5 ms·

Why do so many of these news demos require old versions of CUDA? It’s quite annoying having to juggle the installs and disk usage of CUDA 11.6,11.7,11.8,12.1,12

Why do so many of these news demos require old versions of CUDA? It’s quite annoying having to juggle the installs and disk usage of CUDA 11.6,11.7,11.8,12.1,12.2,12.3

gwervc 3y ago
Because it takes time to write a paper and publish it. And researchers usually don't care about code so the version used when starting the project is never upgraded.
- speedgoose 3y ago
  Never upgrading the dependencies is not specific to research. For example the npm installations of react show that old versions are still downloaded a lot in the last 7 days: https://www.npmjs.com/package/react?activeTab=versions https://www.npmjs.com/package/react?activeTab=versions jQuery on a random CDN with stats also shows that many websites are likely never going to update it: https://www.jsdelivr.com/package/npm/jquery?tab=stats https://www.jsdelivr.com/package/npm/jquery?tab=stats You also sees that with system administrators: here is curl on debian from people who shares about it: https://qa.debian.org/popcon-graph.php?packages=curl&show_installed=on&show_old=on&show_recent=on&want_legend=on&want_ticks=on&from_date=&to_date=&hlght_date=&date_fmt=%25Y-%25m&beenhere=1 https://qa.debian.org/popcon-graph.php?packages=curl&show_in...
fxtentacle 3y ago
Because new CUDA versions often come with new bugs, so as a researcher with a publication deadline, you're incentivised to never upgrade. I still vividly remember how messy it was to downgrade a large TensorFlow project to a previous version which still supported an older CUDA because the current CUDA cuBLAS couldn't do matrix multiplication without silent memory corruption. (I believe it was 9.x to 10.0 that broke it) And we lacked the connections to directly talk to any engineer at NVIDIA, which meant that nobody even confirmed the bugs existence after we submitted a GitHub issue. Eventually, we found a way to trigger a crash on Google Colab using the bug and then someone from Google filed it again with NVIDIA and then things finally got fixed. About 6 months after the research paper was finished.

Why do so many of these news demos require old versions of CUDA? It’s quite annoying having to juggle the installs and disk usage of CUDA 11.6,11.7,11.8,12.1,12.2,12.3

gwervc 3y ago

Because it takes time to write a paper and publish it. And researchers usually don't care about code so the version used when starting the project is never upgraded.

speedgoose 3y ago
Never upgrading the dependencies is not specific to research. For example the npm installations of react show that old versions are still downloaded a lot in the last 7 days: https://www.npmjs.com/package/react?activeTab=versions https://www.npmjs.com/package/react?activeTab=versions jQuery on a random CDN with stats also shows that many websites are likely never going to update it: https://www.jsdelivr.com/package/npm/jquery?tab=stats https://www.jsdelivr.com/package/npm/jquery?tab=stats You also sees that with system administrators: here is curl on debian from people who shares about it: https://qa.debian.org/popcon-graph.php?packages=curl&show_installed=on&show_old=on&show_recent=on&want_legend=on&want_ticks=on&from_date=&to_date=&hlght_date=&date_fmt=%25Y-%25m&beenhere=1 https://qa.debian.org/popcon-graph.php?packages=curl&show_in...

fxtentacle 3y ago

Because new CUDA versions often come with new bugs, so as a researcher with a publication deadline, you're incentivised to never upgrade. I still vividly remember how messy it was to downgrade a large TensorFlow project to a previous version which still supported an older CUDA because the current CUDA cuBLAS couldn't do matrix multiplication without silent memory corruption. (I believe it was 9.x to 10.0 that broke it) And we lacked the connections to directly talk to any engineer at NVIDIA, which meant that nobody even confirmed the bugs existence after we submitted a GitHub issue. Eventually, we found a way to trigger a crash on Google Colab using the bug and then someone from Google filed it again with NVIDIA and then things finally got fixed. About 6 months after the research paper was finished.