Pith. sign in

REVIEW 4 major objections 6 minor 40 references

Exploiting Parallelism for Fast Feynman Diagrammatics

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A GPU port of the combinatorial summation algorithm evaluates Feynman diagram series about 1000 times faster than the original CPU code.

desk verdict A useful GPU port of the CoS diagram-summation kernel with credible speedup numbers, but the headline 'orders of magnitude' applies to a microbenchmark, not the full DiagMC workload. read the letter →

arxiv 2501.00675 v1 pith:WMRJAOW5 submitted 2024-12-31 cond-mat.str-el hep-thmath-phmath.MPquant-ph

John Sturt , Evgeny Kozik This is my paper · ORCID
classification cond-mat.str-elhep-thmath-phmath.MPquant-ph
keywords FeynmandiagramscombinatorialsummationGPUaccelerationCUDAdiagrammaticMonteCarlostronglycorrelatedelectronsparallelcomputingGreen'sfunctions
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper reports that the combinatorial summation (CoS) algorithm for adding up all Feynman diagram integrands of a given perturbation order, originally developed for CPUs, can be restructured to run on graphics processing units (GPUs) with orders-of-magnitude speed gains. On a server-grade H100 the new implementation sums the full set of $n$-th-order diagram integrands at least a thousand times faster than the original CPU code, and even a consumer RTX 3090 gives hundreds-fold acceleration. Because the CoS graph is a layered multiply-accumulate structure, its evaluation maps naturally onto the single-instruction-multiple-thread model of GPUs, with one thread block per sampled vertex configuration. The paper argues this makes high-order diagrammatic many-body calculations, including diagrammatic Monte Carlo, practical on far more modest hardware than large computing clusters.

What carries the argument

The central object is the CoS directed graph: a layered network whose top node holds the value 1, whose edges multiply a node's value by a Green's function $G_{\alpha\beta}$ (and by $-1$ when closing a fermion loop) and add the result to a downstream node, and whose bottom node accumulates the sum of all diagram integrands. A distinct path through the graph corresponds to one diagram, but because each edge is shared by many paths, the factorial number of terms is collected in $O(n^2 2^n)$ operations. The paper's contribution is to flatten this graph into per-level arrays of edges and evaluate every edge of a level concurrently with a CUDA thread block, using a window of up to 1024 threads that sweeps through the level when it is wider than the block. One block handles one full graph evaluation for one set of vertex coordinates, so a grid of blocks evaluates many Monte Carlo configurations at once.

What would settle it

Run an end-to-end diagrammatic Monte Carlo calculation with the original CPU CoS code and the GPU implementation using identical truncation order, vertex-coordinate integration scheme, and target statistical error, then compare wall-clock times; a speedup far below three orders of magnitude would show that the graph-evaluation benchmark does not control total runtime.

Watch

Extended reading notes

Core claim

The central claim is that the recently introduced combinatorial summation (CoS) algorithm, which evaluates the sum of all connected Feynman diagram integrands of order $n$ in $O(n^2 2^n)$ floating-point operations by sharing edges of a directed graph instead of enumerating all $O(n!)$ diagrams, is naturally parallel and ports cleanly to GPU hardware. The authors show that flattening the graph into node-agnostic arrays of edges already gives about a $5\times$ speedup on a CPU, and that evaluating each graph level with a window of GPU threads makes the exponential edge count a parallel workload rather than a serial one. In head-to-head benchmarks against the original CoS CPU code, the GPU version evaluates the diagram series of an expansion order roughly eight higher in the same wall-clock time, corresponding to acceleration factors of about $100$ on consumer hardware and above $1000$ on the H100. They also report that single-precision arithmetic is numerically safe, with about $10^{-5}\%$ error at order 10, and that memory bandwidth rather than compute capacity is the limiting resource.

Load-bearing premise

The three-orders-of-magnitude claim rests on the assumption that graph evaluation dominates the runtime of practical diagrammatic calculations; if Monte Carlo sampling, integration, or data transfer costs are comparable, the acceleration of a full calculation would be smaller than the isolated-graph benchmark.

Editorial extensions

If this is right

  • In a fixed wall-clock time the accessible expansion order increases by about 8, which can push diagrammatic calculations into regimes where the residual series error is small enough for the asymptotic polynomial scaling with error to set in.
  • A consumer-grade GPU such as the RTX 3090 performs within noise of the server A40 in these benchmarks, so high-order diagrammatic Monte Carlo no longer necessarily requires an HPC cluster.
  • Because single-precision arithmetic costs only about $10^{-5}\%$ accuracy at order 10, reduced-precision data types can be used to relieve memory-bandwidth pressure and gain up to 80 percent more speed at intermediate orders.
  • The same parallel graph evaluator fits both stochastic Monte Carlo sampling and deterministic tensor-train integration schemes, since both need many independent integrand evaluations for different vertex configurations.
  • The graph topology matches the multiply-accumulate layers of machine learning, so future ML-oriented hardware such as Tensor Cores or NPUs is a plausible next target for further acceleration.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the reported speedups are for isolated graph evaluations per second, not for end-to-end diagrammatic Monte Carlo runs; real calculations also spend time on proposals, acceptance, integration over vertex coordinates, and host-device transfers, so the full wall-clock speedup could be smaller if graph evaluation is not the dominant cost.
  • Editorial inference: the acceleration curve peaks at intermediate orders and slopes back toward the underlying exponential scaling once graph levels exceed the maximum thread window, so the approximate 8-order gain is demonstrated in the studied range rather than guaranteed asymptotically.
  • Editorial inference: a direct test of the portability of these gains would be to re-run a known diagrammatic Monte Carlo application, such as the self-energy of the doped 2D Hubbard model, with the same truncation order and statistical error on CPU and GPU and compare wall-clock time; if the speedup tracks the benchmarks, previously marginal calculations become routine on modest hardware.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper describes a GPU implementation of the combinatorial summation (CoS) algorithm for evaluating sums of Feynman diagram integrands. The authors flatten the directed CoS graph into level-wise arrays of edges, process edges in parallel with CUDA thread blocks ('windows'), and distribute independent graph evaluations across blocks. They benchmark graph evaluations per second on a CPU and three NVIDIA GPUs (RTX 3090, L40S, H100), reporting up to three orders of magnitude speedup over the original CPU implementation of Ref. [10] and claiming that the GPU code reaches expansion orders about eight higher in the same time. They also report a CPU-only flattening speedup of about 5x, and argue that single-precision arithmetic introduces only about 10^-5% error on an order-10 test configuration. The paper concludes that this provides a platform for more accessible diagrammatic Monte Carlo studies of strong correlations.

Significance. If the reported kernel speedups carry over to production workloads, the result is significant: it would substantially lower the cost of high-order diagrammatic expansions, especially on consumer GPUs, and would make the CoS algorithm practical for a wider range of models. The flattening transformation, the window-based parallel edge evaluation, and the block-grid distribution across configurations are clearly described and constitute a useful engineering contribution. The paper also separates the effect of flattening (Fig. 3) from the GPU parallelization (Figs. 5-6), which is a methodologically helpful distinction. However, the main evidence is a microbenchmark of isolated graph evaluations; no end-to-end DiagMC calculation or raw timing data is provided, and the numerical-stability test is very limited. The speedup claim is therefore plausible but not yet fully demonstrated for the workloads the paper motivates.

major comments (4)
  1. [Section III, Figs. 5 and 6] The central performance claim is based on a microbenchmark that measures isolated graph evaluations per second with propagators and interaction values already available in memory. In a full DiagMC calculation, each Monte Carlo move changes vertex coordinates and requires recomputation or interpolation of O(n^2) propagator and interaction values before the graph kernel is launched, followed by Metropolis acceptance logic and host-device communication. None of these costs are measured. Because the abstract and introduction frame the result as an acceleration of DiagMC computation time, the paper needs either an end-to-end DiagMC benchmark or a careful restatement of the claims as applying only to the graph-evaluation kernel.
  2. [Section III, Figs. 5 and 6] The timing methodology is underreported. The figures show no error bars, no statement of the number of repetitions, no compiler or optimization flags, no CPU model for the baseline, and no clarification of whether the Ref. [10] CPU baseline used single or double precision. Since the headline is a quantitative speedup of three orders of magnitude and an order increase of about eight, these details are necessary to assess the robustness and fairness of the comparison; if the baseline used double precision, the comparison would conflate algorithmic/hardware acceleration with precision reduction.
  3. [Section III, Numerical stability] The single-precision stability test is limited to one configuration at order 10, with no definition of the error metric, no statistics over multiple configurations, and no check at higher orders. The paper nevertheless recommends single-precision arithmetic throughout and uses it for all benchmark results. Given that the speedup partly relies on the reduced memory footprint of 32-bit data, this evidence is too thin to support the claim that single-precision errors are negligible across the range of orders where the GPU speedup is reported.
  4. [Section III, 'Graph evaluations per second'] The statement that during the time the original algorithm sums all diagram integrands for order n, the hardware-accelerated flattened CoS evaluates the series for an order greater by approximately 8 is not derived from the data shown. The acceleration factor in Fig. 6 is strongly n-dependent, so the order offset between the curves in Fig. 5 is not a single constant. Please specify how this number is extracted, over which n range it holds, and how sensitive it is to the choice of hardware and block/window size.
minor comments (6)
  1. [Section I, Introduction] The sentence 'provided they and based on simple and invariable instructions' is ungrammatical and should be rewritten.
  2. [Fig. 3] The axes of Fig. 3 are not labelled; please specify the measured quantity, the hardware and precision used, and whether each point is a single measurement or an average.
  3. [Section III, 'Graph evaluations per second'] The text says 'This is reflected in the inset of Fig. 2', but the FLOPS inset appears in Fig. 5; please correct the cross-reference.
  4. [Section II C, Windowing] The block/window size is the main tunable parameter of the implementation, but the paper does not report the actual window sizes used for each order and GPU; please provide this information for reproducibility.
  5. [Section IV, Discussion] The text contains the typo 'FGPAs'; this should read 'FPGAs'.
  6. [General] The manuscript does not state whether the benchmark code or raw timing data will be made available; for a quantitative performance claim of this magnitude, providing them would substantially strengthen reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the GPU speedup is a directly measured benchmark against the authors' own prior CPU code; the self-citation to Ref. [10] supplies the ported algorithm and baseline, not the conclusion.

full rationale

The paper's central claim is an engineering benchmark, not a derivation: the three-orders-of-magnitude speedup (Figs. 5-6) is the measured ratio of graph evaluations per second between the authors' GPU-flattened CoS implementation and the original CPU implementation of Ref. [10] on the same task. No parameter is fitted to data and then renamed a prediction; the 'order greater by approximately 8' statement is simply the measured horizontal offset between the two benchmark curves at equal evaluation rates. The self-citation to Ref. [10] (co-author Kozik, Nature Commun. 2024) supplies the CoS algorithm being ported and the baseline implementation being beaten; it does not assert the speedup result, which is freshly measured and reproducible, so it is independent, parameter-free support rather than load-bearing circularity. The flattening transformation is benchmarked separately on the CPU (Fig. 3), isolating the implementation gain from the parallelization gain. Caveats do exist: there is no end-to-end DiagMC calculation, the claim that Monte Carlo proposal/acceptance overhead 'provides very little overhead in our testing' is asserted without presenting that testing, and the numerical-stability test uses a single configuration. These limit how far the kernel-level speedup carries over to realistic workloads, but they are evidence-strength and generalizability concerns, not circularity. No equation in the paper is defined in terms of its own output, and no result reduces by construction to a fit.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The central speedup claim rests on the correctness and complexity of the CoS algorithm from Ref. [10], on the CUDA execution model, and on the benchmark choices (flattened graph, window/block sizing, FP32 arithmetic). No new physical entities or fitted physical parameters are introduced.

free parameters (1)
  • GPU block/window size = smallest multiple of 32 exceeding average edges per level, capped at 1024
    Chosen per graph order to balance occupancy and level coverage; the reported speedups depend on this tuning, and exact per-benchmark choices are not tabulated.
assumptions (5)
  • domain assumption CoS algorithm correctly sums all connected Feynman diagram integrands of order n in O(n^2 2^n) operations.
    Taken from Ref. [10] without re-derivation; all GPU results assume this algorithm is correct.
  • domain assumption Graph structure is independent of vertex coordinates and can be stored once for many Monte Carlo configurations.
    Section II.B and II.C; this property is essential to the proposed parallel evaluation strategy.
  • domain assumption CUDA SIMT execution, memory hierarchy, and coalescing behave as described, so the flat edge array yields the claimed parallel speedups.
    Section II.C and Appendix A; hardware behavior is assumed rather than independently verified.
  • ad hoc to paper Benchmark baseline 'Ref. [10] CPU' is a fair state-of-the-art comparison.
    The comparison depends on the authors' prior code and is not externally benchmarked; if that baseline is suboptimal, the speedup factor would be inflated.
  • domain assumption Single-precision FP32 error of about 10^-5 percent on one repeated order-10 configuration generalizes to production DiagMC runs.
    Section III 'Numerical stability'; the test does not cover long Markov chains or varying configurations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploiting Parallelism for Fast Feynman Diagrammatics." pith.science (2026). https://pith.science/paper/WMRJAOW5

@misc{pith2026250100675,
  author       = {Pith},
  title        = {Pith review of: Exploiting Parallelism for Fast Feynman Diagrammatics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WMRJAOW5}},
  note         = {Machine review of arXiv:2501.00675}
}
read the original abstract

Diagrammatic expansions are a paradigmatic and powerful tool of quantum many-body theory. Their evaluation to high order, e.g., by the Diagrammatic Monte Carlo technique, can provide unbiased results in strongly correlated and challenging regimes. However, calculating a factorial number of terms to acceptable precision remains very costly even for state-of-the-art methods. We achieve a dramatic acceleration of evaluating Feynman's diagrammatic series by use of specialised hardware architecture within the recently introduced combinatorial summation (CoS) framework. We present how exploiting the massive parallelism and concurrency available from GPUs leads to orders of magnitude improvement in computation time even on consumer-grade hardware. This provides a platform for making probes of novel phenomena of strong correlations much more accessible.

Figures

Figures reproduced from arXiv: 2501.00675 by the authors.

Figure 1
Figure 1. FIG. 1: Illustration of the technique for combinatorial summation (CoS) of the integrands of all connected Feynman [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Illustration of the graph flattening transformation for parallel processing. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3: Speed-up seen from transforming the graph to [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: FIG. 4: Illustration of how a window iterates through the flattened-representation of the graph. If the size of the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5: Number of evaluations of the sum of all [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6: Acceleration factor achieved on the specified [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7: Number of connected Feynman diagrams [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 30 canonical work pages

  1. [10]

    Prokof’ev and B

    N. Prokof’ev and B. Svistunov, Bold Diagrammatic Monte Carlo Technique: When the Sign Problem Is Wel- come, Phys. Rev. Lett. 99, 250201 (2007)

    work page 2007

  2. [1]

    Single Instruction, Single Data

    Execution Model In the typical model of computation, a set of instruc- tions (e.g. arithmetic operations) are executed in a single sequential stream known as a thread. Each thread there- fore processes one instruction at a time over a single el- ements of a set of data, leading to the common name “Single Instruction, Single Data” (SISD). Most modern CPUs ...

    work page

  3. [2]

    bare metal

    SMs and Occupancy The CUDA programming model provides several lay- ers of abstraction from the “bare metal” hardware, but it is crucial for good performance to keep in mind the underlying architecture. Device kernels require, at min- imum, the number of threads requested for each block, and how many of these independent blocks to launch. Due to the physic...

    work page

  4. [3]

    Synchronicity Thread execution within each warp occurs at the same time - however with branching and conditional code there may be additional synchronisation involved. Each warp within a block is not guaranteed to execute in order of their ‘index’, and neither is there any guarantee over exe- cution order between blocks - unless the programmer im- plement...

    work page

  5. [4]

    constant

    Memory Hierarchies There exists two main addressable regions of memory on CUDA devices. Firstly, global memory (often referred to as GMEM). This is the main bulk of memory avail- able on the GPU, when cards are advertised as having a capacity of 48GB, for example, it is the GMEM which is being talked about. GMEM is addressable from all threads on the devi...

    work page

  6. [5]

    super-chips

    Host-Device Latency Just as with memory transactions, we want to min- imise long-distance or high-latency communications, and this includes the Host-Device channel. Most systems in- 10 terface the host and device with PCIe connections, which while suitable for most graphics use cases can be a signif- icant bottleneck for scientific applications which requ...

    work page

  7. [6]

    Struct of Arrays

    Cache and Locality The message of the previous sections is that data lo- cality and re-use are crucial to the performance of GPU algorithms. In order to write fast and energy-efficient code we therefore need to be mindful of our memory ac- cess patterns and how this translates to cache utilisation. It is best practice to layout memory in a manner that sat...

    work page

  8. [7]

    E. Y. Loh, J. E. Gubernatis, R. T. Scalettar, S. R. White, D. J. Scalapino, and R. L. Sugar, Sign problem in the nu- merical simulation of many-electron systems, Phys. Rev. B 41, 9301 (1990)

    1990

Show all 40 references
  1. [8]

    Troyer and U.-J

    M. Troyer and U.-J. Wiese, Computational complexity and fundamental limitations to fermionic quantum monte carlo simulations, Phys. Rev. Lett. 94, 170201 (2005)

    2005

  2. [9]

    N. V. Prokof’ev and B. V. Svistunov, Polaron Problem by Diagrammatic Quantum Monte Carlo, Phys. Rev. Lett. 81, 2514 (1998)

    1998

  3. [11]

    Van Houcke, E

    K. Van Houcke, E. Kozik, N. Prokofev, and B. Svis- tunov, Diagrammatic Monte Carlo, Physics Procedia 6, 95 (2010)

    2010

  4. [12]

    Kozik, K

    E. Kozik, K. Van Houcke, E. Gull, L. Pollet, N. Prokof’ev, B. Svistunov, and M. Troyer, Diagram- matic Monte Carlo for correlated fermions, Europhys. Lett. 90, 10004 (2010)

    2010

  5. [13]

    A. A. Abrikosov, L. P. Gorkov, and I. Y. Dzyaloshin- skii, Quantum Field Theoretical Methods in Statistical Physics., Vol. 51 (1967) pp. 183–184

    1967

  6. [14]

    R. E. V. Profumo, C. Groth, L. Messio, O. Parcollet, and X. Waintal, Quantum monte carlo for correlated out- of-equilibrium nanoelectronic devices, Phys. Rev. B 91, 245154 (2015)

    2015

  7. [15]

    Rossi, Determinant Diagrammatic Monte Carlo Al- gorithm in the Thermodynamic Limit, Phys

    R. Rossi, Determinant Diagrammatic Monte Carlo Al- gorithm in the Thermodynamic Limit, Phys. Rev. Lett. 119, 045701 (2017)

    2017

  8. [16]

    Kozik, Combinatorial summation of Feynman dia- grams, Nature Communications 15, 7916 (2024)

    E. Kozik, Combinatorial summation of Feynman dia- grams, Nature Communications 15, 7916 (2024)

    2024

  9. [17]

    Rossi, N

    R. Rossi, N. Prokof’ev, B. Svistunov, K. Van Houcke, and F. Werner, Polynomial complexity despite the fermionic sign, EPL (Europhysics Letters) 118, 10004 (2017)

    2017

  10. [18]

    A. J. Kim, N. V. Prokof’ev, B. V. Svistunov, and E. Kozik, Homotopic action: A pathway to convergent diagrammatic theories, Phys. Rev. Lett. 126, 257001 (2021)

    2021

  11. [19]

    W. Wu, M. Ferrero, A. Georges, and E. Kozik, Control- ling Feynman diagrammatic expansions: Physical nature of the pseudogap in the two-dimensional Hubbard model, Phys. Rev. B 96, 041105(R) (2017)

    2017

  12. [20]

    ˇSimkovic, J

    F. ˇSimkovic, J. P. F. LeBlanc, A. J. Kim, Y. Deng, N. V. Prokof’ev, B. V. Svistunov, and E. Kozik, Ex- tended crossover from a fermi liquid to a quasiantiferro- 11 magnet in the half-filled 2d hubbard model, Phys. Rev. Lett. 124, 017003 (2020)

    2020

  13. [21]

    A. J. Kim, F. Simkovic, and E. Kozik, Spin and charge correlations across the metal-to-insulator crossover in the half-filled 2d hubbard model, Phys. Rev. Lett. 124, 117602 (2020)

    2020

  14. [22]

    Lenihan, A

    C. Lenihan, A. J. Kim, F. ˇSimkovic IV., and E. Kozik, Entropy in the non-fermi-liquid regime of the doped 2D hubbard model, Phys. Rev. Lett. 126, 105701 (2021)

    2021

  15. [23]

    imkovic, R

    F. imkovic, R. Rossi, A. Georges, and M. Ferrero, Origin and fate of the pseudogap in the doped hubbard model, Science 385, eade9194 (2024), https://www.science.org/doi/pdf/10.1126/science.ade9194

    2024 doi

  16. [24]

    Lenihan, A

    C. Lenihan, A. J. Kim, F. ˇSimkovic, and E. Kozik, Eval- uating Second-Order Phase Transitions with Diagram- matic Monte Carlo: N´ eel Transition in the Doped Three- Dimensional Hubbard Model, Physical Review Letters 129, 107202 (2022)

    2022

  17. [25]

    Garioud, F

    R. Garioud, F. ˇSimkovic, R. Rossi, G. Spada, T. Sch¨ afer, F. Werner, and M. Ferrero, Symmetry-broken perturba- tion theory to large orders in antiferromagnetic phases, Phys. Rev. Lett. 132, 246505 (2024)

    2024

  18. [26]

    Chen and K

    K. Chen and K. Haule, A combined variational and dia- grammatic quantum Monte Carlo approach to the many- electron problem, Nat. Commun. 10, 3725 (2019)

    2019

  19. [27]

    P. Hou, T. Wang, D. Cerkoney, X. Cai, Z. Li, Y. Deng, L. Wang, and K. Chen, Feynman diagrams as computa- tional graphs (2024), arXiv:2403.18840 [hep-th]

    2024 arXiv

  20. [28]

    C. C. Chang, S. Gogolenko, J. Perez, Z. Bai, and R. T. Scalettar, Recent advances in determinant quan- tum Monte Carlo, Philosophical Magazine 95, 1260 (2015)

    2015

  21. [29]

    Menczer, K

    A. Menczer, K. Kap´ as, M. A. Werner, and . Legeza, Two- dimensional quantum lattice models via mode optimized hybrid CPU-GPU density matrix renormalization group method, Physical Review B 109, 195148 (2024)

    2024

  22. [30]

    K. A. Wendt, J. E. Drut, and T. A. L¨ ahde, Toward large- scale Hybrid Monte Carlo simulations of the Hubbard model on graphics processing units, Computer Physics Communications 182, 1651 (2011)

    2011

  23. [31]

    Siro and A

    T. Siro and A. Harju, Exact diagonalization of the Hubbard model on graphics processing units, Computer Physics Communications 183, 1884 (2012)

    2012

  24. [32]

    Melnick, P

    C. Melnick, P. S´ emon, K. Yu, N. D’Imperio, A. M. Tremblay, and G. Kotliar, Accelerated impurity solver for DMFT and its diagrammatic extensions, Computer Physics Communications 267, 108075 (2021)

    2021

  25. [33]

    J. A. Anderson, C. D. Lorenz, and A. Travesset, Gen- eral purpose molecular dynamics simulations fully im- plemented on graphics processing units, Journal of Com- putational Physics 227, 5342 (2008)

    2008

  26. [34]

    We are also witnessing the advent of unified CPU- GPU ‘superchips’ which continue to erode the current downsides of hardware acceleration (limited memory and throughput between host and device) [34]

  27. [35]

    A. Lee, C. Yau, M. B. Giles, A. Doucet, and C. C. Holmes, On the Utility of Graphics Cards to Per- form Massively Parallel Simulation of Advanced Monte Carlo Methods, Journal of Computational and Graphical Statistics 19, 769 (2010)

    2010

  28. [36]

    AMD, High Performance Computing Tuning Guide for AMD EPYC 9004 Series Processors (2024)

    2024

  29. [37]

    NVIDIA, Nvidia H100 Tensor Core GPU Datasheet (2024)

    2024

  30. [38]

    N´ u˜ nez Fern´ andez, M

    Y. N´ u˜ nez Fern´ andez, M. Jeannin, P. T. Dumitrescu, T. Kloss, J. Kaye, O. Parcollet, and X. Waintal, Learning Feynman Diagrams with Tensor Trains, Physical Review X 12, 041018 (2022)

    2022

  31. [39]

    M¨ uller, F

    T. M¨ uller, F. Rousselle, J. Nov´ ak, and A. Keller, Real- time neural radiance caching for path tracing, ACM Trans. Graph. 40, 10.1145/3450626.3459812 (2021)

    2021

  32. [40]

    Kenyon and C

    C. Kenyon and C. Capano, Apple Silicon Perfor- mance in Scientific Computing, 2022 IEEE High Per- formance Extreme Computing Conference, HPEC 2022 10.1109/HPEC55821.2022.9926315 (2022)

    2022

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.