What is Faster - CPU, GPU or FPGA? Our Investigation of Leading Alignment Algorithms

Let us give the punchline so that impatient readers can move on to doing other things. **The question - ‘which technology among CPU, GPU and FPGA is faster’

• is too simplistic and cannot be easily answered.** The details will follow in several commentaries.

Over the last few days, we have been studying various alignment algorithms in full depth (i.e. at the implementation level) to understand relative merits of various hardware technologies. For GPU, we picked Ruibang’s SOAP3-dp paper recently published in PLOS-One. For CPU, we studied Heng Li’s BWA-MEM paper recently ‘unpublished’ in arxiv. For FPGA, we chose Shepard accepted in Proceedings of the International Conference on Formal Methods and Models for Codesign (MEMOCODE), 2012. Also we carefully studied all relevant earlier papers to understand where various algorithmic ideas came from. For example, BWA-MEM paper provides very little details on the algorithm, whereas Heng Li’s Fermi paper covered them with gory details. We thank Heng and Ruibang for kindly replying to our email questions, when we got stuck.

Our discussion will constitute of several parts broken into multiple commentaries. Roughly we will cover the following topics.

(i) CPU vs GPU vs FPGA - a General Introduction of Three Technologies

This commentary provides you with a general introduction of the three hardware technologies without referring to any bioinformatics algorithm.

(ii) Philosophical Origin of Alignment Algorithms in BWA-MEM, SOAP3-dp and Shepard

We go into the details of three alignment algorithms and explain how the choice of hardware technology affected the implementation

(iii) Comparison of Implementation Details

This commentary includes back-of the-envelope comparison of speed, potential future speed, bottlenecks, etc.

(iv) Other Factors Such as Code Availability, Upgrade and Commoditization

If you want to play, do not forget that three worlds follow three different sets of rules. We discuss them here.

(v) Summary of Comparison

Sorry folks. You expected to hear terms like Burrows Wheeler transform, indels and clock speed, but we have to leave you with jargons such as ‘patent right’, ‘engineering’, ‘business decision’ and ‘total cost of ownership’.

-———————————–

The above points will be covered in several forthcoming commentaries.