-
Notifications
You must be signed in to change notification settings - Fork 21
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
QUESTION: How do I build with/for CUDA? #35
Comments
Hi, cmake -DCMAKE_BUILD_TYPE=Release -Dracon_build_cuda=ON .. && make Best regards, |
Any options recommendations for the CUDA specific commands? Or just start with the defaults?
Something like:
|
Also, Ubuntu 18.04 LTS will need an updated PPA for cmake: https://blog.kitware.com/ubuntu-cmake-repository-now-available/ |
You should first run with |
Okay, I'm running dual (nVIDIA SLI) RTX TITANS (24 GB each). Just to clarify; the GPUs are used for the polishing/alignment step(s) of RACON, correct? |
Yes, only polishing (POA) and alignment (pairwise). |
Nope. Build failed. Are you sure it's:
and not
|
My bad, it should be enable. |
CUDA Build fails; here's the terminal output:
|
Attached is the Build Error Log; looks like there's a pthreads issue. Is it this bug: https://stackoverflow.com/questions/24813827/cmake-failing-to-detect-pthreads-due-to-warnings/25130590#25130590 |
This file |
Can you compile Raven without option |
YES. Using this file in place of the gitclone repo's CMakeLists.txt makes the compile work; one has to specify <-pthread> at the beginning for the C and CXX compiler. Using Ubuntu 18.04.5 LTS - the rest of the information is in the terminal output attached below. Here's the terminal output (there were/was no CMakeErrorLog.txt) generated. raven_build_terminal_NO-CUDA.txt I believe the CUDA compile error has to do with the <-pedantic> flag indicating to treat ALL warnings as errors, thus causing the CUDA compiler to barf (see above). |
FIXED. It was a bad set of quotes in </raven/build/_deps/genomeworks-src/3rdparty/spdlog/include/spdlog/fmt/bundled/format.h> Pulled solution from here: https://www.gitmemory.com/issue/yuzu-emu/yuzu/2597/507715224 Changed Line 3475 from: To (remove quotes): Re-ran compile command & ran |
Okay, so now...getting back to the original question: CUDA parameters. If I have 24 GB per card (two cards) gives me 24,000 MB; and each band is 200 MB, that works out to (24,000/200=120) 120 "bands". To be safe, should I be setting both -a and -c to 110? Something like:
|
You can try with |
So, the full command would then be:
Correct? If so, can you please update the documentation <$ raven -h> file and the <ReadMe.md> file? TIA |
Yes. It is already written (see below) but I can emphasize this more.
|
Yeah, okay, my bad (not a real geek) & I missed that! So, I'm running:
Which should give me both a terminal output AND a written file output <raven_asm.fasta> |
This will also put sequences to stdout, which you do not want if they are big. Just redirect them to a file as showed above, and on stderr (directly to terminal) you will get the log. If you want the log in a file, redirect stderr with |
So: |
FYI: NVIDIA-Genomics-Research/GenomeWorks#637 GenomeWorks will be fixing this (CUDA compile) issue in the next release. |
Not sure how to both redirect to file and see it in terminal 😕. |
If I want to build with CUDA capacity (on Ubuntu 18.04 LTS), would the command be?:
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_BUILD_TYPE=racon_enable_cuda .. && make
The text was updated successfully, but these errors were encountered: