These are recorded tutorial runs on Windows 11, R 4.4.1 and an Intel Core i7-1365U with about 16 GiB RAM. Memory is the whole R process’s peak resident working set, including required inputs. It is not the size of the returned result or the additional memory needed by the solver.
The benchmark overview also includes recovered separate simulated scoring timings and memory, updated Reactome Bayesian MAGMA and full-data PoPS models with separate preparation/BLR clocks, and separately timed 22-fold PoPS validation, a controlled 40-cell scoring grid, and remaining measurement gaps.
The simulated panel has 5,000 people and 37,991 filtered markers. Analysis times exclude downloading, data preparation and process startup. The mapping rows summarize three serial repeats; the other rows are single runs.
| Tutorial / operation | Threads | Time | Peak resident memory |
|---|---|---|---|
| Database download and setup | — | 9 min 4 s | 2.68 GiB |
| Linear regression | 4 | 1.64 s | 153 MiB |
| Gaussian LOCO | 4 | 57.00 s | 386 MiB |
| Prior-selected LOCO | 4 | 124.12 s | 392 MiB |
| Four regional BayesC fits and posterior sets | 1 | 5.48 s | 0.56 GiB |
| Genome-wide BayesC fit from prepared LD | 1 | 2.36 s | 0.81 GiB |
| Genome-wide credible-set grouping | 1 | 3.95 s | 0.52 GiB |
Regional mapping used one chain, 1,000 post-burn sweeps and 100 burn-in sweeps; the four regions contained 850, 785, 1,091 and 1,399 markers. Its observed time range was 5.22–6.28 s. Genome fitting used one chain and 100 sweeps without burn-in (2.03–2.38 s); grouping took 3.83–4.16 s. The tutorial shows two-chain calls; their separate current measurements and completion status are in the benchmark overview. The overview also records the chromosome-component glma checks. Database setup covers 2.57 GiB of transfers and retains 3.04 GiB locally, including a 0.41 GiB ZIP archive; non-ZIP files occupy 2.63 GiB. The separate genotype reference, serialized descriptor and package installation are excluded from that footprint.
Genetic correlation uses the common 7,221,923-marker availability selection and stored LD scores. The reference has 503 EUR individuals. Time excludes input loading and saving; memory is the peak sampled during analysis with the required inputs already loaded.
| Operation | Threads | Time | Peak resident memory |
|---|---|---|---|
| LDSC | — | 9.3 s | 3.02 GiB |
| GNOVA | — | 6.4 s | 3.09 GiB |
| SumHer baseline | — | 12.9 s | 3.53 GiB |
| Chromosome-set LDSC preparation and fit | — | 29.9 s | 3.27 GiB |
| HESS regional preparation | 4 | 18 min 49 s | 3.18 GiB |
| HESS pair refit from compact moments | — | 3.3–3.6 s | 0.43 GiB |
The score-based fitters expose no thread argument. HESS prepared 1,702 regions and 18,696 retained components; its two refits reused the compact moments. Preparation and fitting are separate costs.
For the full VEGAS gene analysis, 17,862 genes and two traits shared the same regional spectra:
| Operation | Time | Peak resident memory |
|---|---|---|
| Marker-set LD eigenvalues | 10 min 43 s | — |
| Prepare reusable VEGAS object | 1 min 20 s | — |
| Initial two-trait VEGAS fit | 1 min 38 s | — |
| Refit from restored preparation | 1 min 11 s | — |
| Complete initial workflow, including preparation and alignment | 15 min | 5.88 GiB |
VEGAS’s memory figure covers the complete worker, including input loading; per-stage peaks were not recorded. Its preparation used four decoding threads; gene regions and the eigensolver were serial. The complete-workflow time is not the VEGAS fit time. Refitting excludes loading the restored preparation.
Runtimes depend on the data and settings. The native prepared-summary BLR timing, supplementary PoPS validation timings and the status of chromosome-component examples are reported in the benchmark overview. The earlier rows above retain their original settings and measurement boundaries.