[Bug 293362] graphics/nvidia-drm-kmod: No devices on 15.0

[email protected]
Newsgroups gmane.os.freebsd.devel.x11
Message-ID <[email protected]/bugzilla/>
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=293362

--- Comment #42 from Tomoaki AOKI <[email protected]> ---
(In reply to Martin Cracauer from comment #40)
Not sure your hello-world-cuda2 is relevant or not, but CUDA on FreeBSD has at
least one limitation.

NVIDIA driver sets for FreeBSD does NOT have nvidia-uvm.ko, which is needed for
unified virtual memory accessing. So anything really require the
functionalities cannot work.
This applies to linked libraries, so if something called from the program code
require UVM, it shouldn't work.

Note that, IIUC, /compat/linux/usr/share/ai-ml-env/dummy-uvm.so installed by
science/linux-ai-ml-env mimics as if nvidia-uvm.ko is loaded and functional,
thus, checking the existence of the functionalities would pass. But it's
actually a dummy, so actual call to the functionalities should fail. But
anything actually does NOT require it can work.

Example (benchmark/clpeak):
% nv-sglrun clpeak
/usr/local/lib/libc6-shim/libc6.so: shim init

Platform: NVIDIA CUDA
  Device: NVIDIA RTX A400
    Driver version  : 595.58.03 (FreeBSD)
    Compute units   : 6
    Clock frequency : 1762 MHz

    Global memory bandwidth (GBPS)
      float   : 80.20
      float2  : 82.82
      float4  : 84.59
      float8  : 84.71
      float16 : 86.47

    Single-precision compute (GFLOPS)
      float   : 2469.95
      float2  : 2590.07
      float4  : 2614.91
      float8  : 2558.61
      float16 : 2579.89

    No half precision support! Skipped

    Double-precision compute (GFLOPS)
      double   : 43.39
      double2  : 43.21
      double4  : 43.16
      double8  : 42.83
      double16 : 42.60

    Integer compute (GIOPS)
      int   : 1269.62
      int2  : 1354.34
      int4  : 1371.44
      int8  : 1389.18
      int16 : 1322.86

    Integer compute Fast 24bit (GIOPS)
      int   : 1309.57
      int2  : 1352.53
      int4  : 1361.72
      int8  : 1308.12
      int16 : 1328.59

    Integer char (8bit) compute (GIOPS)
      char   : 1179.97
      char2  : 1241.14
      char4  : 1231.74
      char8  : 1025.40
      char16 : 967.08

    Integer short (16bit) compute (GIOPS)
      short   : 1179.57
      short2  : 1269.67
      short4  : 1245.17
      short8  : 1035.91
      short16 : 964.43

    Transfer bandwidth (GBPS)
      enqueueWriteBuffer              : 9.29
      enqueueReadBuffer               : 9.34
      enqueueWriteBuffer non-blocking : 9.22
      enqueueReadBuffer non-blocking  : 9.10
      enqueueMapBuffer(for read)      : 11.49
        memcpy from mapped ptr        : 10.74
      enqueueUnmap(after write)       : 12.89
        memcpy to mapped ptr          : 12.48

    Kernel launch latency : 6.54 us

-- 
You are receiving this mail because:
You are the assignee for the bug.
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.