Now that Microsoft’s Windows on ARM emulation layer can also run 64-bit x86 code, inevitably there will be questions as to how it compares with the Apple M1. Qualcomm’s 8cx platform and its minimal refresh earlier this fall have never been known for speed — these chips have been sold on the basis of excellent battery life, not pure performance.
Even with that said, the M1 eats Windows on ARM-powered laptops for breakfast. That’s the conclusion PCWorld reached after comparing the Snapdragon 8cx against the M1, with a Core i5 (4C/8T, 1GHz base, 3.6GHz boost) machine tossed in for good measure. Moving to 64-bit apps on Windows doesn’t seem to do anything for the 8cx’s overall performance.
It must be said that this isn’t particularly surprising, but it highlights the critical weakness in the nascent Windows on ARM ecosystem. Qualcomm essentially punted on updating its SoC this year, and while there technically was a refresh for devices like the Surface Pro X, the performance is practically identical. There’s been no architectural update.
Early benchmarks must, as always, be taken with grains of salt, but the Surface Pro X is a pretty known quantity at this point, and its performance is already known to be inferior to x86. And one of the problems Windows on ARM is struggling with, despite having existed for several years at this point, is that it has fewer native applications. Cinebench R23, for example, has Apple M1 native code paths, but nothing equivalent for the Surface Pro X. That means we’re effectively comparing native performance on M1 with emulated x86 performance via Windows on ARM.
Both the Core i5 and the SQ1 lose badly here, but the Core i5 manages 39 percent of the M1’s performance in multi-threaded, while the SQ1 manages ~23 percent of the M1 in both. Things don’t improve in Handbrake:
There are results for the M1 in Handbrake 1.4.0 using videotoolbox as well as H.265. The x86 systems were tested with the latest version for the platform, 1.3.3. The M1 is 2.87x faster than the Intel Core i5, but the i5 is more than 2x faster than the SQ1.
I’ve said before that Windows on ARM might become more important if ARM chips prove to have enduring advantages over x86, but these results show just how cavernous the gap is. It’s true that the SQ1 is running emulated code here, while the M1 isn’t, but it barely matters. If you eyeball the SQ1 and mentally improve its scores by 1.5x, it’s still nowhere near the performance of M1 or x86.
In short, Microsoft needs an ARM hardware developer that’s willing to invest in improving its platform on a regular cadence with real performance advances. Besides Qualcomm, the top three potential players would be Nvidia, AMD, and Samsung. Samsung shut down its homegrown CPU effort, so they seem out of the running. AMD has built ARM chips relatively recently, but it’s shown no interest in pulling focus away from Ryzen and competing against its own x86 CPUs. That leaves Qualcomm and Nvidia. Qualcomm could launch a true successor to the 8cx this coming year, and probably boost single-thread performance by 1.4x or more, since the CPU would be at least two generations more advanced.
Right now, Qualcomm seems like the best option for a competitive product, but I’m not convinced we should count out the idea of a future Windows on ARM system powered by Nvidia hardware. I’ve got absolutely zero inside information backing that up, but Nvidia is a competitive company that had some genuine plans for itself on the desktop, once upon a time. The company’s name literally derives from the Roman goddess of envy, Invidia. If ARM CPUs take the lead in a way Intel and AMD are unable to answer — and yes, that’s a very big “if” — I can see Nvidia throwing its hat into the metaphorical ring.
Either way, PCWorld’s testing makes clear that we’re a long way from any Windows on ARM laptop that can compete with the Apple M1.