Scalable Many-Core Algorithms for Tridiagonal Solvers

Balogh, Daniel; Flynn, Toby; Laizet, Sylvain; Mudalige, Gihan R.; Reguly, István Z.

doi:10.1109/mcse.2021.3130544

Cited by 1 publication

(2 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, we compare performance on the FPGA to an Nvidia Tesla V100 GPU using the tridiagonal solver library, tridsolver implemented by László et al [2,15] using its batched version presented by Reguly et al [22]. This GPU library has been shown [7] to provide matching or better performance than the two current batch tridiagonal solver functions in Nvidia's cuS-PARSE library [6, 27] ś cusparse<t>gtsv2StridedBatch() and cusparse<t>gtsvInterleavedBatch(). Our experiments also confirmed these results for the applications evaluated in this paper.…”

Section: Performance Evaluationmentioning

confidence: 99%

See 1 more Smart Citation

High throughput multidimensional tridiagonal system solvers on FPGAs

Kamalakkannan

Mudalige

Reguly

et al. 2022

Proceedings of the 36th ACM International Conference on Supercomputing

Self Cite

View full text Add to dashboard Cite

We present a high performance tridiagonal solver library for Xilinx FPGAs optimized for multiple multi-dimensional systems common in real-world applications. An analytical performance model is developed and used to explore the design space and obtain rapid performance estimates that are over 85% accurate. This library achieves an order of magnitude better performance when solving large batches of systems than previous FPGA work. A detailed comparison with a current state-of-the-art GPU library for multidimensional tridiagonal systems on an Nvidia V100 GPU shows the FPGA achieving competitive or better runtime and significant energy savings of over 30%. Through this design, we learn lessons about the types of applications where FPGAs can challenge the current dominance of GPUs. CCS CONCEPTS• Computer systems organization → Reconfigurable computing; Multicore architectures; • Mathematics of computing → Mathematical software performance.

show abstract

Section: Performance Evaluationmentioning

confidence: 99%

“…Note also that we have selected the number of interleaved systems and interleaved reduced systems to be equal (i.e. 𝑔 = 𝑔 𝑟 in relation to (7)). The final term in (15) and ( 16) are the latencies for processing a batch of B systems.…”

mentioning

confidence: 99%