These representative measurements describe GACT website workflows and simulated workloads. They were recorded on an Intel Core i7-1365U workstation with approximately 16 GiB RAM and R 4.4.1 on Windows. Exact software versions vary between runs; historical evidence is labeled separately. These timings describe updated workflows, not the original UK Biobank manuscript analyses. The scoring grid below varies a bounded set of simulated workload controls.
| Operation | Threads | Wall time | Resident memory |
|---|---|---|---|
| Database download and setup | 1 requested | 544.20 s | 2.68 GiB |
| VEGAS complete initial analysis | 4 decoding; serial spectra | 899.94 s | 5.88 GiB |
| LDSC | No fitter thread argument | 9.28 s | 3.02 GiB |
| GNOVA | No fitter thread argument | 6.39 s | 3.09 GiB |
| SumHer equal-marker baseline | No fitter thread argument | 12.87 s | 3.53 GiB |
| Chromosome-set LDSC preparation and fit | No fitter thread argument | 29.91 s | 3.27 GiB |
| HESS regional preparation | 4 | 1128.67 s | 3.18 GiB |
| HESS raw pair refit | No fitter thread argument | 3.61 s | 0.41 GiB |
| HESS partial pair refit | No fitter thread argument | 3.31 s | 0.43 GiB |
Database setup transfers 2.57 GiB and retains 3.04 GiB locally (2.63 GiB non-ZIP files and a 0.41 GiB ZIP archive, excluding the separate genotype reference and serialized descriptor); its 544.20-second clock is a published earlier-version download/setup measurement, separate from package installation. Network conditions affect this measurement.
VEGAS covers 17,862 genes, two traits and a 503-person EUR reference. The 899.94-second complete initial analysis includes regional spectra, prepared-object conversion and alignment; input-file loading is separate. Its 5.88 GiB is the complete-worker peak including loading. Preparation uses four decoding threads while regional products/eigensolvers are serial. Component clocks are 643.36 s for spectra, 79.52 s for conversion and 97.69 s for the initial fit. A separate prepared refit takes 71.44 s, excluding the 17.26-second object-loading cost. Do not add this alternative refit to the initial workflow.
Correlation fits use 7,221,923 common marker rows. Their memory is a sampled analysis peak with inputs loaded before the clock, rather than a whole-process high-water mark. HESS prepares 1,702 regions and 18,696 components; refits reuse compact moments. SumHer is an equal-marker baseline, not an upstream LDAK comparison. Scores are reused; estimator-specific assumptions remain in the correlation tutorial.
The panel contains 5,000 people and 37,991 filtered markers. Downloading, simulation, QC, startup and whole-genome sparse-LD preparation are excluded from these fit times.
| Operation | Threads | Wall time | Process peak memory |
|---|---|---|---|
| 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 (5.22–6.28) | 0.56 GiB |
| Genome-wide BayesC fit from prepared LD | 1 | 2.36 s (2.03–2.38) | 0.81 GiB |
| Genome-wide credible-set grouping | 1 | 3.95 s (3.83–4.16) | 0.52 GiB |
Association rows are single runs. Fine-mapping rows are medians of three serial repeats; memory is the median whole-process peak working set. Regional BayesC uses one chain, 1,000 post-burn sweeps and 100 burn-in sweeps over four regions containing 850, 785, 1,091 and 1,399 markers. The genome fit uses one chain and 100 sweeps without burn-in. The tutorial’s two-chain calls have not been timed. These short runs measure computation, not convergence or calibrated inference.
The following historical standalone measurements separate score application from fitting. They are suitable as labeled representative workload evidence. They do not establish the cost of every current software version.
| Operation | Simulated workload | Threads requested | Wall time | Process peak memory |
|---|---|---|---|---|
| Total score | Artificial-LD panel; 2,000 recipients; 50,000 SNPs | 1 | 8.03 s | 192.3 MiB |
| Three synthetic pathway scores | Artificial-LD panel; 2,000 recipients; 50,000 SNPs | 1 | 6.30 s | 192.8 MiB |
| Total score | Artificial-LD panel; 2,000 recipients; 50,000 SNPs | 4 | 3.23 s | 192.9 MiB |
| Three synthetic pathway scores | Artificial-LD panel; 2,000 recipients; 50,000 SNPs | 4 | 2.66 s | 192.7 MiB |
| Total score | 1000G-derived gsim panel; 2,000 recipients; 5,360 SNPs | 1 | 1.03 s | 157.2 MiB |
| Three synthetic pathway scores | 1000G-derived gsim panel; 2,000 recipients; 5,360 SNPs | 1 | 2.41 s | 157.8 MiB |
The artificial-LD panel uses binomial/block-copy genotypes and gsim phenotypes. The 1000G-derived panel uses gsim reference-conditioned simulation from 503 EUR donors on chr22; filtering retains 5,360 SNPs. Both have 2,000 held-out score recipients and three synthetic pathway memberships. The reference-derived scores use fixed simulated effect weights, not posterior weights from a model fit. No UK Biobank data are used for these scoring measurements.
Operation clocks exclude recipient preparation, input loading and serialization; the memory peaks cover the complete R process including loading and saving. Thread counts are requested limits; runtime worker counts were not observed. Each setting is one execution. The reference-derived workload superseded artificial LD for the real-LD benchmark objective; the historical artificial-LD thread comparison remains useful with this qualification.
A different 50,000-marker workload scored a total plus 20 artificial gene-marker sets in 14.88 s at one thread and 5.31 s at four threads. Its driver did not record peak memory; memory from the rows above cannot be assigned to that combined workload.
These reruns use 17,955 saved gene scores and 1,324 Reactome sets from the already installed hsa.0.0.1 database, gsea 0.1.2 and gbayes 0.1.6. Gene-association statistics were not recomputed: their upstream cost is excluded, not measured as zero. Cached fixture loading is separate from the calls below; no new database acquisition, genotype processing or simulation was performed.
| Full-data model | Preparation (s) | BLR fit (s) | Whole call (s) | Process peak (GiB) |
|---|---|---|---|---|
| Bayesian MAGMA: single-trait BayesC | 1.33 | 4.75 | 6.07 | 0.16 |
| Bayesian MAGMA: joint GWAS1/GWAS2 BayesC | 37.81 | 30.99 | 68.86 | 0.52 |
| PoPS: one full-data BayesC model | 1.02 | 6.77 | 7.82 | 0.15 |
| PoPS: one full-data BayesR model | 1.81 | 84.88 | 86.72 | 0.15 |
Preparation includes input checks, set mapping and model-specific summaries. Scalar independent-gene set models compute cross-products directly without forming X. The joint-trait model retains its general projected preparation. Native BLR time includes checks, sampling, posterior accumulation and original-unit coefficient back-transformation. Whole-call time also includes posterior summaries, prediction where relevant and remaining interface overhead; serialization and process startup are excluded. Memory is the complete R worker’s lifetime peak, including loading and saving, and cannot be assigned to an individual phase.
All rows use two chains, seeds 10/11 and one requested thread. Single-trait MAGMA uses BayesC with pi=.01 and 5,000 sampling plus 1,000 burn-in sweeps; joint MAGMA uses 1,000 plus 200. MAGMA retains its default fixed hyperparameters; PoPS learns slab variance, residual variance and mixture weights. The two single full-data PoPS models use their current PoPS priors and learning flags, pi=.001, 5,000 sampling plus 1,000 burn-in sweeps. For BayesR that scalar pi is divided among three active components; it is not the .95/.02/.02/.01 mixture from the earlier qgg-oriented probe. Pathway residual normalization remains n-1 for an intercept-only fit. These are representative current workflows, not original manuscript model reruns or convergence evidence.
| Validation workflow | Preparation (s) | Sum of BLR fits (s) | Prediction (s) | Whole call (s) | Process peak (GiB) |
|---|---|---|---|---|---|
| PoPS BayesC: 22 held-out chromosome folds | 33.64 | 156.90 | 1.27 | 192.97 | 0.24 |
| PoPS BayesR: 22 held-out chromosome folds | 35.82 | 1555.23 | 1.30 | 1593.50 | 0.26 |
These are 22 training models, each with two chains and 5,000 sampling plus 1,000 burn-in sweeps. They are validation costs and are excluded from the single-fit table. Saved gene scores are reused in every fold; preparation uses only training genes. The native fitting column is the sum over folds, with preparation and held-out prediction recorded separately. Each setting has one execution; there is no repeated-run timing uncertainty estimate.
The updated independent set path needs no dense gene-by-feature
matrix. The 1,536 MiB preparation limit is retained for consistency; it
is a matrix-payload guard, not measured process memory. The joint model
still uses projected preparation. Earlier preparation-included runs and
the separate one-chain 14.21-second native probe remain historical
evidence, not current rows or matched speedup baselines. A direct
sufficient-statistic R interface is now available as
gsea::blr(), backed by native gbayes. The qgg R/native
vb/vg positional mismatch is not reproduced; these timings make no
external qgg or MAGMA/FUMA comparison.
See the rerun report and phase records for settings, per-fold timing, binary/input hashes, memory definitions and the excluded duplicate-membership harness diagnostic.
| People | Available SNPs | Sets | Adjacent overlap | Union SNPs | Seconds 1 / 4 threads | Process MiB 1 / 4 threads |
|---|---|---|---|---|---|---|
| 500 | 10000 | 0 | 0 | 10000 | 1.77 / 0.72 | 129.3 / 129.6 |
| 500 | 10000 | 3 | 0 | 1500 | 0.47 / 0.28 | 128.3 / 129.4 |
| 500 | 10000 | 3 | 0.5 | 1000 | 0.36 / 0.21 | 129.7 / 129.4 |
| 500 | 10000 | 20 | 0 | 10000 | 3.57 / 1.11 | 134.3 / 135.4 |
| 500 | 10000 | 20 | 0.5 | 5250 | 1.66 / 0.64 | 132.2 / 136.1 |
| 500 | 50000 | 0 | 0 | 50000 | 13.62 / 3.77 | 155.7 / 155.8 |
| 500 | 50000 | 3 | 0 | 1500 | 0.53 / 0.28 | 155.7 / 156.2 |
| 500 | 50000 | 3 | 0.5 | 1000 | 0.42 / 0.27 | 155.5 / 155.7 |
| 500 | 50000 | 20 | 0 | 10000 | 4.17 / 1.19 | 158.8 / 159.0 |
| 500 | 50000 | 20 | 0.5 | 5250 | 2.10 / 0.75 | 167.9 / 159.2 |
| 2000 | 10000 | 0 | 0 | 10000 | 1.75 / 0.68 | 129.6 / 125.4 |
| 2000 | 10000 | 3 | 0 | 1500 | 0.48 / 0.25 | 129.5 / 129.7 |
| 2000 | 10000 | 3 | 0.5 | 1000 | 0.37 / 0.19 | 129.9 / 130.1 |
| 2000 | 10000 | 20 | 0 | 10000 | 3.83 / 1.04 | 134.8 / 135.4 |
| 2000 | 10000 | 20 | 0.5 | 5250 | 1.67 / 0.66 | 134.0 / 132.2 |
| 2000 | 50000 | 0 | 0 | 50000 | 13.67 / 3.78 | 156.2 / 155.7 |
| 2000 | 50000 | 3 | 0 | 1500 | 0.59 / 0.31 | 156.3 / 156.3 |
| 2000 | 50000 | 3 | 0.5 | 1000 | 0.45 / 0.27 | 156.3 / 155.8 |
| 2000 | 50000 | 20 | 0 | 10000 | 4.75 / 1.35 | 159.0 / 166.4 |
| 2000 | 50000 | 20 | 0.5 | 5250 | 2.39 / 0.73 | 159.2 / 159.7 |
The new grid uses current gbase 0.1.3 and gscore 0.1.1 on frozen gsim-derived effects and artificial-LD genotypes. Each synthetic set contains 500 SNPs. Available SNP count and the number actually contributing to set scores are distinct. Weights are fixed, coding is standardized, and completed scores are unscaled. Each of 40 cells is one execution; the grid supplies bounded workload sensitivity, not repeated-run confidence intervals or a universal scaling law. Preparation and loading precede the operation clock; peaks cover the entire R worker including serialization. No UK Biobank data or new model fitting is used for scoring.
| Operation | Status | Seconds | Process peak GiB |
|---|---|---|---|
| Chromosome-component Gaussian LOCO | completed | 90.26 | 0.35 |
| Chromosome-component prior-selected LOCO | completed | 67.07 | 0.37 |
| Two-chain genome-wide BayesC from cached LD | completed | 2.91 | 0.67 |
| Two-chain regional BayesC | completed | 6.13 | 0.34 |
See the dated evidence report for controls, validation and failures.
The overview focuses on gsuite ecosystem workflows. It does not measure or compare external MAGMA/FUMA execution.
The consolidated evidence table retains component clocks, qualifications and source references. Original file locations and checksums identify the imported evidence; consolidation source hashes identify the selected records. Scripts are preserved in benchmarks with their original execution assumptions. Large datasets and source archives remain outside this small-file collection.
Selected records: artificial-LD measurements, reference-derived simulation measurements, historical gene-set and PoPS records, and correlation timing and memory.
Memory is resident working set, not result size, incremental solver allocation or machine RAM. Missing memory stays missing. Historical records and the new 19 October ecosystem runs are identified separately. No benchmark result is a claim of Bayesian convergence or external-tool equivalence.