GPU deconvolution: impressive Huygens GPU performance with NVIDIA's brand-new Volta-based Tesla V100 card

Thursday 12 of October, 2017
SVI has recently performed GPU deconvolution benchmarks for several 3D image dimensions with three state-of-the-art GPU cards.
The brand-new NVidia Volta-based Tesla V100 shows an impressive acceleration compared to the the (Pascal) P100 and GTX Titan Xp cards, which had already demonstrated extremely fast deconvolution performance in previous Huygens benchmark tests.

The acceleration factor shown in the chart below is relative to a 6 core (12 threads) Intel(R) Xeon(R) CPU E5-1650 v4 @ 3.60GHz.
The acceleration factor was determined by comparing the average time per iteration over a total of 100 iterations using the Huygens CMLE algorithm.

500x500x40 1000x1000x50 1000x1000x100 1600x1600x100 1600x1600x200 0 5 10 15 20 25 Huygens deconvolution GPU acceleration *Acceleration relative to Intel(R) Xeon(R) CPU E5-1650 v4 @ 3.60GHz. Higher is better V100 P100 GTX Titan Xp Image pixel dimension (X * Y * Z) Acceleration factor * NVIDIA GPU cards: www.svi.nl