Skip to content

Latest commit

 

History

History
42 lines (31 loc) · 1.96 KB

rocm-bin.md

File metadata and controls

42 lines (31 loc) · 1.96 KB

Basecalling on AMD GPUs

With slorado, now you can do some nanopore basecalling on AMD GPUs. We have some compiled binaries which should work on Linux if you have GLIBC >= 2.28 (invoke ldd --version to check). You should also have the GPU driver installed. Currently, we have tested the binaries only on systems in Note 1. If you encounter a problem feel free to open an issue.

Getting Started

First, download and extract the slorado rocm Linux binaries tarball. Note that this is for testing only and the link will not be persistent!!!!

VERSION=v0.2.0-beta
wget "https://github.com/BonsonW/slorado/releases/download/$VERSION/slorado-$VERSION-x86_64-rocm-linux-binaries.tar.gz"
tar xvf slorado-$VERSION-x86_64-rocm-linux-binaries.tar.gz
cd slorado-$VERSION
bin/slorado --help

Download the test dataset with 20,000 reads and run slorado:

wget -O PGXXXX230339_reads_20k.blow5 https://slow5.bioinf.science/hg2_prom_5khz_subsubsample
./bin/slorado basecaller models/[email protected] PGXXXX230339_reads_20k.blow5  -o out.fastq -x cuda:all

Refer to troubleshoot for help on resolving common problems.

Optional Testing

Test if the output maps and identity scores are good (required minimap2, the human genome and datamash):

minimap2 -cx map-ont hg38noAlt.fa out.fastq --secondary=no -t16  | awk '{print $10/$11}' | datamash mean 1 median 1 count 1

It should print the mean identity score, median identity score and the number of alignments. The numbers are expected to be close to the following (would not be identical due to floating point deviations):

0.94113261325097        0.9771185       24768

Note 1

Currently, we have tested these binaries on following systems:

  1. O/S: SUSE Linux Enterprise Server 15; GPU: Instinct MI250X
  2. O/S: Ubuntu 20; GPUs: Instinct MI50/MI60, Instinct MI100 and Instinct MI210
  3. O/S: Ubuntu 22; GPU: Instinct MI210