GPU #0 COMPUTE ERROR

[zer0gra]

• ABOUT XMRig/6.20.0 gcc/11.2.0
• LIBS libuv/1.44.2 OpenSSL/1.1.1s hwloc/2.9.0
• HUGE PAGES permission granted
• 1GB PAGES unavailable
• CPU AMD Ryzen 5 3500X 6-Core Processor (1) 64-bit AES
  L2:3.0 MB L3:32.0 MB 6C/6T NUMA:1
• MEMORY 7.9/31.9 GB (25%)
  DIMM_A0: 16 GB DDR4 @ 3600 MHz CMW32GX4M2Z3600C18
  DIMM_B0: 16 GB DDR4 @ 3600 MHz CMW32GX4M2Z3600C18
• MOTHERBOARD Micro-Star International Co., Ltd. - H510M-A PRO (MS-7D22)
• DONATE 1%
• ASSEMBLY auto:ryzen
• POOL 128tx exploit #1 gulf.moneroocean.stream:10128 algo auto
• COMMANDS hashrate, pause, resume, results, connection
• OPENCL disabled
• CUDA 9.2/12.3/6.3.1
• NVML 12.546.17/546.17 press e for health report
• CUDA GPU #0 29:00.0 NVIDIA GeForce RTX 3060 1837/7501 MHz smx:28 arch:86 mem:11255/12287 MB
  [2023-11-22 15:38:20.378] net use pool gulf.moneroocean.stream:10128 51.75.64.249
  [2023-11-22 15:38:20.378] net new job from gulf.moneroocean.stream:10128 diff 23898 algo rx/0 height 125782
  [2023-11-22 15:38:20.378] cpu use argon2 implementation AVX2
  [2023-11-22 15:38:20.530] msr register values for "ryzen_17h" preset have been set successfully (152 ms)
  [2023-11-22 15:38:20.532] randomx init dataset algo rx/0 (6 threads) seed e354a769597c6c19...
  [2023-11-22 15:38:20.534] randomx allocated 2336 MB (2080+256) huge pages 100% 1168/1168 +JIT (1 ms)
  [2023-11-22 15:38:27.309] randomx dataset ready (6774 ms)
  [2023-11-22 15:38:27.311] nvidia use profile rx (1 thread) scratchpad 2048 KB
  | # | GPU | BUS ID | INTENSITY | THREADS | BLOCKS | BF | BS | MEMORY | NAME
  | 0 | 0 | 29:00.0 | 1792 | 32 | 56 | 6 | 25 | 3584 | NVIDIA GeForce RTX 3060
  [2023-11-22 15:38:27.658] nvidia READY threads 1/1 (342 ms)
  [2023-11-22 15:38:35.854] nvidia GPU #0 COMPUTE ERROR
  [2023-11-22 15:38:53.730] nvidia GPU #0 COMPUTE ERROR
  [2023-11-22 15:38:59.383] net new job from gulf.moneroocean.stream:10128 diff 23898 algo rx/0 height 125783

[Moreno-Borsalino]

Same problem with Cuda and Opencl. Suddenly appeared that error. Any idea ?

[SChernykh]

This will be fixed in the next release of the XMRig CUDA plugin.

[Moreno-Borsalino]

This is great news but the problem also occurs on graphics cards that use OPENCL (for example AMD RX Vega). And for OPENCL you don't use the Xmrig Cuda plugin. The problem is somewhere else

[SChernykh]

OpenCL code had exactly the same problem, it's fixed in #3399

[Moreno-Borsalino]

It is good news! So we have to wait for the next release for Xmrig too. Thanks Schernykh!

[Moreno-Borsalino]

This will be fixed in the next release of the XMRig CUDA plugin.

@SChernykh I return to write about this error because I did some tests with the version that should resolve the error.
I tried to compile a version of xmrig-cuda using the version taken here https://github.com/SChernykh/xmrig-cuda/tree/dev which solves the GPU #0 COMPUTE ERROR issue.
In fact the problem is solved but after about 6 hours of continuous operation without interruptions of xmrig the error appears again.
I tried on two different computers with two different Nvidia cards but the behavior is identical on both; after about 6 hours the error returns. You have to close xmrig and restart it to regain error-free operation but after 6 errors the error returns.
It's really strange behavior.