RAM benchmark: read, write and copy bandwidth plus access latency, in addition, it detects your hardware configuration via

SIV (CPU, mainboard, memory modules, timings) and lets you submit your score to the online database.

 

Real-World mode (default):
Measures what normal applications actually get. Single-threaded, no priority tricks, median of multiple runs. Bandwidth is averaged over six different

instruction-kernel variants on a 512 MB working set (bigger than any cache). Latency is a fully random pointer chase over 128 MB including address-translation

(TLB) costs – because real software pays them too.

 

MAX mode:
Try to measures the hardware limit (>95%). All physical cores (P-cores preferred, spread across chiplets), threads pinned, high priority, power throttling off, best value instead

of median. WRITE/COPY auto-tune thread count and store method and report the winning combination. Latency uses a prefetcher-proof, translation-amortized

pattern and reports the near-pure DRAM latency.

 

Rule of thumb:
Real-World = what your software experiences. MAX = what your hardware can do. Both are correct – they answer different questions. That's why switching modes

resets all results: scores are never a mix of both.

What gets uploaded?

Only what you see and confirm: your hardware configuration (as detected by the scan), the benchmark settings and score, and a screenshot of the result window.

Upload is only possible for completed, verification-passed runs, and only once per run.