Skip to content
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

nvidia plugin CUDA8 works maximum with 16 threads #22

Closed
berezinevgeniy opened this issue Nov 28, 2019 · 1 comment
Closed

nvidia plugin CUDA8 works maximum with 16 threads #22

berezinevgeniy opened this issue Nov 28, 2019 · 1 comment

Comments

@berezinevgeniy
Copy link

Cuda 8 version is works for me? but only with 16 threads max.
I have several GPU 640M, 720M, 740M, 800M, 920M....with Fermi, as i think
It`s works only with CUDA 8. I use 5.0.1 xmrig with cuda 8 plugin.
When i set more than 16 thread (24, 32, 48, 64) app will crash with:

thread #0 failed with error <cryptonight_core_gpu_hash>:752 "an illegal memory access was encountered"

number of blocks doesn`t metter. Almost gpu has 2 SMX with 2Gb memory. And bxt 16x16 is not enough for full throttle. 16x32 and 16x64 will get more hash. There is no problem with xmr-stak cuda8 version, but randomx version of xmr-stak has no cuda 8 -(
Can you help me with xmrig?

p.s. log
[2019-11-28 23:13:08.170] Huge pages support was successfully enabled, but reboot required to use it

  • ABOUT XMRig/5.0.1 gcc/9.2.0
  • LIBS libuv/1.31.0 OpenSSL/1.1.1c hwloc/2.0.4
  • HUGE PAGES unavailable
  • CPU Intel(R) Core(TM) i3-4160T CPU @ 3.10GHz (1) x64 AES
    L2:0.5 MB L3:3.0 MB 2C/4T NUMA:1
  • DONATE 1%
  • ASSEMBLY auto:intel
  • POOL Cleanup everything in GPU context when releasing it #1 10.54.0.1:3333 algo auto
  • COMMANDS hashrate, pause, resume
  • OPENCL disabled
  • CUDA 8.0/9.1/2.0.1-beta
  • NVML 9.388.73/388.73 press e for health report
  • CUDA GPU #0 01:00.0 GeForce 800M 1475/900 MHz smx:1 arch:21 mem:1691/2048 MB
    [2019-11-28 23:13:08.217] net use pool 10.54.0.1:3333 10.54.0.1
    [2019-11-28 23:13:08.217] net new job from 10.54.0.1:3333 diff 1200 algo cn/r height 1976945
    [2019-11-28 23:13:08.217] nv use profile cn (1 thread) scratchpad 2048 KB
    | # | GPU | BUS ID | I | T | B | BF | BS | MEM | NAME
    | 0 | 0 | 01:00.0 | 256 | 32 | 8 | 8 | 25 | 512 | GeForce 800M
    [2019-11-28 23:13:08.420] nv READY threads 1/1 (205 ms)
    [2019-11-28 23:13:09.608] nv thread #0 failed with error <cryptonight_core_gpu_hash>:752 "an illegal memory access was encountered"
    [2019-11-28 23:13:11.404] SIGHUP received, exiting
    [2019-11-28 23:13:11.404] nv stopped (9 ms)
@Spudz76
Copy link
Contributor

Spudz76 commented Dec 18, 2019

Try inverse, 32x16 or 64x16
I usually get good perf with between 32x8 and 48x8 actually

@xmrig xmrig closed this as completed Aug 31, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants