Summary

  • GPU-only pipelines complete sequence analysis workflows ~22 to ~60 times faster than CPU-only pipelines

  • Performance of the GPU-only pipeline vary depending on the available GPU platform

    • Pipeline run ~2.6x faster in A100 compared to GeForce-RTX 3090

  • GPU-only and CPU-only pipelines both agrees on the DNA-changes detected in a given sample

    • Over 99.2% DNA changes detected on a give sample are identified in both GPU- and CPU-only pipelines.

  • Source code is not public and software is available in for use in the non-commercial settings (commercial applications need to go for a paid license agreement ).

Concerns

  • Source code is not public

    • Distributed as container image. Which means we can not make any changes to the wrappers and must use it as a whole.

    • When we got stuck in several occasions due to resource limitation we had to spend a lot of time figuring out solutions

    • We don’t have the option to compile the software on a GPU platform (i.e, Tesla M60) and fix cuda-compatibility issues from our-side (we have to depend on NVIDIA - Parabricks developers)

  • Software is available for use in the non-commercial settings.