Error GPU 0

[pfreindorf]

Hi,

I got into another problem when starting the miner with my GeForce GTX 1050Ti and cuda 9 version : [CUDA] Error gpu 0: <C:/mining/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:244

Before this happens my screen gets black, then image comes back and both xmr.stack.exe and monero wallet stop working.

Tried to modify the nvidia.txt with the following values without success :+1:

"gpu_threads_conf" : [ // gpu: GeForce GTX 1050 Ti architecture: 61 // memory: 3785/4096 MiB // smx: 6 { "index" : 0, "threads" : 64, "blocks" : 18, "bfactor" : 0, "bsleep" : 0, "affine_to_cpu" : false, },

],

Any ideas ?

Pierre

[psychocrypt]

Do you changed the auto suggested config? Bfactor should be at least 6 and bsleep should be 25 in windows.

[psychocrypt]

Readuce threads or increase bfactor to 8 should also help

[pfreindorf]

Much better now thanks. Except the GPU I have a very old hardware some upgrade in perspective.

Pierre

[photis]

I'm having the same issue on a Kaby Lake i7 7700K processor with 6 x NVIDIA 1070 GPU's. I can't give you the exact error message as my main screen is frozen and CTRL-C doesn't work. From a SSH session I can see that xmr-stak is taking 100% CPU, as is an unnamed process and XORG. My nvidia.txt is

"gpu_threads_conf" :
[
  // gpu: GeForce GTX 1070 architecture: 61
  //      memory: 8005/8114 MiB
  //      smx: 15
  { "index" : 0,
    "threads" : 86, "blocks" : 45,
    "bfactor" : 0, "bsleep" :  0,
    "affine_to_cpu" : false,
  },
  // gpu: GeForce GTX 1070 architecture: 61
  //      memory: 8005/8114 MiB
  //      smx: 15
  { "index" : 1,
    "threads" : 86, "blocks" : 45,
    "bfactor" : 0, "bsleep" :  0,
    "affine_to_cpu" : false,
  },
  // gpu: GeForce GTX 1070 architecture: 61
  //      memory: 8005/8114 MiB
  //      smx: 15
  { "index" : 2,
    "threads" : 86, "blocks" : 45,
    "bfactor" : 0, "bsleep" :  0,
    "affine_to_cpu" : false,
  },
  // gpu: GeForce GTX 1070 architecture: 61
  //      memory: 8005/8114 MiB
  //      smx: 15
  { "index" : 3,
    "threads" : 86, "blocks" : 45,
    "bfactor" : 0, "bsleep" :  0,
    "affine_to_cpu" : false,
  },
  // gpu: GeForce GTX 1070 architecture: 61
  //      memory: 8005/8114 MiB
  //      smx: 15
  { "index" : 4,
    "threads" : 86, "blocks" : 45,
    "bfactor" : 0, "bsleep" :  0,
    "affine_to_cpu" : false,
  },
  // gpu: GeForce GTX 1070 architecture: 61
  //      memory: 8005/8114 MiB
  //      smx: 15
  { "index" : 5,
    "threads" : 86, "blocks" : 45,
    "bfactor" : 0, "bsleep" :  0,
    "affine_to_cpu" : false,
  },

],

This is the default configuration as created by xmr-stak the first time. Any suggestions?

Running on Ubuntu 16.04.

[awesomazing]

I have the same problem on single GPU test instance on Windows.

OS Name Microsoft Windows 10 Pro Version 10.0.15063 Build 15063 Processor Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz, 4200 Mhz, 4 Core(s), 8 Logical Processor(s)

Nvidia configuration

"gpu_threads_conf" : [ // gpu: GeForce GTX 1070 architecture: 61 // memory: 6782/8192 MiB // smx: 15 { "index" : 0, "threads" : 72, "blocks" : 45, "bfactor" : 6, "bsleep" : 25, "affine_to_cpu" : false, },

],

[psychocrypt]

Please resuce the thraeds. It could be that the window manager increases the memory usage over time but the miner has already allocated the full memory. For linux, disable the x window manager and work only on the terminal.

[Pseudothink]

FYI, I tried running the x64 current version of (unified) XMR-STAK briefly before installing the Windows 10 Fall Creator Update. It seemed to work fine, though I didn't do extensive testing with it. Then immediately after installing the Windows 10 Fall Creators Update, I started getting the error in the original post. It wouldn't even really run, it would crash within seconds of starting. I tried reducing my "threads" value from 54 to 36, and it started working again. If I increased it to 42, it would run for a bit, but the error would come back quickly, within a minute or two. I'm still testing it to see how long it runs at 36.

[mvpel]

Pseudo - I had the same problem because the Creators Update replaced the previously-working driver with a different version.

[Pseudothink]

That makes sense! Did you reinstall the original driver (or install a new/different driver), and if so, did it resolve this issue? So far, mine's been running reliably at a threads value of 36, for the past 2 days. I'm not sure how much it would increase its performance to go through the minor hassle of fixing the driver issue then increasing the threads count back to 54, but I'm willing to find out for the sake of "why the heck not".

[HiceS]

I'm having a similar issue but the response is:

[CUDA] Error gpu 0: <C:/Users/widera/Downloads/release/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:266

That's not a path on my local machine but I will fork and see if I can come up with a fix.

[mvpel]

Pseudo - my issue was with AMD GPUs, where the Fall Creators Update "upgraded" the beta Blockchain Driver from AMD, so it's not strictly germane to this issue, but perhaps the same sort of situation happened here - maybe the Windows update moved the system to the 9.x NVIDIA driver when stak needs the 8.x version? Just a guess. I'm dealing with some Volta cards at the moment and the 9.x driver kit is moving very quickly.

[pjdouillard]

I too have this error: "[CUDA] Error gpu 0: <C:/Users/widera/Downloads/release/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:266"

It seems to be a hardcoded path in the .exe somehow. Would be nice if this could be fixed.

[LordMrcS]

@pjdouillard I had that error, just increase bfactor from 6 to 8.

[borzaka]

Same here with 2.2.0: [CUDA] Error gpu 0: <C:/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:265[CUDA] Error gpu 1

Default nvidia.txt:

"gpu_threads_conf" :
[
  // gpu: GeForce GTX 1080 Ti architecture: 61
  //      memory: 9312/11264 MiB
  //      smx: 28
  { "index" : 0,
    "threads" : 54, "blocks" : 84,
    "bfactor" : 6, "bsleep" :  25,
    "affine_to_cpu" : false, "sync_mode" : 3,
  },
  // gpu: GeForce GTX 1080 Ti architecture: 61
  //      memory: 9312/11264 MiB
  //      smx: 28
  { "index" : 1,
    "threads" : 54, "blocks" : 84,
    "bfactor" : 6, "bsleep" :  25,
    "affine_to_cpu" : false, "sync_mode" : 3,
  },
  // gpu: GeForce GTX 1080 Ti architecture: 61
  //      memory: 9313/11264 MiB
  //      smx: 28
  { "index" : 2,
    "threads" : 54, "blocks" : 84,
    "bfactor" : 6, "bsleep" :  25,
    "affine_to_cpu" : false, "sync_mode" : 3,
  },
],

And I also had this message:

Application has been blocked from accessing Graphics hardware.

Changed bfactor from 6 to 8 is helped. I have a dedicated iGPU, the 3 1080 Ti has no monitor plugged in.

[SteakisGood]

I have tried all of the above, yet mine still fails. Bfactors and bsleeps havent stopped the thing from crashing. Ive tried everything from switching the b factor and b sleep balues to using 1 thread. Help is appreciated, I'll pay you if you can help me get it to work.

[borzaka]

Please, copy your current nvidia.txt here. Also try to delete your nvidia.txt and start the miner again, and copy your newly created nvidia.txt here.

[SteakisGood]

I have it attached on this email.

On Tue, Feb 6, 2018 at 2:03 AM, Attila notifications@github.com wrote:

Please, copy your current nvidia.txt here. Also try to delete your nvidia.txt and start the miner again, and copy your newly created nvidia.txt here.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/fireice-uk/xmr-stak-nvidia/issues/187#issuecomment-363372470, or mute the thread https://github.com/notifications/unsubscribe-auth/Aigh94raVzaIByWjqp9cKdruTT6_TvK2ks5tSCNpgaJpZM4QqccW .

[borzaka]

I can't see it. Attach it on GitHub, not via email reply.

[SteakisGood]

[borzaka]

Try to change the following:

• bfactor to 12 (you can try 8, 10, but for me 12 is where my computer stays usable and not freezing)
• bsleep to 100 (optional, but you can try)

[SteakisGood]

What about the threads and blocks? Will they cause the system to crash/run more stable?

Additionally: When mining this, is it better to overclock memory or clockspeed? Which is cryptonight more heavy in?

[borzaka]

Leave the threads and blocks for now, try what I wrote. CryptoNight is more memory dependent.

[PashaKai]

changed 'threads' to 30 - work for me at 1080ti. If I change to 31+, I have that 265 error