Nvidia Geforce GT 525 M

CritSin commented 5 years ago

support r not?

psychocrypt commented 5 years ago

you must compile it by hand if this is a fermi gpu. else update your driver to 9.X

CritSin commented 5 years ago

My nvidia deiver upgraded but not work again

Spudz76 commented 5 years ago

I have NVS 5400M which is nearly identical, arch 2.1, and you must compile with CUDA 8.0 SDK, set CUDA_ARCH=20 for best results.

You can't run newer than last CUDA 8.0 driver, which was 386.28 for Windows (you never told us your OS) 390 and 400 series will never work (other than for graphics) Newer drivers than that only run the 2D/3D with no CUDA, unless you built with cudart 8.0 (cudart64_80.dll) support, which is the only way to run old CUDA on newer drivers (but I have less luck backward compatible, and you have to add custom link flags for cudart, just run the exactly matching driver)

Spudz76 commented 5 years ago

Also pending accept, new docs section here summarizing the above

CritSin commented 5 years ago

Please write link for downloading best cuda nvidia gt 525m

Spudz76 commented 5 years ago

Windows?

Linux?

Spudz76 commented 5 years ago

Win7 Driver Win10 Driver

SDK 8.0 - assuming you can navigate from there based on your OS

Zenitur commented 5 years ago

I have Linux and NVIDIA GeForce 560 Ti. Driver version is 390.48. Xmr-stak 2.5.1. If my xmrstak_cuda_backend.so was compiled with CUDA 9.1, then miner cannot start to mine.

[2018-11-01 03:29:25] : Mining coin: cryptonight_v8
[2018-11-01 03:29:25] : NVIDIA: try to load library 'xmrstak_cuda_backend_cuda10_0'
WARNING: NVIDIA Insufficient driver!
WARNING: NVIDIA no device found
[2018-11-01 03:29:25] : NVIDIA: try to load library 'xmrstak_cuda_backend_cuda9_2'
WARNING: NVIDIA Insufficient driver!
WARNING: NVIDIA no device found
[2018-11-01 03:29:25] : NVIDIA: try to load library 'xmrstak_cuda_backend'
NVIDIA: found 1 potential device's
[2018-11-01 03:29:26] : Starting NVIDIA GPU thread 0, no affinity.
CUDA [9.1/9.1] GPU#0, device architecture 21: "GeForce GTX 560 Ti"... WARNING: skip device - binary does not contain required device architecture
[2018-11-01 03:29:26] : Setup failed for GPU 0. Exiting.

// Also I cannot type any symbol in the Vitrual Terminal after this message shown But if it was compiled with CUDA 8.0, it's starting to mine:

[2018-11-01 03:31:29] : Mining coin: cryptonight_v8
[2018-11-01 03:31:29] : NVIDIA: try to load library 'xmrstak_cuda_backend_cuda10_0'
WARNING: NVIDIA Insufficient driver!
WARNING: NVIDIA no device found
[2018-11-01 03:31:29] : NVIDIA: try to load library 'xmrstak_cuda_backend_cuda9_2'
WARNING: NVIDIA Insufficient driver!
WARNING: NVIDIA no device found
[2018-11-01 03:31:29] : NVIDIA: try to load library 'xmrstak_cuda_backend'
NVIDIA: found 1 potential device's
[2018-11-01 03:31:29] : Starting NVIDIA GPU thread 0, no affinity.
CUDA [9.1/8.0] GPU#0, device architecture 21: "GeForce GTX 560 Ti"... device init succeeded
[2018-11-01 03:31:29] : NVIDIA: use library 'xmrstak_cuda_backend'
[2018-11-01 03:31:29] : MEMORY ALLOC FAILED: mlock failed
[2018-11-01 03:31:30] : WARNING: No AMD OpenCL platform found. Possible driver issues or wrong vendor driver.
[2018-11-01 03:31:30] : WARNING: backend AMD (OpenCL) disabled.
[2018-11-01 03:31:30] : Fast-connecting to stratum+tcp://xmr.pool.minergate.com:45700 pool ...
[2018-11-01 03:31:30] : Pool stratum+tcp://xmr.pool.minergate.com:45700 connected. Logging in...
[2018-11-01 03:31:30] : Difficulty changed. Now: 1063.
[2018-11-01 03:31:30] : Pool logged in.
[2018-11-01 03:31:33] : New block detected.

My build: https://yadi.sk/d/i7IR3eUxvsjx8A (don't afraid of this link, this is a cloud service). All 4 libraries (10.0, 9.2, 9.1 and 8.0) are included in this tar.xz file. And there are not stripped.

Spudz76 commented 5 years ago

Yeah, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0

Best results run the last CUDA 8.0 driver 386.28 as I have also said (yours as you see is 9.1 running in 8.0 compatibility mode... compatibility mode is never optimal in anything)

vtec888 commented 5 years ago

hi Spudz76m, i am newbie to mining..recently i have stuck with the setup also almost same issue with above.. if you dont mind can you guide us from driver install, what item to download such as cuda version, which visual studio version required and etc in-order to let the miner start work.. My current laptop in window 10, geforce gtx 540m before i format my laptop that was fine..these have stuck me for quite sometimes...

Thanks

Des

Spudz76 commented 5 years ago

All of that is standard, in the docs, other than the table of driver-to-cuda is not included yet For CUDA8 you must install v140 Toolset in addition to normal MSVSCE-2017 (but it says so in the docs)

mechanator commented 5 years ago

Yeah, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0, Fermi requires CUDA 8.0

Best results run the last CUDA 8.0 driver 386.28 as I have also said (yours as you see is 9.1 running in 8.0 compatibility mode... compatibility mode is never optimal in anything)

Then who can make a windows binary latest release with CUDA 8 for older Fermi cards? Assuming you have the correct drivers for Win 7/10 for Tesla M2050 through M2090 and GTX 400/500 series. It would be nice to mine CN v2 and newer variants. XMRIG did it. One thing I noticed is that the Tesla cards have 3-6GB of ram, but are thread limited. Even the GTX 465 has 1.5GB, and the GTX 580 / 590 has 1.5GB X 2 with two cores. But the amount of ram used by the cuda call is much smaller than that. There are stlll plenty of the gpus available and assuming cheap power usage or surplus quantities of cards available.

Spudz76 commented 5 years ago

Then who can make a windows binary latest release with CUDA 8 for older Fermi cards?

Literally anyone. I have never run a Release binary there is too much incentive to self-compile, and it is pretty easy to rebuild new pulls once the toolchain/SDK stack is set up.

The way it builds releases now could be extended to make a Release binary pack for all CUDA major versions (8, 9.2, 10.0), which I think was a goal related to the recent CUDA-backend-dll multiloader but the CMake/AppVeyor parts need to be completed.

I would host my builds publicly, but they are modded with my own branch hacks. I don't build vanilla master branch, only dev + my mod pack.

psychocrypt commented 5 years ago

No since cuda 8 requires v140 and all other v141 I can not mix it :-(

mechanator commented 5 years ago

Before the last reply, I took an attempt to compile it with the instructions supplied. It comes up with this error message which I tracked back to the jconf.cpp CPU threads being messed up saying "Thread 0 has invalid config." See screenshot here: https://imgur.com/GD4ioDE

It looks similar to the same error message string in the amd section of jconf.cpp. So I made a dare and modified the message and to make it to return true instead of false at:

"for(size_t i=0; i < GetThreadCount(); i++) { if(!GetThreadConfig(i, c)) { printer::inst()->print_msg(L0, "Thread %llu has invalid CPU config.", int_port(i)); return true; } }

return true;

}

With that change. I still get cpu thread errors, See attached screenshot: https://imgur.com/jD3Q3nS Some kind of warning about affinity settings. Although, it will mine using GPUs only with the --noCPU option or commenting out the options in cpu.txt.
see screenshot: https://imgur.com/DJcCstD

System Build information: Intel i5 2500k Asus p8z77 motherboard, Cpu is AES-Ni, BIOS last rev 1402, No overlocks on cards or board. Stable running 2.45 and previous versions for a year. 14GB RAM. Windows 10 new build a week old. Pulled the git code for 2.7.1 on 12/15 Installed either driver 385.28 for gtx 400/500 and Tesla support, installed CUDA 8.1.6.2 , Same error message with CUDA8 custom driver but that doesn't run the GTX cards. Same error message on 386.28 for Tesla only cards. Followed your compilation instructions exactly. Installed VS 2017 Community with the options mentioned. Tested version: cmake 3.9 installed. Installed that AMD OCL Light kit,. from here: https://github.com/GPUOpen-LibrariesAndSDKs/OCL-SDK/releases I did install this:

Dependencies OpenSSL/Hwloc and Microhttpd

For CUDA 8*:
- Download version 1 of the precompiled binary from https://github.com/fireice-uk/xmr-stak-dep/releases/download/v1/xmr-stak-dep.zip
- Version 1 of the pre-compiled dependencies is not compatible with Visual Studio Toolset v141
Extract archive to C:\xmr-stak-dep - ok did that. The build script I made:

mkdir build cd build set CMAKE_PREFIX_PATH=C:\xmr-stak-dep\hwloc;C:\xmr-stak-dep\libmicrohttpd;C:\xmr-stak-dep\openssl cmake -G "Visual Studio 15 2017 Win64" -T v140,host=x64 .. cmake --build . --config Release --target install cd bin\Release copy C:\xmr-stak-dep\openssl\bin* .

I also tried excluding components with the cmake switches to no avail still error messages about affinity settings out range when mining on CPU combined. Using any of -DMICROHTTPD_ENABLE=OFF -DOpenSSL_ENABLE=OFF -DOpenCL_ENABLE=ON -DCMAKE_LINK_STATIC=ON , Didn't make a difference.

These card I know work since I can run xmr-stak 2.45 with cuda 9.2 , but want the variant 2 algos.

At least the opencl mode runs, but at performance hit, but CUDA8 for arqma v7 cn-lite was only 10% slower. I tried the CUDA 8 Toolkit driver, 385.28 for GTX/Tesla combos, and just the 386.28 driver for Tesla M/S series with each attempt at building and running it. I also tried the --noAMD switch, and the --openCLVendor NVIDIA --noAMDCache options. Yes, I cleared the .opencl cache directory every time I ran it. I hope this helps figuring out what causes the cpu thread crash.

Spudz76 commented 5 years ago

Did you delete all config files and regenerate? Many new (and required) options were added and will not parse "old version" configs. Check topmost line of each config file, it will say what version had created it. If it doesn't say any version then it's REEEEAL old config file format, definitely won't work.

mechanator commented 5 years ago

You are correct sir. Restarted from scratch. It runs without the strange thread error. The good news is that you can still insall CUDA 8 Toolkit, the older drivers, and still run XMR-STAK 2.7.1 with all the latest updates.

Spudz76 commented 5 years ago

The config file parsing could be a bit more helpful when it can't find a required setting in your current JSON structure. Partially the reason for all-or-nothing behavior is that RapidJSON is called with the whole content of the file and returns the result (or fails). There is not much per-line state happening to be able to tell what's wrong.

mechanator commented 5 years ago

I buillt the CUDA 8 for 2.71 and made a release for it here: https://github.com/mechanator/XMR-STAK-CUDA-8-2.7.x

fireice-uk / xmr-stak