NVIDIA GPGPU with Abaqus 2017

A review of our benchmark results of Abaqus 2017 with the new nVIDIA GP100

After a lot of interest was expressed from our 2016 GPGPU studies of Abaqus with AMD and NVIDIA hardware , including our presentation at Science in the Age of Experience and our webinar, we decided to pursue our studies further with the newest versions of Abaqus and the latest and greatest hardware from NVIDIA.

We leave our AMD FirePro W9100 and NVIDIA Tesla K40 behind and continue our journey into GPU Computing with Abaqus 2017 and NVIDIA's latest Pascal-based GPU, the Quadro GP100. The GP100 specs speak for themselves, this is a hybrid monster half-GPU half-accelerator. Best of both worlds for sure.

nVIDIA GP100 Quick Specs


This time around, we will focus on a more substantial model, a 26M degree of freedom of a jet engine turbine. The analysis includes a typical turbine pressure and temperature non-linear statics analysis.




The tests will be performed on one of our workstations (Dell T7910, dual 6-core Xeon with 128GB RAM running Windows 10). As we traditionally do with these tests, we studied several combinations of number of cores, with and without GPU. In this particular test, we will use 4-core, 8-core and 12-core combinations. As usual, we notice good scalability of the Abaqus solver, around 30% solution time decrease as the core count increases. Furthermore, the addition of the NVIDIA Quadro GP100 brings a systematic 45%-50% decrease in solution time across all configurations as we can see on the graph below:

NVIDIA Quadro GP100 Abaqus benchmark

It is also interesting to notice the benefits of GPGPU while comparing a 4-core solve with GPU to an 8-core solve without: the 4-core + GPU outperforms the 8-core solution by around 33%. This is new data as our previous tests with Quadro K4000 showed slightly lower performance, while our tests with Tesla K40 showed an increase of performance of around 10% over a double core count. This additional 25% boost in performance can certainly be attributed to the superior NVIDIA Quadro GP100 hardware.




blog comments powered by Disqus