[CUDA] Error gpu 0: <cryptonight_gpu_hash>:878 "a PTX JIT compilation failed"

[cyvitor]

Hi I did all the instalation. But i got this erro:

root@Frank:/home/suporte/xmrig-nvidia/build# ./xmrig-nvidia
 * ABOUT        XMRig-NVIDIA/2.14.4 gcc/5.4.0
 * LIBS         libuv/1.8.0 CUDA/8.0 OpenSSL/1.0.2g microhttpd/0.9.44
 * CPU                  Intel(R) Core(TM) i3-2120 CPU @ 3.30GHz x64 -AES
 * GPU #0       PCI:0000:01:00 GeForce GTX 750 Ti @ 1110/2700 MHz 4x40 0x0 arch:50 SMX:5 MEM:1958/1999 MiB
 * GPU #1       PCI:0000:04:00 GeForce GTX 750 Ti @ 1110/2700 MHz 4x40 0x0 arch:50 SMX:5 MEM:1959/2000 MiB
 * ALGO         cryptonight, donate=5%
 * POOL #1      donate.v2.xmrig.com:3333 variant auto
 * COMMANDS     hashrate, health, pause, resume
[2019-06-07 23:25:58] use pool donate.v2.xmrig.com:3333  159.89.38.204
[2019-06-07 23:25:58] new job from donate.v2.xmrig.com:3333 diff 1000225 algo cn/r height 1852142
[CUDA] Error gpu 0: <cryptonight_gpu_hash>:878 "a PTX JIT compilation failed"
terminate called after throwing an instance of 'std::runtime_error'
  what():  [CUDA] Error: a PTX JIT compilation failed
Abortado (imagem do núcleo gravada)

On Ubuntu 16.04 Plis help

[Spudz76]

Do you have CUDA lib "NVRTC" installed and available (required for CN-R)? This handles all the JIT compiling. Also for temp/fan reporting (-HWMON option) the "NVML" libs also need to be available. And they all have to match the CUDA8 version. And you can't run any newer driver than the last one with CUDA8 (390.87 or so)

You don't have to run CUDA8 (and such old driver) unless it's arch 20/21 (Fermi) it would be better to use 9.1 minimum on a Kepler, which would also allow a more recent linux distro (with gcc > 5.4.0).

You could also try the ubuntu toolchain ppa which provides newer gcc on old distros, I have not bothered trying without it (also updates the gcc5 to 5.5.0 which was a lot less buggy than 5.4.0).