RequantTESTNET
Requant research

Matrix-Free Poisson and Heat Operators on GPUs

Abstract

Across 384 attempts, matrix-free operators are compared with direct CUDA, libCEED and cuBLAS controls. The study checks discretization, complete Poisson solves and heat-step trajectories.

Study 12 of the Consumer Blackwell programme. See the research methodology for timing and replay boundaries.

This is a bounded measured study, not a hardware ceiling, novelty claim or a quality-safe dispatcher. External context is in PDE_SOURCES.md; executable details and limits are in PDE_REPRODUCTION.md.

Executed scope and numerical closure

The full matrix contains 24 independent cohort/geometry collections and 384 declared configurations. Including separate smoke, fresh CPU replay covered 400 attempted / 382 available configurations, 11840 operator observations, 5920 complete Poisson solve observations, 3072 heat trajectories and 18176 retained heat steps. It checked 11858 distinct full final-vector records and 430 complete retained GEMM witnesses.

Each full geometry has two independent sine-mode cohorts; E=1/2/4/8, q=4/8/16. Changing smooth and rough inputs include nonzero operator boundaries. Heat is explicitly bounded to E=1/2, q=4/8/16, final time0.01 and4/8 steps. The remaining larger geometries cover operators and Poisson, not unexecuted large heat trajectories.

Direct FP32/64 CUDA, pinned libCEED cuda/ref/shared/gen, grouped native FP4/FP8/FP16/BF16/strict-FP32, and ungrouped short-q cuBLAS references were actually attempted. Half/BF16 candidates require recorded eligibility flags; numerical flags/API types/root resources do not constitute an instruction trace. Short-q controls do not inherit the proposal's artificial128-wide padding.

Complete online operator at fixed error

Every table choice uses the worst independently checked error among all observations in that timing route, and the fastest eligible non-FP4 implementation. Effective cost includes common benchmark preparation and construction/source load/calibration/first use divided by the stated reuse, plus median complete operation time. Common preparation includes analytic reference-field generation: this is benchmark-inclusive, not minimal deployment initialization.

BoundaryError budgetReuseFP4 feasible groupsReference feasible groupsFP4 wins
resident0.001100000/2424/240/24
resident0.01100000/2424/240/24
resident0.1100002/2424/240/24
host0.001100000/2424/240/24
host0.01100000/2424/240/24
host0.1100002/2424/240/24

The complete tables also include reuse1/100. Padding, dynamic source packing, quantization, actual native product, all-axis weighting, deterministic global assembly and context acknowledgment are not replaced by a prepacked microbenchmark. Host service includes H2D and D2H; its CUDA event span ends before host D2H and is not the same boundary as host wall time.

Poisson: original residual and continuum forward error

Recursive convergence of an input-dependent/nonsymmetric low operator is not physical convergence. Fixed runs retain this distinction. Reliable runs pay for bounded low-operator defect corrections, true residual tests, rejected progress and FP64 fallback. FP4-assisted/fallback rows are not pure-FP4 solves.

The full data retain 1628 Poisson observation(s) missing their nominal original residual target and 172 FP4-assisted Poisson observations using fallback. No failed repetition is silently dropped from a configuration's frontier eligibility.

BoundaryResidual budgetContinuum mass-L2 budgetReuseFP4-assisted/fixed feasibleReference feasibleFP4-labelled wins
resident0.0010.011000019/2421/240/24
resident1e-080.011000019/2421/240/24
host0.0010.011000019/2421/240/24
host1e-080.011000019/2421/240/24

Discretization and temporal refinement

The following independently replayed strict fixed-CUDA-FP64 controls show continuum mass-L2 error, not merely an internally small algebraic residual. At high order, iterative stopping and floating-point error can dominate the remaining spatial discretization error; monotonic convergence is not asserted below that floor.

CohortEqWorst original residualWorst continuum mass-L2 error
a142.141778e-160.01303956
a241.18588e-150.0006494484
a446.08952e-152.271411e-05
a842.208934e-147.278705e-07
a187.896608e-151.894045e-07
a285.348962e-091.636015e-09
a487.245568e-095.336853e-11
a888.504597e-091.813636e-11
a1167.838373e-091.236693e-10
a2168.772635e-092.07999e-10
a4167.787941e-093.218917e-11
a8167.996211e-097.88523e-12
b141.065891e-161.680864
b241.308054e-150.0605276
b447.121209e-090.003099077
b848.102881e-090.0001017411
b182.688511e-090.003657159
b284.090404e-091.656554e-05
b487.762741e-093.627192e-08
b887.971073e-091.553481e-10
b1166.287105e-091.913364e-09
b2169.286973e-099.690416e-10
b4169.247177e-095.373842e-10
b8168.694172e-098.274099e-11

Heat tables retain each step state and RHS and validate RHS = M*previous_state/dt against the preceding retained vector. Continuum decay and sampled continuum backward-Euler targets are separate, exposing time discretization rather than equating a solved implicit system with the exact heat PDE. Frontier step residuals are recomputed from every retained state/RHS, not taken from scalar GPU claims. Frontiers require every step residual and final continuum forward error and may choose between the actually executed 4/8-step trajectories.

DatasetStepsMaximum step residualContinuum mass-L2 errorError against continuum BE target
a-e1-q446.064554e-170.0068286960.003633984
a-e1-q481.291584e-160.001577590.003764261
a-e2-q445.631665e-100.010524850.000465411
a-e2-q484.302338e-090.0053983450.0004888059
a-e1-q841.65826e-090.010500841.663075e-07
a-e1-q886.267415e-090.0053620351.831392e-07
a-e2-q844.172605e-090.010500843.065198e-09
a-e2-q885.261417e-090.0053620367.359424e-09
a-e1-q1647.980944e-090.010500844.653204e-09
a-e1-q1688.586193e-090.0053620363.392483e-09
a-e2-q1649.443818e-090.010500843.415621e-09
a-e2-q1689.94945e-090.0053620367.620412e-09
b-e1-q448.923691e-171.3209130.9483076
b-e1-q482.056438e-161.2893491.080724
b-e2-q445.103955e-090.23837390.06010021
b-e2-q481.322375e-090.14814260.06681127
b-e1-q847.659991e-090.20586860.004081265
b-e1-q883.331709e-090.10668880.00450431
b-e2-q848.280975e-090.20890561.885377e-05
b-e2-q887.565249e-090.10980981.939737e-05
b-e1-q1649.121471e-090.20890561.092981e-08
b-e1-q1688.922668e-090.10980982.329457e-08
b-e2-q1649.961207e-090.20890565.562938e-09
b-e2-q1689.159662e-090.10980988.345791e-09

The strict FP64 controls distinguish the temporal error reduced by refinement from persistent under-resolved spatial error. No monotonic low-precision or unresolved-mesh convergence claim is inferred from a small recursive residual.

BoundaryResidual budgetContinuum mass-L2 budgetReuseFP4-labelled feasibleReference feasibleFP4-labelled wins
resident1e-080.1100006/126/120/12
resident1e-080.01100006/126/120/12
host1e-080.1100006/126/120/12
host1e-080.01100006/126/120/12

The full heat data retain 4510 step observation(s) missing their nominal original residual target. Fixed low-precision misses remain negative evidence, not successful physical acceleration.

Method-local memory categories

The following E2/q8 sample excludes shared benchmark input/reference buffers. Library request lengths remain separate, not invented opaque-owned-byte totals. Temporary arrays, reduction buffers, allocator/JIT caches and physical peaks are not claimed audited.

ImplementationPython named bytesNative named bytesCEED qdata request bytesCEED index request bytes
0:cuda:fp321177190not measurednot measured
3:cuda:fp641341030not measurednot measured
5:ceed:fp6484375not measured13107216384
7:tensor:fp433071120480not measurednot measured
9:tensor:fp1633071133619968not measurednot measured
13:small:fp1623266333579136not measurednot measured

Retained limits and negatives

There are 18 full configuration availability failures. The pinned cuda/ref q16 basis exceeds its static shared-allocation limit, while optimized shared/gen operators are retained. The largest grouped geometry encounters the original native bridge N<=8192 guard. These are respectively a specific kernel/compiler allocation failure and an implementation guard, not a blanket GPU/Tensor Core limitation. Unavailable inherited heat combinations are not separately executed heat failures.

All negative records are in results/pde-negative-results.json. CUDA, native, libCEED and short-q setup and allocation/request scopes remain visible in master tables. Opaque libCEED memory is not ranked against another family's fuller named accounting; no physical peak-memory saving is established.

This common host-controlled Jacobi-PCG experiment stages each callback and performs FP64-vector reductions; it is not a best-of-all fully GPU-resident solver claim. First use follows explicit calibration, and solver measurements follow operator warmup. Import/device-discovery, warm disk/JIT/allocator history, diagnostics between requests and shared-device activity prevent cold-system/exclusive/saturated-throughput claims.

Every timed final vector is retained losslessly, including nondeterministic variants. Numeric replay reconstructs original-data encoding/products/post-processing and final original-system residuals. It does not certify every unretained CG intermediate or measure board/system energy. Full source30,4binary,fixture pre/post and child-exit closures are retained; exact numerical/source/cost receipts are prerequisites for these tables.

Artifacts

Evidence and source revision

Cite this report

Requant. “Matrix-Free Poisson and Heat Operators on GPUs.” Technical report, 19 November 2026. Evidence revision d12bcf1c3270f28fe85828725fb2f2c5c790b986. Canonical report.

This is an organizational technical report, not a peer-reviewed publication.