 39.8 |
| 2024 | Apple | Mac mini | M4 Pro | arm64 | UTM | 4 | 8 | macOS | Safari | 39.9 |
| 2024 | Apple | Mac mini | M4 Pro | arm64 | UTM | 4 | 8 | Windows 11 ARM | Edge | 21.1 |
| 2024 | Apple | Mac mini | M4 Pro | arm64 | VMware Fusion | 4 | 8 | Windows 11 ARM | Edge | 30.7 |
| 2024 | Lenovo | Yoga Slim 7x | Snapdragon X Elite | arm64 | host | 12 | 32 | Windows 11 ARM | Edge | 25.0 |
| 2024 | Lenovo | Yoga 9i 2-in-1 | Intel Core i7 (13th Gen) | x64 | host | 12 | 16 | Windows 11 | Edge | 23.3 |
| 2024 | Apple | MacBook Pro | M4 Max | arm64 | host | 16 | 128 | macOS | Safari | 49.0 |
My trusty baseline reference (fully-loaded 2019 27" Intel iMac) was not exactly a slouch, but these numbers opened my eyes quite a bit. Because this test favors single-core performance, both the M4 and M4 Pro performed about the same as each other within measurement error. They both achieve over double the performance of my baseline.
After getting over the shock of the raw M4 performance, what really surprised me next was the virtualization testing. Running Windows 11 ARM in VMware Fusion on M4 Pro performs better than running Windows 11 ARM natively on Snapdragon X Elite. I'm still laughing as I write this and I'm sure other tests would be a different story. And running macOS in UTM on M4 Pro outperforms that. The virtualized macOS doesn't give up much performance compared to native. Native M4 performance and some virtualization cases handily beat the Intel Core i7 13th Gen. The Snapdragon also outperformed the similarly priced Intel laptop.
After doing this browser testing, I did some 4K and 1080p H.264 and HEVC video transcoding tests on the Mac minis. Both Mac minis achieved roughly 4x the transcoding frame rate as my iMac with Radeon Pro Vega 48 graphics. The Windows laptop hardware encoders wouldn't do HEVC 10-bit, so I did not fully compare those.
I also compared some tasks in Ubuntu Server, which seemed blazing fast in both UTM and Fusion. Both outperformed WSL on ARM and Intel. I didn't capture lots of careful measurements for that because the trend echoed the Speedometer results.
I'm sure other types of testing will either level the paying field or show other clear winners. I have seen some other types of tests where the multi-core performance of the Snapdragon X Elite does very well, but that isn't the focus of the testing I'm doing here.
I just heard about some benchmarks of the upcoming Snapdragon X2 Elite that look very impressive. Those Windows ARM laptops might be able to give the M4 and M5 some real competition.