Tango uses state of the art optimizations to minimize the overhead of binary translation. The Technology page describes these optimizations in more detail.

In this page we present a few benchmark results which show the performance of Tango. The benchmark platforms support both 32-bit and 64-bit ARM in hardware to allow for a fair comparison. All benchmarks were compiled as 32-bit ARM code and the resulting binaries were run with Tango enabled ("Translated") and disabled ("Native").

Geekbench 4 on AWS

  • System: AWS a1.xlarge dedicated instance
  • OS: Ubuntu 18.04 LTS
  • CPU: 4x Cortex-A72
  • Image: AWS Tango reference image
  • Benchmark: Geekbench 4.0.2 built for GNU/Linux and ARMv7

Geekbench 4 on Nexus 5X

  • System: LG Nexus 5X
  • OS: AOSP Android 8.1
  • CPU: 2x Cortex-A57 & 4x Cortex-A53
  • Image: Nexus 5X Tango reference image
  • Benchmark: Geekbench 4.3.4 APK with 64-bit native libraries stripped out.

SPEC CPU 2006 on AWS

  • System: AWS a1.large dedicated instance
  • OS: Ubuntu 18.04 LTS
  • CPU: 2x Cortex-A72
  • Image: AWS Tango reference image
  • Benchmark: SPEC CPU2006 built from source for GNU/Linux and ARMv7 (build options "-O3 -mfpu=neon-fp-armv8")
SPEC CPU2006 on AWS, 2019.06