mfaktc: a CUDA program for Mersenne prefactoring

[lolapus]

Originally Posted by Dylan14

The pkgbuild for mfaktc has been updated. A new CUDA version, 11.2 came out which requires a new driver which has not been fully released (it is in beta). Hence the dependencies have been updated.
For anyone running the beta driver (presently 460.27.04), here is a mfaktc executable with CUDA 11.2 support on Linux.

I know I asked this a while ago but would you be able to compile mfaktc for cuda 11.2 on windows so I can use my 3090 to do tf instead of prp? I'm using my own desktop pc so installing another os wouldn't be an option for me.

[lolapus]

Ok, after a few hours of tinkering with cuda 11.2 and visual studio 2012 I managed to get a 64-bit cuda 11.2 build of mfaktc. I'm not sure if I'm supposed to include a dll for cuda 11.2 with the build so let me know if I do. Attached is the build with the readme, changelog, copying, ini, and woktodo files. Hopefully this helps out a few people with the RTX 3000s on windows.

[kriesel]

Originally Posted by lolapus

Ok, after a few hours of tinkering with cuda 11.2 and visual studio 2012 I managed to get a 64-bit cuda 11.2 build of mfaktc.

Is that v0.21 with the gpusievesize max extended to 2047M?

[lolapus]

Originally Posted by kriesel

Is that v0.21 with the gpusievesize max extended to 2047M?

When I run mfaktc it says the gpu sievesize is 64Mi bits but I think you can change the sievesize in the ini file. I didn't change anything from the source except for the cuda and vs build tools locations for compiling so I guess the answer is no.

[lolapus]

I recompiled mfaktc for cuda 11.2 on windows that has the GPUSievesizemax increased to 2047 and now the value in the ini file is set to 2047. I did see significant performance improvements and might be looking to increase the value even further in the future.

[joejoefla]

I tried to run the program on my machine. I have nVidia driver 461.72 and installed the cudart64_110.dll file from the CUDA toolkit. Getting this error:

mfaktc v0.21 (64bit built)

Compiletime options
THREADS_PER_BLOCK 256
SIEVE_SIZE_LIMIT 32kiB
SIEVE_SIZE 193154bits
SIEVE_SPLIT 250
MORE_CLASSES enabled

Runtime options
SievePrimes 25000
SievePrimesAdjust 1
SievePrimesMin 5000
SievePrimesMax 100000
NumStreams 3
CPUStreams 3
GridSize 3
GPU Sieving enabled
GPUSievePrimes 82486
GPUSieveSize 2047Mi bits
GPUSieveProcessSize 16Ki bits
Checkpoints enabled
CheckpointDelay 30s
WorkFileAddDelay 600s
Stages enabled
StopAfterFactor bitlevel
PrintMode full
V5UserID (none)
ComputerID (none)
AllowSleep no
TimeStampInResults no

CUDA version info
binary compiled for CUDA 11.20
CUDA runtime version 11.20
CUDA driver version 11.20

CUDA device info
name GeForce GTX 1080
compute capability 6.1
max threads per block 1024
max shared memory per MP 98304 byte
number of multiprocessors 20
clock rate (CUDA cores) 1733MHz
memory clock rate: 5005MHz
memory bus width: 256 bit

Automatic parameters
threads per grid 655360
GPUSievePrimes (adjusted) 82486
GPUsieve minimum exponent 1055144

running a simple selftest...
ERROR: cudaGetLastError() returned 209: no kernel image is available for execution on the device

[James Heinrich]

Originally Posted by joejoefla

CUDA device info
name GeForce GTX 1080
compute capability 6.1

I think the above build is targeted for RTX 30x0 hardware only, I don't think you'd see any improvement over the previous 2047-enabled builds.

[joejoefla]

Originally Posted by James Heinrich

I think the above build is targeted for RTX 30x0 hardware only, I don't think you'd see any improvement over the previous 2047-enabled builds.

Sorry my bad. I'll switch back.

[ZacHFX]

Originally Posted by lolapus

I recompiled mfaktc for cuda 11.2 on windows that has the GPUSievesizemax increased to 2047 and now the value in the ini file is set to 2047. I did see significant performance improvements and might be looking to increase the value even further in the future.

Just got this up and running on my 3080. Sent benchmarks to James. Let me know if there's something specific I can do to assist.

[James Heinrich]

Originally Posted by ZacHFX

Just got this up and running on my 3080. Sent benchmarks to James. Let me know if there's something specific I can do to assist.

Thanks! 6 months after launch you're the first person to submit a 3080 benchmark...

[ZacHFX]

Originally Posted by James Heinrich

Thanks! 6 months after launch you're the first person to submit a 3080 benchmark...

Mining ether is more profitable than factoring, sadly.