Intel's iBOT does more than code rearrangement - Geekbench notes 14.6x more vector instructions in new tests

A new update from Geekbench sheds light on how Intel's new iBOT (Intel Binary Optimization Tool) works under the hood. The team conducted a series of tests on a Panther Lake-equipped laptop and found that it uses checksum-based matching to run verified apps. Most notably, the analysis suggests the tool goes beyond simple instruction reordering. Instead, it replaces unoptimized instruction sequences with optimized alternatives, for example, vectorized code.

iBOT was announced alongside Intel's debut of its Core Ultra 200 Plus processor lineup. Intel benchmarks or profiles select applications in its labs to identify unoptimized instruction sequences. If it finds instructions that can be optimized to better fit the architecture, it restructures and reorganizes them. Intel called this analogous to Tetris. The company explicitly stated that iBOT does not skip any instructions and that no reverse engineering is involved, as the optimization is performed post-link while the tool runs with user-mode privileges.

4

One of the supported applications for this utility was surprisingly Geekbench, which, on average, increased scores by 8%. Due to a lack of official technical details, Geekbench flagged all benchmarks conducted on CPUs that supported iBOT as potentially invalid. Following a week of rigorous internal testing, the developers have implemented a more precise detection method that isolates iBOT-optimized runs from standard ones.

It seems Intel is using binary hashing to whitelist specific applications for iBOT optimization. A cryptographic hash changes if even a single bit of code is altered, giving Intel more granular control over applications that can run iBOT. This was discovered when the team found iBOT optimized Geekbench 6.3 while ignoring Geekbench 6.7. Likewise, iBOT was attributed to a 40-second startup delay before the program started. Subsequent runs only had a 2-second delay.

4

Using a utility to monitor which instructions are executed during a program run, the team ran over 100 iterations of the HDR workload on Geekbench 6.3. They discovered a 14% reduction in total code executed, suggesting that iBOT is doing much more than simply reordering instructions. Most strikingly, the data showed a massive 1,366% surge in vectorized instructions (a 14.64x increase) during optimized runs, showing that the utility is swapping scalar code for high-performance vector operations.

Geekbench says its benchmarks are written to reflect the diversity and unpredictability of real-world code. According to them, iBOT undermines this by replacing standard or generic code with hand-tuned alternatives. This effectively causes the benchmark to measure the CPU's peak performance rather than its real-world performance. Another fundamental issue is that iBOT only works with select Intel-approved applications rather than with every x86 workload. For the vast majority of real-world software, which lacks iBOT optimization, these performance gains simply won't exist, which makes the benchmark, Geekbench in this case, a poor predictor of daily world performance.

4

From Intel's perspective, iBOT is a non-intrusive way to extract maximum performance from a workload without altering game binaries or requiring developer intervention. It guides the hardware towards optimized instruction streams. While this certainly provides a tangible performance boost for everyday apps and gaming, Geekbench's decision to flag these results is equally logical. Geekbench is a tool that is used to measure everyday performance across a range of workloads, and iBOT creates a best-case scenario that most software cannot replicate.

Geekbench will include built-in telemetry to detect if iBOT optimizations are active during a benchmark. Versions 6.6 and older will remain flagged indefinitely because they lack tools to confirm whether iBOT has modified the code.