DMSD Horizon Edition v0.3 — Compute Capacity Overlay

Magnitude ≠ work · HPL / HPCG / FP32 / AI TOPS remain separate benchmark lanes.
Capacity tier
Work model
work
10^24 compatible ops
efficiency
100%
Compatible throughput
—
Compute lower bound
—
Power
—
Energy lower bound
—
Profiles
Selected profile
Warning
LineShine benchmark split
MetricThroughputMeaning
HPL2.1984 EFLOP/sdense linear-algebra benchmark
HPCG22.0049 PFLOP/smemory/communication-sensitive benchmark
Ratio≈99.9×same machine, radically different workload throughput