MFAKTC wont run on a GTX 980

[Peter]

Hi everybody,

I just got my new GTX 980 up and running but sadly when I try to run MFAKTC, I end up with the following output:

C:\Users\Peter\Desktop\mfaktc-0.20>mfaktc-win-64.exe
mfaktc v0.20 (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              64Mi bits
  GPUSieveProcessSize       16Ki bits
  WorkFile                  worktodo.txt
  Checkpoints               enabled
  CheckpointDelay           30s
  Stages                    enabled
  StopAfterFactor           bitlevel
  PrintMode                 full
  V5UserID                  (none)
  ComputerID                (none)
  AllowSleep                no
  TimeStampInResults        no

CUDA version info
  binary compiled for CUDA  4.20
  CUDA runtime version      6.0
  CUDA driver version       6.50
ERROR: CUDA runtime version must match the CUDA toolkit version used during compile!

I was not able to find MFAKTC-binaries that have been compiled for:

CUDA version info
  binary compiled for CUDA  4.20
  CUDA runtime version      6.0
  CUDA driver version       6.50

Can somebody help me out with this? :-)

[Mark Rose]

Grabbing a newer CUDA library should work: https://developer.nvidia.com/cuda-do...geforce-gtx9xx

[kracker]

Originally Posted by Mark Rose

Grabbing a newer CUDA library should work: https://developer.nvidia.com/cuda-do...geforce-gtx9xx

I think mfaktc just needs to be recompiled for the new CUDA 6.x.

[Peter]

Originally Posted by kracker

I think mfaktc just needs to be recompiled for the new CUDA 6.x.

Can I do this without having to install some full blown development environment?
Or could you maybe do me this favour? :)

[Batalov]

Originally Posted by Peter

Can I do this without having to install some full blown development environment?

More or less, yes if you are running lin*x.
Almost certainly, no if you are running wind*ws.

Originally Posted by Peter

Or could you maybe do me this favour? :)

You may want to ask user FlashJH. Try Options::Send new message. (Options tab at the top left.)

[Peter]

Thanks Batalov.
I hope he's going to answer me. :)

[flashjh]

PM received :)

I have the ability to compile; I'll try it later tonight and I'll keep you posted.

[axn]

Originally Posted by kracker

I think mfaktc just needs to be recompiled for the new CUDA 6.x.

If you compile with the right set of options, the compiler will leave some intermediate code in the binary, from which all future CUDA driver versions can recompile an appropriate version -- no need to recompile for each an every CUDA release.

[flashjh]

Agreed, I compile with all available real GPU and virtual PTX architecture switches. In this case the GTX 9XX series requires the newest version of CUDA 6.5 that was just released.

I attached the newest 64-bit Windows version. Give it a test and post your results. Make sure you're using the newest drivers. If you run into a missing library file, download it from here.

Happy TFing

Can you post this updated info from your test run:

CUDA version info
  binary compiled for CUDA  4.20
  CUDA runtime version      6.0
  CUDA driver version       6.50

Thanks.

[Peter]

Thank you FlashJH. It is working now.
Its running at mindblowing 515-535 GHz-d/day.
Finishing off a 70-71 search takes about 6m20s.
Great!

[VictordeHolland]

Do you happen to have a wattmeter, I'm curious how much it consumes running mfaktc?