mfaktc: a CUDA program for Mersenne prefactoring

[James Heinrich]

Originally Posted by Icecold

Will do, I should be able to send over cudalucas and gpuowl in the next couple days, and mfaktc as soon as I can get that working... this is a fresh install of Windows 10.

Thank you kindly. If you need a CUDA 8.6 build of mfaktc to get it working, and you do get one built, please share it (attached here is fine) and I'll put it up on download.mersenne.ca

[Icecold]

Originally Posted by axn

Yes, that's what I meant. Did that solve the "no kernel image" issue? I'm assuming you're able to use it as anonymous platform under BOINC

It did not, unfortunately. I had done that prior(or around the same time) to rebirther posting here about the issue. It was giving me an error compiling until I added 86 in there (I should have screenshotted the error, but I didn't and don't remember what it was)

[rebirther]

Is someone able to compile a cuda11 version for windows and linux?

[aheeffer]

I just received a RTX 3060 ti card (Gigabyte Aorus) but can't get it to run mfaktc


CUDA version info
binary compiled for CUDA 10.0
CUDA runtime version 10.0
CUDA driver version 11.10

CUDA device info
name GeForce RTX 3060 Ti
compute capability 8.6
max threads per block 1024
max shared memory per MP 102400 byte
number of multiprocessors 38
clock rate (CUDA cores) 1800MHz
memory clock rate: 7001MHz
memory bus width: 256 bit

Automatic parameters
threads per grid 622592
GPUSievePrimes (adjusted) 82486
GPUsieve minimum exponent 1055144

########## testcase 1/2867 ##########
Starting trial factoring M50804297 from 2^67 to 2^68 (0.59 GHz-days)
Using GPU kernel "75bit_mul32_gs"
Date Time Pct ETA | Exponent Bits | GHz-d/day Sieve Wait
Dec 04 15:37 0.1 n.a. | 50804297 67-68 | n.a. 82485 n.a.%
ERROR: cudaGetLastError() returned 48: no kernel image is available for execution on the device


What is wrong? Do I need a version compiled for CUDA 11? Who has one for Windows?

[clowns789]

Originally Posted by aheeffer

What is wrong? Do I need a version compiled for CUDA 11? Who has one for Windows?

That would be nice. I also have a 3060 Ti (MSI) but the window closes as soon as I open it. It worked fine on a GTX 1080.

[aheeffer]

It took me a while, but I succeeded in building a 64-bit Windows 10 executable using the NVIDIA GPU Computing Toolkit v.11.1, Visual Studio 2019 Community and Msys64 for mfaktc adding the new 8.0 and 8.6 architecture (which my RTX 3060 ti uses).

Now, I get a CUDA runtime error.

CUDA version info
  binary compiled for CUDA  11.10
  CUDA runtime version      11.10
  CUDA driver version       11.10

CUDA device info
  name                      GeForce RTX 3060 Ti
  compute capability        8.6
  max threads per block     1024
  max shared memory per MP  102400 byte
  number of multiprocessors 38
  clock rate (CUDA cores)   1800MHz
  memory clock rate:        7001MHz
  memory bus width:         256 bit

Automatic parameters
  threads per grid          622592
  GPUSievePrimes (adjusted) 82486
  GPUsieve minimum exponent 1055144

########## testcase 1/2867 ##########
Starting trial factoring M50804297 from 2^67 to 2^68 (0.59 GHz-days)
Using GPU kernel "75bit_mul32_gs"
Date    Time | class   Pct |   time     ETA | GHz-d/day    Sieve     Wait
Dec 06 14:53 | 3387   0.1% |  0.001    n.a. |      n.a.    82485    n.a.%
ERROR: cudaGetLastError() returned 98: invalid device function

Any advice?

BTW, I also get the same error running on a RTX 2060 card.

[pinhodecarlos]

Go here http://srbase.my-firewall.org/sr5/index.php, if I’m not mistaken they have a working version.

[rebirther]

Originally Posted by pinhodecarlos

Go here http://srbase.my-firewall.org/sr5/index.php, if I’m not mistaken they have a working version.

nope, one has compiled a win version and run into the same error while linux is ok.

[axn]

Ok, so a bit of a long shot.

In mfaktc.c, under int tf(mystuff_t *mystuff, int class_hint, unsigned long long int k_hint, int kernel)

there is a section...

        if(cudaError != cudaSuccess)
        {
          printf("ERROR: cudaGetLastError() returned %d: %s\n", cudaError, cudaGetErrorString(cudaError));
          return RET_CUDA_ERROR; /* bail out, we might have a serios problem (detected by cudaGetLastError())... */
        }

modify that to...

        if(cudaError != cudaSuccess && cudaError != cudaErrorInvalidDeviceFunction)
        {
          printf("ERROR: cudaGetLastError() returned %d: %s\n", cudaError, cudaGetErrorString(cudaError));
          return RET_CUDA_ERROR; /* bail out, we might have a serios problem (detected by cudaGetLastError())... */
        }

to ignore that specific error. See if that allows you to get past the error and pass the self tests (-st and -st2). If all checks out, try a few tests with knows factors. If all of that works out, keep going with it.

I'm assuming that that particular error is benign (considering linux works).

[Dylan14]

Two things:

1. I have created a PKGBUILD for mfaktc for use on Arch Linux. You can find it here: https://aur.archlinux.org/packages/mfaktc/

2. Attached is a build of mfaktc compiled on cuda 11.1 on Linux, with a max GPU sieve size of 2047. It supports all CC's that is supported by this version of cuda:
3.5 (some Kepler cards, like the K80)
5.x (Maxwell)
6.x (Pascal)
7.0 (the Titan V and the Tesla V100)
7.5 (Volta)
8.0 (the Tesla A100)
8.6 (Ampere)

[aheeffer]

Originally Posted by axn

Ok, so a bit of a long shot.

ignore that specific error. See if that allows you to get past the error and pass the self tests (-st and -st2). If all checks out, try a few tests with knows factors. If all of that works out, keep going with it.

I'm assuming that that particular error is benign (considering linux works).

Well, I gave it a try....

Ignoring the CudaError, the program goes through all the test TF's and they all fail.