Closed jetson24 closed 3 weeks ago
Hi @jetson24,
Apologies for the issues you are seeing. We are currently discussing the level of support for Jetson devices - our main priority is TX2 due to minKNOW's Mk1C variant.
This missing libcupti appears to be a packaging error in this distribution. You can install the cuda-cupti-11-4
apt package yourself to create this file - you may need to update your LD_LIBRARY_PATH
to include /usr/local/cuda-11.4/targets/aarch64-linux/lib
.
Regarding the compile errors:
libcufft_static_nocallback
- please use CMake >= 3.24
liblapack_static.a
- this should be provided by libcusolver-dev-11-4
apt pacakge, which should come as part of the cuda toolkit
-lnuma
- I'm not aware that we require the numa libs - possibly this is due to one of your other dependency versions?
-lstdc++
- this should be automatically detected, possibly a CMake version issue as well?
Regarding the unit test errors:
/home/jay/dorado/tests/gpu_monitor_test.cpp:359: FAILED:
REQUIRE( info->current_throttling_reason.has_value() )
with expansion:
false
with message:
info->current_throttling_reason_error := "Not Supported"
This appears to be an oversight in our test setup - we don't run these specific tests on the Jetson devices, and Jetsons don't support these queries. These errors can be safely ignored.
/home/jay/dorado/tests/IndexFileAccessTest.cpp:256: FAILED:
REQUIRE( header == EXPECTED_REF_FILE_HEADER )
with expansion:
"@SQ SN:read_0 LN:1,898"
==
"@SQ SN:read_0 LN:1898"
This appears to be a locale issue. Please try setting export LC_ALL=C
. We will look at enforcing this in future versions (I'm not clear how e.g. samtools handles thousand-separators in sam files?)
[2024-05-01 11:07:31.508] [error] CuBLAS error 15
This is the concerning one for me. This indicates an internal failure in cublas - possibly an Out of Memory, or some misconfiguration with the drivers. Note that we build this variant internally using Cuda 11.4. Unless you also see this during basecalling, I would not worry about it for the time-being. If it also occurs during basecalling, please try reducing the selected batchsize using the -b
parameter.
Alright! Thank you very much! I will try the suggestions in both pre-built as well as compiled version and keep you posted!
Thanks a ton, again!
Hi @malton-ont I wonder if it is possible to run dorado on Jetson TX2 with basecalling of SUP mode?
Hi @JMencius,
No, sup basecalling isn't supported on TX2.
hi @malton-ont, I tried your suggetions, and the the latest versions are working well on my Jetson AGX orin.
Thanks again!
Issue Report
Please describe the issue:
So I am trying to use the pre-compiled version of Dorado (downloaded from https://cdn.oxfordnanoportal.com/software/analysis/dorado-0.6.1-linux-arm64.tar.gz ) for basecalling.
Errors in pre-compiled version
However, when I untar the file and try to run
dorado
from the./dorado-0.6.1-linux-arm64.tar.gz/bin/
it is showing the following error:For this, I also tried the solution given on the github to update the
LD_LIBRARY_PATH
to/dorado-0.6.1-linux-arm64/lib
but the error was still persisting.Thus, I went to other online forums where people had identified the issue as CUDA update and suggested to compile the tool from the source. Thus, I started compiling
dorado
as per the instructions on github,which also had some errors along the way but got resolved after installing and finally I was able to compile it with gcc-9.
Errors during compilation:
The errors were:
Solved via introducing a link of
libcufft_static_nocallback.a
to/usr/lib64
.Solved via installing
liblapack-dev
solved via installing
libnuma-dev
and configuring theld
withldconfig
.libstdc++.so.6
Solved via introducing
-lstdc++
incMakelist.txt
file.Problem persists
However, even after compiling while running the
ctest --test-dir cmake-build
, the test is failing. The error log for the same is as follows and there are multiple errors for which what to do I am not able to understand.LastLog.txt
Help in this or in the pre-compiled version's error is very much needed.
Thanking all in advance!