Qualcomm Snapdragon X2 Elite mitigates Windows on Arm translation penalty

Microsoft's Prism binary translator lets Arm Windows PCs run x86-64 software, but at a performance cost. A detailed analysis finds that translated apps execute roughly twice as many instructions. Qualcomm's new Snapdragon X2 Elite cores, with up to 9-wide pipelines at 5 GHz, are powerful enough to absorb much of that overhead, per chipsandcheese.com.

Bottom line — Binary translation still imposes a performance penalty, but Qualcomm's high-performance cores can offset it by several generations.

Go deeper (7)

• Prism translates x86-64 instructions to aarch64 in a straightforward, often unoptimised way, replicating complex x86 addressing modes with multiple Arm instructions, per chipsandcheese.com.
• A Geekbench 7 workload running through Prism executes roughly twice as many aarch64 instructions as the native x86-64 version, the analysis found.
• The penalty varies by workload: heavily vectorised tasks like video playback suffer most because 256-bit AVX has to be emulated with 128-bit NEON instructions.
• Qualcomm's Snapdragon X2 Elite P-Cores (9-wide, 5 GHz) and E-Cores (6-wide, 3.6 GHz) show smaller relative penalties than older Arm cores like Neoverse N1, thanks to wider pipelines and larger caches.
• Prism is optimised for Qualcomm's Snapdragon X series, using features like FEAT_LPCPC2 for more efficient load-acquire instructions, though the instruction count penalty is similar across Arm CPUs.
• The translation cache (XtaCache) stores generated code to avoid retranslation, but Microsoft hasn't documented the format, making it hard to analyse optimisations, per chipsandcheese.com.
• Other articles (Wikipedia, GkGigs, Cyberockk) provide general background on binary translation techniques, but none add new facts specific to this Qualcomm/Prism analysis.

Read the reporting