Intel BOT reshapes code execution through vectorization

Intel’s Binary Optimization Tool, launched alongside the Arrow Lake Refresh processor lineup that includes the Core Ultra 5 250K Plus and Core Ultra 7 270K Plus, is designed to improve application performance by altering how .exe programs execute. The tool may appeal to gamers seeking extra performance, but it also creates complications for benchmark developers because it changes runtime behavior in ways that affect result comparability.

Primate Labs found that Geekbench runs using BOT will now be flagged, and deeper testing showed that Intel’s optimization layer can deliver sizable gains in selected workloads. In applications such as Object Remover and HDR, performance increased by up to 30%. Testing indicates that these gains come from deep vectorization performed behind the scenes, suggesting BOT does more than apply minor tuning and instead meaningfully transforms the instruction mix used during execution.

Using Intel’s Software Development Emulator to inspect execution behavior, Primate Labs measured a total of 1.26 trillion instructions for a standard Geekbench 6 run, while a BOT-enhanced run completed with 1.08 trillion instructions. This is an impressive 14% reduction. The breakdown by instruction type showed an even larger shift in how work was processed. The number of scalar instructions needed to execute a program fell from 220 billion to 84.6 billion, while the number of vector instructions increased from 1.25 billion to 18.3 billion, a 13.7x increase.

Those results indicate that BOT converts inefficient scalar code into vectorized instructions that Intel CPUs can process more efficiently, including through instruction sets such as SSE2 and AVX2. The scale of the shift points to a much more complex optimization process than initially assumed. To account for that change in execution behavior, the Geekbench v6.7 update will include a BOT flag so future results can be clearly identified as BOT-enhanced or not.