Not long now until our Long-Read Research Symposium on November 13th. See below the amazing line-up of speakers and talks. It's not too late to register for this free event - click the link --> www.eventbrite.com.au/e/garvan-lon...

A more pretty one from a while back

only the first time ;)
Once you have a good template set up, things are super easy. My stuff is mostly screenshots and diagrams with basic text anyway, explaining algorithms, pipelines, data structures, or pretty annotated squiggles/sequences

(here is a not so pretty example)

And, you can still do pretty things with it

I built and installed this absolute beast of a machine for
@nanoporetech.com
basecalling
@petermaccc.bsky.social

Threadripper
4x5090, liquid cooled
128gb DDR5 ram
~40Tb of nvme and SSD storage
TWO power supplies, a 1200W and a 2500W
All in a Corsair 9000D case

~5^7 samples/second with dorado.

I think I hit a new high score for basecalling speed on a 5090 machine with Dorado v1.0.2 and the v5.2.0 sup model

1.79e+07 samples/s

That's ~123gbp/day

Wow!
@nanoporetech.com have done a great job with the speedups.

We also implemented our own chunk method using intermediate files and a merge step that was very performant, and would fit in perfectly with their file writing/staging structure during a run.
Honeslty, slow5 is just a better fit for purpose format overall, and this paper shows that with evidence.

Just casually doing some SUP basecalling on an AMD 7900XTX GPU 😅 with ROCm in a workstation at home.

A work in progress by Hasindu and Bonson in the lab.