Building CCMiner 1.3.2 on Linux with Cuda 11.2

[chandlerpl]

Hello, I was wondering if there was any support for CCMiner to run on Linux with CUDA 11.2, I tried to compile it myself but I get this error, I am using Debian 10 as well and can't find any supporting CUDA drivers for CUDA 10.1 as suggested, any help on this would be great :)

nvcc -gencode=arch=compute_61,code=\"sm_61,compute_61\" -gencode=arch=compute_52,code=\"sm_52,compute_52\" -gencode=arch=ute_75,code=\"sm_75,compute_75\" -I/usr/local/cuda/include -I. -I./compat/bos-jansson -std=c++11 --ptxas-options="-v" --megcount=128 -o cuda_mtp/cuda_mtp_forlib.o -c cuda_mtp/cuda_mtp_forlib.cu
cuda_mtp/cuda_mtp_forlib.cu:894: warning: "G" redefined
 #define G(r,i,a,b,c,d) \

cuda_mtp/cuda_mtp_forlib.cu:232: note: this is the location of the previous definition
 #define G(r,i,a,b,c,d) \

cuda_mtp/cuda_mtp_forlib.cu(349): error: function "__ldlu(const uint4 *)" has already been defined

1 error detected in the compilation of "cuda_mtp/cuda_mtp_forlib.cu".
make[2]: *** [Makefile:2606: cuda_mtp/cuda_mtp_forlib.o] Error 1
make[1]: *** [Makefile:2031: all-recursive] Error 1
make: *** [Makefile:656: all] Error 2

[djm34]

Hello,

you need to remove the line
device forceinline uint4 __ldlu(const uint4 * restrict ptr) { uint4 ret; asm volatile ("ld.global.lu.v4.u32 {%0,%1,%2,%3}, [%4];" : "=r"(ret.x), "=r"(ret.y), "=r"(ret.z), "=r"(ret.w) : __LDG_PTR(ptr)); return ret; }
which is around line 325 in cuda_mtp/cuda_mtp_forlib.cu

Also if you are compiling for 11.2, I assume you want to use an rtx 30x card,

in this case you need to modify the nvcc command line and add support for those card ie:

-gencode=arch=compute_80,code="sm_80,compute_80"
-gencode=arch=compute_86,code="sm_86,compute_86" (I think the 86 target only the 3080/3090)

Regarding debian particular matter, I can't really help you, I work only with ubuntu, but you should be able to get drivers for nvidia website for various linux distro (including debian)

[djm34]

small correction:
only this gencode is required for rtx 30xx: (not the other one)
-gencode=arch=compute_86,code="sm_86,compute_86"

[chandlerpl]

Hello, thanks for the fast response, I have performed these changes and it compiled successfully however the miner now crashes with Segmentation fault a couple of seconds into mining

[djm34]

what does it say when it crashes ?

[chandlerpl]

All it says is Segmentation fault and closes immediately

[djm34]

yeah, I have the same problem in windows, I am looking into it...
Alternatively, I would suggest to install cuda 11.1 (which is working, was tested yesterday...)

[djm34]

ok I committed a change to the .cu for mtp which correct those crashes

[chandlerpl]

Hello! Thank you for the fast work, it also fixed my segmentation fault, my issue is now that the hashrate is far lower than usually, on my 1080 on Windows I could get 2.6Mh/s but now I barely get 900Kh/s, using another miner I also get 2.6Mh/s, do you have an idea of why that is happened? Sorry for all the questions :)

[djm34]

I am not sure may-be the new cuda version (I can't check at the moment, I will check a bit later). I would suggest to try cuda 11.1 may-be that will solve the problem.
Can it be a power issue ?

[chandlerpl]

Hi I will install CUDA 11.1 either tonight or tomorrow and give you an update it, thanks for all your support, I don't believe it can be a power issue

[djm34]

Ok you are lucky my air cooled 1070 fit below the tube of the water cooled 3090 :D
So I ran a small test with 11.2 it does around 400kh/s and with 11.1 2.5MH/s... so there is a problem with cuda 11.2

I'll try to understand the problem but for now, it is probably better to switch the cuda version to 11.1

[chandlerpl]

A watercooled 3090?? very nice! Been after one myself but having a hard time hahaha, thanks I will downgrade later, want me to close this ticket now or wait for a fix to the problem?

[djm34]

https://www.gigabyte.com/Graphics-Card/GV-N3090AORUSX-WB-24GD#kf :)

nice card, a little problematic with large fitting though (because of a built-in leak detector)
yeah they don't stay for sale really long have to keep watching

[STRATZ-Ken]

Any chance anyone can build this for Windows? I cannot run MTP on my 3070.

[djm34]

I'll put later in the day in release binaries for windows but will be based on cuda 11.1.
But technically there is already one in the latest release...
https://github.com/firoorg/ccminer/releases/download/1.3.2/ccminer_cuda11.exe
This release support already rtx 30xx

[STRATZ-Ken]

Weird, when I try to run it on my 3070 on the latest, it just sits there, no errors. I tried on my 2070 (same version) and it works fine.

[djm34]

what do you mean by the latest ?
with cuda 11.1, both card works for me:

[2021-01-26 11:23:01] GPU #1: Intensity set to 21.1719, 2457600 cuda threads number of multiproc 15
[2021-01-26 11:23:02] GPU #0: Intensity set to 23.2812, 10747904 cuda threads number of multiproc 82
[2021-01-26 11:23:05] GPU #1: EVGA GTX 1070, 653.14 kH/s
[2021-01-26 11:23:05] GPU #0: Gigabyte RTX 3090, 2664.97 kH/s
[2021-01-26 11:23:09] GPU #1: EVGA GTX 1070, 2528.56 kH/s
[2021-01-26 11:23:10] GPU #0: Gigabyte RTX 3090, 7089.09 kH/s

make sure you don't select a particular card with "-d " in the command line. may-be
I can't check (before next week) with a combination of rtx20xx and rtx30xx needs more fitting to put back my 2080ti in the system... but usually cards are treated independently...

[bitfabrikken]

Just tried compiling and when starting it also just sits there. No activity in nvtop.
Doesn't matter if I use -d or not.
And seems like no matter which algo I use, it says it's using blake.

Debian GNU/Linux bullseye/sid, Linux 5.10.0-3-amd64
ASRock Z68 Pro3 with bios P.2.30 + Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz
GPU's: RTX 2060 rev a, GTX 960, GTX 980 Ti
Cuda compilation tools, release 11.2, V11.2.142
Build cuda_11.2.r11.2/compiler.29558016_0
ccminer-1.3.2

me@mine:~/mine/ccminer-firoorg/ccminer-1.3.2$ ./ccminer -a mtp --devices 0
*** ccminer 1.3.2L-djm34 for nVidia GPUs by djm34 ***
    Built with the nVidia CUDA Toolkit 11.2

  Originally based on Christian Buchner and Christian H. project based on tpruvot 1.8.4 release
  Include algos from alexis78, djm34, sp, tsiv and klausT.
  *** News (07/06/2018): MTP algo for ZCoin

  MTP algo based on krnlx kernel

  BTC donation address: 1NENYmxwZGHsKFmyjTc5WferTn5VTFb7Ze (djm34)
  ZCoin donation address: aChWVb8CpgajadpLmiwDZvZaKizQgHxfh5 (djm34)

[2021-02-28 20:06:40] POOL 0: stratum.eu.miners-pool.eu:8421 USER ShGCAFrspnLDo414EzUyJm4k5Qr2Kkrmvh -s 30
[2021-02-28 20:06:40] Starting on stratum+tcp://stratum.eu.miners-pool.eu:8421
[2021-02-28 20:06:40] 1 miner thread started, using 'blake' algorithm.