All components are too new!!

[basaltzhang]

CUDA Driver >= 11.7 (Not tested on 12.0)
GCC >= 11.0
Python >= 3.7
cmake >= 3.22

Difficult to compile from source code, I have to update ALL my tools to fit these minimum versions. Are these really needed?

Example, if I don't need #include to use c++20 <=> , I can use gcc version 8, am i right?

[lji72]

+1

[lji72]

It's difficult to use in some project with lower gcc.

[edric1261234]

especially GCC

[shiremathNV]

Hello, Thanks for your feedback on the minimum versions.

On the new release that went live this week (v0.3.0), we have lowered the minimum versions on CUDA toolkit and GCC.

• CUDA Toolkit >= 11.3 support for core library build and run
• GCC >= 9.0 supported for core library build and run

Hope this helps. For lower GCC version than 9.0, will using the gcc-9 pre-builds to link against meet your requirements?
@basaltzhang @lji72 @edric1261234

[davidhopper2003]

Hello, Thanks for your feedback on the minimum versions.

On the new release that went live this week (v0.3.0), we have lowered the minimum versions on CUDA toolkit and GCC.

• CUDA Toolkit >= 11.3 support for core library build and run
• GCC >= 9.0 supported for core library build and run

Hope this helps. For lower GCC version than 9.0, will using the gcc-9 pre-builds to link against meet your requirements? @basaltzhang @lji72 @edric1261234

According to the GCC and CUDA versions you gave, it cannot be compiled successfully on Ubuntu 20.04. Please refer to my post: #95

[chencuber]

yes，very difficult to integrate into current working project!

Ubuntu22.04 is too high, even nvidia ngc container only support Ubuntu20.04 with cuda<12。
just suggest that all building（not just core library） uses Ubuntu20.04 with cuda<=11.6 and gcc >= 9.0。