Geekbench Updates: All Tests with Intel BOT Active Will Be Marked as 'Invalid'

Primate Labs has updated its benchmarking tool, Geekbench, to version 6.7, which introduces a mechanism designed to manage one of the most discussed innovations of recent Intel CPUs. We are, of course, referring to the Binary Optimization Tool (BOT), a technology supported by the Core Ultra 200 Plus (Arrow Lake refresh) and Core Ultra 300 (Panther Lake) processors, which enables performance boosts in certain application scenarios.

According to analyses conducted by Primate Labs, BOT can influence benchmark results unevenly. Initial tests showed increases of up to 40% in specific workloads, while subsequent checks have revised that figure down to about 5.5% in both single-core and multi-core scenarios. Some tests, such as HDR processing, still show improvements of up to 30%, indicating behavior that is highly dependent on the type of workload.

The central issue is the representativeness of the results. Geekbench was created with the goal of simulating a wide variety of real-world workloads, but BOT intervenes by replacing the executed code with targeted optimized binaries. This approach benefits only a few compatible desktop CPUs in specific contexts and does not reflect everyday experience across a wide range of applications.

Technically, BOT uses a system based on checksums of executables. Essentially, it analyzes the executable and compiles optimized code to enhance performance with that specific application. This behavior prompted Primate Labs to consider the altered results invalid, as extremely specific scenarios were affecting the overall result, which was no longer representative of the everyday experience in its entirety.

With Geekbench 6.7, there arrives a clear response: all tests run with BOT active will be marked as invalid. Currently, BOT does not have optimized binaries for Geekbench, so the differences would be marginal. However, with the feature active, all tests will be marked as invalid, even when it is optimized for Geekbench, to avoid skewed results.

Beyond the Intel issue, Geekbench 6.7 improves the recognition of Android SoCs, now providing precise model and architecture indications, introduces the name display for RISC-V processors, and enhances the stability of multi-thread workloads on Arm-based Linux systems.

Overall, the new version of the benchmark aims to ensure greater consistency and reliability in comparisons across different platforms, in a context where increasingly aggressive software optimizations risk compromising the interpretation of performance in real-world situations.