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 v2.0.0. We run the benchmark on CPUs supporting both 32-bit and 64-bit ARM 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").
Note that newer ARM cores such as the Neoverse V1 and Cortex-X1 have highly optimized AArch64 execution. For example, on the Pixel 6's Cortex-X1, Geekbench 5 runs 40% faster in 64-bit mode than in 32-bit mode. As such, Tango translated code can take advantage of this and deliver better performance than original 32-bit code. When this happens, the performance gains are shown as positive percentages in Geekbench scores and negative percentages in SPEC run time differences.
Please click here for Tango v1.0.x performance page.
AWS Graviton 3
System: AWS c7g.xlarge, 4 cores, 8GB RAM
CPU core: 4x ARM Neoverse V1
OS: Ubuntu 22.04 LTS
Image: AWS Tango reference image
Geekbench 4
Benchmark: Geekbench 4.0.2 built for GNU/Linux and ARMv7
Geekbench 5
Benchmark: Geekbench 5.4.0 Preview for Linux ARMv7
SPEC CPU2006
AWS Graviton 2
System: AWS c6g.xlarge, 4 cores, 8GB RAM
CPU core: 4x ARM Neoverse N1
OS: Ubuntu 22.04 LTS
Image: AWS Tango reference image
Geekbench 4
Benchmark: Geekbench 4.0.2 built for GNU/Linux and ARMv7
Geekbench 5
Benchmark: Geekbench 5.4.0 Preview for Linux ARMv7