search

fireice-uk / xmr-stak

Free Monero RandomX Miner and unified CryptoNight miner
GNU General Public License v3.0
4.05k stars 1.79k forks source link

NVIDIA Compute not working. #686

Closed goose closed 6 years ago

goose commented 6 years ago

Intel Core I5 6600 Windows 10 64-bit Gigabyte GTX970 G1

I cannot start the miner with Compute mode enabled on my GTX970, I have tried both with Cuda 8 with patch and cuda 9 latest installer.

I can however start either version of the miner if i switch the card back to normal mode. Please advise.

"gpu_threads_conf" : [ // gpu: GeForce GTX 970 architecture: 52 // memory: 3387/4096 MiB // smx: 13 { "index" : 0, "threads" : 16, "blocks" : 39, "bfactor" : 6, "bsleep" : 25, "affine_to_cpu" : false, "sync_mode" : 3, },

],

xmr-stak 2.1.0 30888a3

Brought to you by fireice_uk and psychocrypt under GPLv3. Based on CPU mining code by wolf9466 (heavily optimized by fireice_uk). Based on NVIDIA mining code by KlausT and psychocrypt. Based on OpenCL mining code by wolf9466.

Configurable dev donation level is set to 2.0%

You can use following keys to display reports: 'h' - hashrate 'r' - results 'c' - connection

[2017-12-23 22:52:27] : Start mining: MONERO [2017-12-23 22:52:28] : Starting NVIDIA GPU thread 0, no affinity. [CUDA] Error gpu 0: <C:/Users/Downloads/release/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:221

I have also tried the latest version and get the following:

xmr-stak 2.2.0 c4400d19

Brought to you by fireice_uk and psychocrypt under GPLv3. Based on CPU mining code by wolf9466 (heavily optimized by fireice_uk). Based on NVIDIA mining code by KlausT and psychocrypt. Based on OpenCL mining code by wolf9466.

Configurable dev donation level is set to 2.0%

You can use following keys to display reports: 'h' - hashrate 'r' - results 'c' - connection

[2017-12-23 22:54:32] : Start mining: MONERO [2017-12-23 22:54:33] : Starting NVIDIA GPU thread 0, no affinity. [CUDA] Error gpu 0: <C:/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:207

psychocrypt commented 6 years ago

please change bfactor to 8 or reduce the threads in the file nvidia.txt

goose commented 6 years ago

"gpu_threads_conf" : [ // gpu: GeForce GTX 970 architecture: 52 // memory: 3387/4096 MiB // smx: 13 { "index" : 0, "threads" : 4, "blocks" : 13, "bfactor" : 8, "bsleep" : 25, "affine_to_cpu" : false, "sync_mode" : 3, },

],

set to this and it still does the same error.

demib72 commented 6 years ago

I got a similiar error ([CUDA] Error gpu 0: <C:/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:207) when I set the "optimize for compute performance" option in the nvidia control panel to on. Turning it off allows it to work once more.

Below are the settings I am using for my GTX 970. bfactor is set where it is because 8 does not smooth out the lagging between computing and rendering the screen when I am using the computer at the same time.

[ // gpu: GeForce GTX 970 architecture: 52 // memory: 3386/4096 MiB // smx: 13 { "index" : 0, "threads" : 13, "blocks" : 39, "bfactor" : 10, "bsleep" : 100, "affine_to_cpu" : false, "sync_mode" : 3, },

],

If I delete the nvidia config file and let the program auto-config, it comes in as blank.

carlthome commented 6 years ago

Same issue here.

eruditej commented 6 years ago

why are you tinkering with blocks and threads in config files?

developer should use grid stride: https://devblogs.nvidia.com/parallelforall/cuda-pro-tip-write-flexible-kernels-grid-stride-loops/

Let grid stride launch the right threads/blocks of your nv kernels automatically.

carlthome commented 6 years ago

@eruditej, Hemi looks cool, but xmr-stak already warns when the number of blocks is not a multiple of the number of processors and it might not be worth a breaking change unless there is a clear performance improvement.

Also, off topic.

goose commented 6 years ago

@eruditej that link doesn't seem relevant? I already know the blocks and threads my GPU will support, however xmr-stak does not support NVIDIA compute and if it does it definitely doesn't on the Gigabyte G1 GTX970.

However any input on how to get past the CUDA issues would be useful? :-)

psychocrypt commented 6 years ago

Could you please post the full error. In the first post there is the last line missing. Bfactor and co is not helping here the error is coming from the l1 disbling call.

goose commented 6 years ago

xmr-stak 2.2.0 c4400d19

Brought to you by fireice_uk and psychocrypt under GPLv3. Based on CPU mining code by wolf9466 (heavily optimized by fireice_uk). Based on NVIDIA mining code by KlausT and psychocrypt. Based on OpenCL mining code by wolf9466.

Configurable dev donation level is set to 2.0%

You can use following keys to display reports: 'h' - hashrate 'r' - results 'c' - connection

[2017-12-28 17:57:55] : Start mining: MONERO [2017-12-28 17:57:55] : Starting NVIDIA GPU thread 0, no affinity. [CUDA] Error gpu 0: <C:/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:207

This is all of the console output that I get given, then the application crashes and closes

psychocrypt commented 6 years ago

I will create a bug fix for this which allows that this line fails. Could you pease tru the release 2.1.0 as I remember this line was not guarded in this release.

goose commented 6 years ago

Here you go:

xmr-stak 2.1.0 30888a3

Brought to you by fireice_uk and psychocrypt under GPLv3. Based on CPU mining code by wolf9466 (heavily optimized by fireice_uk). Based on NVIDIA mining code by KlausT and psychocrypt. Based on OpenCL mining code by wolf9466.

Configurable dev donation level is set to 2.0%

You can use following keys to display reports: 'h' - hashrate 'r' - results 'c' - connection

[2017-12-28 18:07:35] : Start mining: MONERO [2017-12-28 18:07:35] : Starting NVIDIA GPU thread 0, no affinity. [CUDA] Error gpu 0: <C:/Users/XXX/Downloads/release/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:221

carlthome commented 6 years ago

2.1.0 doesn't work either.

[CUDA] Error gpu 0: <C:/Users/XXX/Downloads/release/xmr-stak-master/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:221

That's a weird error by the way because my user name is not XXX and I'm not running from that directory.

goose commented 6 years ago

@carlthome I assume this is just because there is a hard code directory location in the code.

psychocrypt commented 6 years ago

This is the path from the precompiled binary it is hard coded into the binary.

carlthome commented 6 years ago

Interestingly, v2.2.0 works fine with "Optimise for Compute Performance: On" and I get a good improvement on my 980 Ti from 600 H/s to 680 H/s.

carlthome commented 6 years ago

I assume this error was related to CUDA 9.1 compatibility fixes or something.

goose commented 6 years ago

@carlthome please may you describe a little more about your setup such as drivers?

goose commented 6 years ago

If it helps I am running the following driver: 388.71-desktop-win10-64bit-international-whql

goose commented 6 years ago

I got compute mining working :smiley: [7:22 PM] Goose: "gpu_threads_conf" : [ { "index" : 0, "threads" : 12, "blocks" : 52, "bfactor" : 8, "bsleep" : 25, "affine_to_cpu" : false, }, ],

This driver 388.71-desktop-win10-64bit-international-whql

and the following 2 settings in the manage 3D settings DSR - Factors 1.20x Optimize for compute performance on

Gigabyte GTX970 G1 mining electroneum @ 410 H/s

I'll probably get this to 450 H/s once I re-apply my card overclock

Thanks for all the help, I'd advise asking people to change their DSR- Factors in the future if they have this issue with NVIDIA cards.