Solving Lattice QCD systems of equations using mixed precision solvers on GPUs
read the original abstract
Modern graphics hardware is designed for highly parallel numerical tasks and promises significant cost and performance benefits for many scientific applications. One such application is lattice quantum chromodyamics (lattice QCD), where the main computational challenge is to efficiently solve the discretized Dirac equation in the presence of an SU(3) gauge field. Using NVIDIA's CUDA platform we have implemented a Wilson-Dirac sparse matrix-vector product that performs at up to 40 Gflops, 135 Gflops and 212 Gflops for double, single and half precision respectively on NVIDIA's GeForce GTX 280 GPU. We have developed a new mixed precision approach for Krylov solvers using reliable updates which allows for full double precision accuracy while using only single or half precision arithmetic for the bulk of the computation. The resulting BiCGstab and CG solvers run in excess of 100 Gflops and, in terms of iterations until convergence, perform better than the usual defect-correction approach for mixed precision.
This paper has not been read by Pith yet.
Forward citations
Cited by 9 Pith papers
-
Complete lattice QCD calculation of $K^{-}\to \ell^{-}\bar{\nu}_{\ell}\ell^{'+}\ell^{'-}$ form factors
First complete lattice QCD determination of the four structure-dependent form factors for K- → ℓ- ν̄ℓ ℓ'+ ℓ'- decays at physical quark masses with controlled statistical and systematic errors.
-
Reconstructing the full kinematic dependence of GPDs from pseudo-distributions
Lattice QCD pseudo-distributions at m_π=358 MeV are inverted via multidimensional Gaussian process regression to reconstruct the full kinematic dependence of GPDs H^{u-d} and E^{u-d} while directly extracting double d...
-
Charmonium radiative transitions to dileptons from lattice QCD: The case of $h_c \to \eta_c \ell^+\ell^-$ and $\chi_{c1} \to J/\psi\,\ell^+\ell^-$
First fully dynamical lattice QCD yields Γ(h_c → η_c e⁺e⁻) = 5.45(19) keV (3σ above BESIII) and Γ(χ_c1 → J/ψ e⁺e⁻) = 2.869(90) keV, with continuum-extrapolated results and q² distributions.
-
Electromagnetic form factors and structure of the $T_{bb}$ tetraquark from lattice QCD
Lattice QCD on one ensemble yields electromagnetic form factors for T_bb, indicating a compact heavy diquark plus light antidiquark bound state with charge radius smaller than the BB* threshold.
-
Scalar and Tensor Form Factors for $\Lambda \rightarrow p\ell \bar{\nu}_\ell$ from Lattice QCD
Lattice QCD yields the scalar and tensor form factors for Λ→pℓν̄ℓ as functions of q², providing a model-independent input to constrain non-standard charged-current interactions via the predicted R^{μe} ratio compared ...
-
The Lambda 1405 at the $SU(3)$ point in lattice QCD
Lattice QCD at the SU(3) symmetric point extracts energy levels for singlet and octet baryon-meson channels to inform the two-pole structure of Lambda(1405).
-
$D_1$ and $D_2$ resonances in coupled-channel scattering amplitudes from lattice QCD
Lattice QCD at m_π≈391 MeV finds D1 bound state below D*π threshold strongly coupled in S-wave and D1' resonance in elastic D*π region for I=1/2 charmed channels.
-
$F_K/F_\pi$ as a precision test of a new four flavor Domain Wall Fermion action
New four-flavor smeared Möbius Domain Wall Fermion ensembles yield F_K/F_pi = 1.1962(34) as a precision test for inexpensive chiral fermion calculations in lattice QCD.
-
Inverse problem in the LaMET framework
Analysis of non-perturbative lattice data shows that the inverse problem in LaMET introduces significant uncertainties in parton distributions, especially from harmonics around λ=5-15, and that exact asymptotic decay ...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.