Ungapped prefilter died during GPU-accelerated search

[TheChosenOneJG]

When I ran my scrpit below to generate MSA against uniref90 by MMseqs2 GPU-accelerated searching, it reported an error as follows. However, when I replaced the targetDB uniref90 with a smaller one (consisit of about thousands of sequences), such error would not appear. My Linux system contains:

• 40 CPUs,
• 400+ GB ram,
• 4 GPUs,
• enough storage.

Could you please help me figure this out? Thanks in advance for your expert help.

Error

ungappedprefilter /a100_nas/ai4s/MSA/queries/testDB /a100_nas/ai4s/MSA/uniref90DB_gpu.idx /a100_nas/ai4s/MSA/tmp/15602816422822286028/pref_0 --sub-mat 'aa:blosum62.out,nucl:nucleotide.out' -c 0 -e 0.001 --cov-mode 0 --comp-bias-corr 1 --comp-bias-corr-scale 1 --min-ungapped-score 15 --max-seqs 300 --db-load-mode 2 --gpu 1 --gpu-server 1 --prefilter-mode 1 --threads 42 --compressed 0 -v 3 

Index version: 16
Generated by:  16.747c6
ScoreMatrix:  VTML80.out
--gpu-server /dev/shm/8478586279687262130 does not existError: Ungapped prefilter died

part of my script

mmseqs createdb $UNIREFFASTA $UNIREFDB
mmseqs makepaddedseqdb $UNIREFDB $UNIREFDB_GPU

QUERY_DB="${QUERY_DIR}/testDB"
mmseqs createdb $QUERY $QUERY_DB

# searching with GPU 
mmseqs createindex $UNIREFDB_GPU $TMP --index-subset 2
mmseqs gpuserver $UNIREFDB_GPU --gpu 1 &
PID=$!
mmseqs search $QUERY_DB $UNIREFDB_GPU $RESULTS/aln_test $TMP --gpu 1 --gpu-server 1 --db-load-mode 2 --remove-tmp-files 1 --max-seq 10000
kill $PID

[unknow1024]

hello, it seems i have the same problem with you.I wanna know that if you used the command mmseqs makepaddedseqdb for the targetdb and it returns the segment error?

[TheChosenOneJG]

hello, it seems i have the same problem with you.I wanna know that if you used the command mmseqs makepaddedseqdb for the targetdb and it returns the segment error?

Hello, I definitely used mmseqs makepaddedseqdb command, and it succeeded and no error occurred. How large is your targetDB?

[unknow1024]

Nearly 43GB.It is uniclust30-hhsuite and used mmseqs createdb to get the targetdb.