Introduction

Megahit is a de-novo assembler for NGS metagenomics. It is an effective alternative to diamond and bowtie. Here we will look how long it takes to assemble the pair-genome using the version of 1.2.9 of megahit.

Methodology

We will run both single and concurrent runs on

Rome (Dual EPYC 7552 with 48 cores each and 512GB DDR4 RAM combined) and

Turin  (Single EPYC 9655 with 96 cores and 768GB DDR5 RAM) CPUs

Code Block (bash)

module load megahit/1.2.9
r1=/pasteur/appa/scratch/test_scal/test_PE_R1.fastq
r2=/pasteur/appa/scratch/test_scal/test_PE_R2.fastq
megahit -1 ${r1} -2 ${r2} -o megahit_output --num-cpu-threads ${XX} --min-contig-len 100

Results

Threads	Turin	Rome	RAM
96	10h28m	26h	26G
48	11h35m	21h47	26G
24	17h34m	36h	26G
24x4	21h32m	too long	4x26G
32x3	17h	31h	3x26G

Conclusions:

Turin is about twice faster than Rome. Do not use more than 48 threads on any of them. Use 48 on Rome to fit into the 24h queue.