REVIEWS / GENERAL / OWNER INSIGHTS
🦉 WE READ 276 OWNER COMMENTS
Geekbench 6: what owners actually say
Geekbench 6 is widely used for chip comparisons but many owners and enthusiasts consider it unreliable, gameable by manufacturers, and detached from real-world performance.
What owners complain about
- Unreliable for real-world comparisons COMMON
Multiple users call Geekbench 'pretty useless for actual performance comparisons' and 'the least reliable popular benchmark,' requesting alternatives like Cinebench that better represent real workloads.
- Manufacturers game the benchmark COMMON
Users report Intel's Binary Optimization Tool (BOT) applies hand-tuned patches specifically to Geekbench binaries, replacing scalar with vector instructions only Intel CPUs can run. Qualcomm's X Elite was benchmarked on a desktop with intense cooling and a custom Linux build, scores never replicated on real laptop hardware.
- Scores meaningless without power data SOME
Commenters repeatedly note that performance numbers for mobile chips are useless without accompanying power consumption figures; a chip matching an M2 at triple the wattage is a very different proposition.
- Inflated generational jumps from instruction-set support SOME
GB 6.3 added SME support for a subset of tests, making those run ~2x faster on Apple M4. Users note the actual IPC increase on general-purpose code is only ~3%, meaning headline gains mostly reflect clock speed and new instructions rather than real architectural improvement.
- Perceived Apple bias FEW
Users observe that Apple chips increased 10-20% more from GB5 to GB6 than AMD or Intel chips did, suggesting the workload mix or threading model favors fewer fast cores over many slower ones, which aligns with Apple's design but may mislead buyers comparing across ecosystems.
What owners love
- Improved multi-threading model
GB6 switched from a 'separate task' to a 'shared task' model that users say better reflects how most consumer applications actually use multiple cores, rather than rewarding having many slow cores.
- Transparent internals documentation
Users reference and appreciate that Geekbench publishes an internals document explaining the rationale behind scoring changes and workload selection.
- Serves as common reference point
Despite criticisms, Geekbench remains widely used across platforms and reviewers, giving users at least one standardized number to compare wildly different chips.
Surprising patterns
- Intel BOT apparently includes a checksum to identify known binaries and apply optimizations only to them; users note the 40-second 'optimization' delay may be theatrical rather than computational.
- Users point out that applying targeted binary patches to cheat benchmarks has precedent in GPU shader optimization (referencing 'quack3.exe'), and some see Intel's behavior as an expected escalation rather than a shock.
- Multiple commenters treat benchmark leaks from developer platforms (e.g., AMD 'Lilac') as essentially fictional for predicting final consumer hardware performance, yet benchmarks from these platforms still drive coverage.
WHO SHOULD SKIP IT
Buyers who need a single reliable number to compare real-world CPU performance across vendors should skip Geekbench 6 and look at application-specific benchmarks, as scores can be inflated by vendor-specific optimizations, instruction-set additions, and desktop-cooled mobile chips.
Synthesised from 276 real owner comments across 4 platforms. Every point is grounded in the comments — no marketing, no AI guessing. How we do it →