New Windows Build Brings AVX Support To Snapdragon X For Gaming And More
One sore spot in our testing, though, particularly around the top-end Snapdragon X Elite X1E-84-100 chip in the Samsung Galaxy Book4 Edge, was that some of the applications we wanted to test simply could not run on that processor. The version of Windows that runs on Qualcomm's Arm-based SoCs is equipped with an emulation layer known as Prism, that allows programs compiled for x86-64 processors to run on the Arm chips. However, at that time, it lacked support for important SIMD instructions, including both AVX2 as well as the original AVX.
Well, that issue has now been resolved, at least according to Microsoft. In a post on the Windows Insider Blog, the company says that Windows 11 Insider Preview Build 27744 includes a major feature update to Prism that brings along support for both AVX and AVX2, as well as additional instructions, like FMA, BMI, F16C, XSAVE/OSXSAVE, RDRAND, and more.
So, what does that mean for end users? Well, once this update propagates to the mainstream, users on Snapdragon-based Windows machines will be able to run x86 Windows software that requires these instruction sets. This will make applications that weren't compatible before suddenly begin working, although Microsoft notes that only x64 applications can use these CPU features. Older 32-bit x86 applications will not detect them, and this will apply (for example) to games and other applications that use a 32-bit launcher to detect CPU features.
We mention games because obviously, games are one of the major targets of this; very few desktop applications actually require AVX support, and most of those have Arm-native builds. Recent PC games have started to use AVX, though, as the CPU feature provides a major speed-up for certain types of game processing, particularly physics. We found numerous games that simply would not run on our Snapdragon machines because of the AVX requirement.
One thing Microsoft doesn't mention is performance, though. AVX instructions enable extremely dense compute operations, and Qualcomm's Snapdragon processors have their own vector compute units that should be able to handle the same kind of math, but it's not clear if the translation process has a deleterious effect on performance here. We observed performance effects from Prism emulation that were both smaller than feared but larger than we would like; impressively, complex modern full 3D games are quite playable, yet performance still falls off compared to native Arm apps.
At this point, the ball for Snapdragon X Elite gaming is largely in Qualcomm's court. Qualcomm is still finding its way regarding Windows graphics drivers, and many commentators observed issues with the Adreno graphics drivers for the Snapdragon X Elite processors, beyond those related to Prism emulation. Here's hoping these machines continue to see significant software improvements, if Snapdragon is to be a serious competitor over the long haul with AMD and Intel in the Windows laptop space, beyond thin and light productivity machines.
