Closed riyuebao closed 3 years ago
Hi @riyuebao, this is unfortunately intentional behavior at the moment. STAR with multiple threads utilizes OpenMP
for multithreading, and this library can create issues when run on older hardware that does not support AVX2
instructions. To avoid that problem, we've temporarily disabled that option (you can see the message about this in the help text), until we can find time to solve it (hopefully) in the next software release.
In the interim, the only work around if you need multiple threads is to download the last version of cellranger
which supports threading on all supported platforms.
Hello - thanks for getting back to me so quickly!
I was using cellranger 5.0, which I thought was the most recent version which should have the multithreading? if not, would you please point me what is the last version?
Thanks!!
Best, Riyue
On Mon, Dec 28, 2020 at 4:03 PM Nigel Delaney notifications@github.com wrote:
Hi @riyuebao https://github.com/riyuebao, this is unfortunately intentional behavior at the moment. STAR with multiple threads utilizes OpenMP for multithreading, and this library can create issues when run on older hardware that does not support AVX2 instructions. To avoid that problem, we've temporarily disabled that option (you can see the message about this in the help text), until we can find time to solve it (hopefully) in the next software release.
In the interim, the only work around if you need multiple threads is to download the last version of cellranger which supports threading on all supported platforms.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/10XGenomics/cellranger/issues/107#issuecomment-751863307, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABMWBIOBJWCB5NFSTYLZ6I3SXDXATANCNFSM4VMOH2FA .
5.0
is best to use except for this one issue, for which the old version available here:
https://support.10xgenomics.com/single-cell-gene-expression/software/downloads/4.0/
provides the work around.
Great thank you!! Will try :)
Best, Riyue
On Mon, Dec 28, 2020 at 4:20 PM Nigel Delaney notifications@github.com wrote:
5.0 is best to use except for this one issue, for which the old version available here:
https://support.10xgenomics.com/single-cell-gene-expression/software/downloads/4.0/
provides the work around.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/10XGenomics/cellranger/issues/107#issuecomment-751867716, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABMWBIIHXDEHFNXCAPKQNWDSXDZA7ANCNFSM4VMOH2FA .
Hi, I am having the same issue in version 6.0, I'm guessing the issue was not solved with the new release?
Dear @evolvedmicrobe,
this very bad practice! Using the --nthreads option should simply crash if it doesn't work!! I have waited hours for jobs to start (the queueing system has more difficulty finding a node that has e.g. 16 cores available), then paying for 16 cores (while only 1 is being used), then running out of requested run time because I only had one core, wasting everyones time and my money. Why is this not solved, if all cellranger mkref
seems to do is call STAR --runMode genomeGenerate --runThreadN 1 _other_args_
?
Is there any post-processing done by cellranger mkref
after the call to STAR
, the full command line of which is in the Log.out ?
Disgruntled,
Philip
PS: this is cellranger 5.0.1
Thank you all for the input here. I was using cellranger 6.0.2 to build a custome reference genome and spent a lot of time to figure out why the pipelines hung at this step. @evolvedmicrobe, I think it's important to mention it on the cellranger website (https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/advanced/references) even though 10X already included the message in the help text.
Hi @jyang635 and @plijnzaad,
This is indeed an unfortunate situation and one I can assure we were looking to improve. @plijnzaad , in answer to your question, the problem is that simply calling STAR
with multiple threads in our code leads to a SEGFAULT on unsupported hardware, so we disabled this option as we felt it was better for the code path to work slowly than not work at all and/or exit with a very confusing error message.
A new version of Cell Ranger will be released in the next few weeks that re-enables the nthreads
option as well as having several other more meaningful improvements. We expect this will resolve this issue and avoid the need for further documentation around the previously disabled (and now re-enabled) option.
Warm wishes, Nigel
Has this issue been resolved? If so, can you give the specifics? If not, can you please not mark it closed? Thanks, Philip
@plijnzaad the latest version of Cell Ranger passes nthreads
onto STAR and does not show this issue. There are no plans to back port this fix to previous versions.
Hello - I was checking star log files and realized the number of threads specified on cellranger mkref is not passed onto the actual star run. Would you please take a look? The commands and logs are shown below.
Thanks!!
Best, Riyue
cellranger command
cellranger mkref log file shows
--nthreads=16
actual log file from STAR
Log.out
still shows--runThreadN 1