QCD on GPUs: cost effective supercomputing

Clark, Michael A.

doi:10.22323/1.091.0003

Cited by 7 publications

(11 citation statements)

References 23 publications

(57 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To achieve high efficiency for the GPU computation the task assigned to the GPU should be as simple as possible and should contain high parallelism. The GPU computation has been applied to the quark solver and high efficiency has been achieved in the case of a single GPU computation using the mixed-precision (or flexible) preconditioner technique [4]. Based on these success we would like to assign one linear equation Dx = b to one GPU even if we have many GPUs.…”

Section: Gpu Implementationmentioning

confidence: 99%

“…By this modification the single-precision BiCGStab solvers on the GPUs are called in embarrassingly parallel in the preconditioning part of the Gl-BiCGStab. We follow the tuning techniques described in [4] for the GPU solver.…”

Section: Gpu Implementationmentioning

confidence: 99%

“…The application of GPU computing to lattice field theory has been introduced by [2] and has been studied extensively in recent years [3,4,5]. However the parallel GPU computations with field domain decomposition, which is usually employed for large-scale QCD simulations, are still very challenging because there is no efficient device for direct communication among GPUs.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Improving many flavor QCD simulation using multiple GPU's

Ishikawa¹

2011

Proceedings of the XXVIII International Symposium on Lattice Field Theory — PoS(Lattice 2010)

View full text Add to dashboard Cite

We accelerate many-flavor lattice QCD simulations using multiple GPUs. Multiple pseudofermion fields are introduced additively and independently for each flavor in the many-flavor HMC algorithm. Using the independence of each pseudo-fermion field and the blocking technique for the quark solver, we can assign the solver task to each GPU card. In this report we present the blocking technique for the many-flavor dynamical QCD simulations. We investigate the effect of the blocking and the acceleration with the multiple GPUs for the Schrödinger functional simulations with Wilson SU(3) plaquette gauge action and N f = 10 Wilson fermions. Five pseudo-fermion fields are introduced and the quark solver task is distributed in the ratio of 2:3 to two GPUs. We expect a 40% timing reduction from the single GPU case and have observed a 34% timing reduction in the test simulations.

show abstract

Section: Gpu Implementationmentioning

confidence: 99%

Section: Gpu Implementationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Improving many flavor QCD simulation using multiple GPU's

Ishikawa¹

2011

Proceedings of the XXVIII International Symposium on Lattice Field Theory — PoS(Lattice 2010)

View full text Add to dashboard Cite

show abstract

“…Thus, GPUs are quickly being adopted in a myriad of fields, ranging from scientific workload processing [22] to education in the developing world [23]. These architectures present opportunities to explore the utility of GPUs towards improving storage system reliability.…”

Section: Introductionmentioning

confidence: 99%

On the Use of GPUs in Realizing Cost-Effective Distributed RAID

Khasymski

Rafique

Butt

et al. 2012

2012 IEEE 20th International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems

View full text Add to dashboard Cite

Abstract-The exponential growth in user and application data entails new means for providing fault tolerance and protection against data loss. High Performance Computing (HPC) storage systems, which are at the forefront of handling the data deluge, typically employ hardware RAID at the backend. However, such solutions are costly, do not ensure end-to-end data integrity, and can become a bottleneck during data reconstruction. In this paper, we design an innovative solution to achieve a flexible, faulttolerant, and high-performance RAID-6 solution for a parallel file system (PFS). Our system utilizes low-cost, strategically placed GPUs -both on the client and server sides -to accelerate parity computation. In contrast to hardware-based approaches, we provide full control over the size, length and location of a RAID array on a per file basis, end-to-end data integrity checking, and parallelization of RAID array reconstruction. We have deployed our system in conjunction with the widely-used Lustre PFS, and show that our approach is feasible and imposes acceptable overhead.

show abstract

“…The rapidly increasing performance of graphics processing units (GPUs) and the introduction of CUDA that improves the programmability of GPUs for scientific computation has led to a significant effort devoted to efficient numerical computation for lattice quantum chromodynamics (LQCD) using GPUs [1]. LQCD is the method of choice for nonperturbative calculations of the effects of Nature's strong force.…”

Section: Introductionmentioning

confidence: 99%

The Fat-Link Computation on Large GPU Clusters for Lattice QCD

Shi

Babich

Clark

et al. 2012

2012 Symposium on Application Accelerators in High Performance Computing

View full text Add to dashboard Cite

Graphics Processing Units (GPU) are becoming increasingly popular in high performance computing due to their high performance, high power efficiency and low cost. In this paper, we present results of an effort to implement the fatlink computation -an important component of many lattice quantum chromodynamics (LQCD) calculations -on GPU clusters using the QUDA framework. Two implementations, one similar to the original CPU algorithm in the MILC code and one based on the idea of reduced communication by redundant computations, are presented and their relative advantages are discussed. In strong-scaling tests on up to 384 GPUs on Longhorn and 256 GPUs on Keeneland GPU clusters, where the CPU core to GPU ratio is 4:1 in both clusters, we achieved up to 11.4x and 8.7x node speedup when running on the two GPU clusters, respectively.

show abstract

QCD on GPUs: cost effective supercomputing

Cited by 7 publications

References 23 publications

Improving many flavor QCD simulation using multiple GPU's

Improving many flavor QCD simulation using multiple GPU's

On the Use of GPUs in Realizing Cost-Effective Distributed RAID

The Fat-Link Computation on Large GPU Clusters for Lattice QCD

Contact Info

Product

Resources

About