[ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX]

[ltcnim]

ccMiner v0.2 now supports pools with fuguecoin (like dwarfpool) and has a nice boost for Fermi and Maxwell owners.

https://github.com/cbuchner1/ccminer/releases/tag/v0.2

Christian

does it also speed up heavycoin mining? or is the speedup for fugue only? thx for the release!

[ManIkWeet]

Doesn't cost much, to do the launch and grab a few blocks. Just in case.

What would be the scrypt-n parameters though?
I have never been on a N-factor coin launch

you haven't been on any vertcoin, exe, gpuc ? 

Correct

cudaminer.exe --algo=scrypt:2048  -q -m1 -d gtx780 -b32768 -C2 -lt15x32 -i 0 -H 2
(try autotune first, I think for some cards the T kernel is better)

Is it always scrypt:2048 at a coin launch? I'll try with this then.

[bigjme]

Right guys, has anyone got a mining system at hand that's using usb risers?

I'm interested in a few picture of something on the rig. So if you can get pictures and have used usb risers, mainly using them in sockets other then x1 please drop me a message. I want to check something before I go and order some cards and risers

[cbuchner1]

ccMiner v0.2 now supports pools with fuguecoin (like dwarfpool) and has a nice boost for Fermi and Maxwell owners.

https://github.com/cbuchner1/ccminer/releases/tag/v0.2

Christian

does it also speed up heavycoin mining? or is the speedup for fugue only? thx for the release!

heavycoin should be more or less unchanged. Except we switched the hefty1 kernel from sm1.0 to sm2.0 on the Windows build - but I don't think that makes a difference.

Christian

[guitarplinker]

I've heard that AMD based machines have troubles with four cards and CUDAMiner sometimes, is this true?

I'm looking at purchasing an AMD based CPU and board for quad 750 Ti's and don't want to run into any issues with them.

[RavenX]

ccMiner v0.2 now supports pools with fuguecoin (like dwarfpool) and has a nice boost for Fermi and Maxwell owners.

https://github.com/cbuchner1/ccminer/releases/tag/v0.2

Christian

Will try now.

Just a minor correction needed: the README file mention cudaminer.exe in the examples.

;-)

[ltcnim]

is the CudaMiner master branch broken atm.? I'm getting an error when compiling under ubuntu 13.10:

Code:

g++  -g -O2 -pthread -L/usr/local/cuda/lib64  -o cudaminer cudaminer-cpu-miner.o cudaminer-util.o wrapnvml.o cudaminer-sha2.o cudaminer-scrypt.o cudaminer-maxcoin.o cudaminer-blakecoin.o cudaminer-sha3.o cudaminer-scrypt-jane.o salsa_kernel.o sha256.o keccak.o blake.o cudaminer-blake.o fermi_kernel.o kepler_kernel.o test_kernel.o nv_kernel.o nv_kernel2.o titan_kernel.o -L/usr/lib/x86_64-linux-gnu -lcurl compat/jansson/libjansson.a -lpthread  -lcudart -fopenmp -lcrypto -lssl  -lcrypto -lssl
/usr/bin/ld: wrapnvml.o: undefined reference to symbol 'dlclose@@GLIBC_2.2.5'
/lib/x86_64-linux-gnu/libdl.so.2: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status

[djm34]

ccminer does not work well for me.
The power litteraly shut off after a while (goes to 38%), then the driver crashes (not all the time, but I need to restart, then it get back in the same state after 10minutes or so...).

[cbuchner1]

is the CudaMiner master branch broken atm.? I'm getting an error when compiling under ubuntu 13.10:

Code:

g++  -g -O2 -pthread -L/usr/local/cuda/lib64  -o cudaminer cudaminer-cpu-miner.o cudaminer-util.o wrapnvml.o cudaminer-sha2.o cudaminer-scrypt.o cudaminer-maxcoin.o cudaminer-blakecoin.o cudaminer-sha3.o cudaminer-scrypt-jane.o salsa_kernel.o sha256.o keccak.o blake.o cudaminer-blake.o fermi_kernel.o kepler_kernel.o test_kernel.o nv_kernel.o nv_kernel2.o titan_kernel.o -L/usr/lib/x86_64-linux-gnu -lcurl compat/jansson/libjansson.a -lpthread  -lcudart -fopenmp -lcrypto -lssl  -lcrypto -lssl
/usr/bin/ld: wrapnvml.o: undefined reference to symbol 'dlclose@@GLIBC_2.2.5'
/lib/x86_64-linux-gnu/libdl.so.2: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status

A -l dl or -ldl (??)  linker option in the Makefile.am might help  (don't forget ./autogen.sh afterwards)

[AizenSou]

Some reports for Christian:  
ccminer 0.2 fugue256 with dwarfpool

- GTX 780: 180Mhs
- GTX 680: 90Mhs
- GTX 580: 120-130Mhs   (better than 680)
- GTX 570: 100Mhs
- GTX 750Ti: 90-95Mhs. On other rig with 4x750Tis i get 500Mhs, that's mean 125Mhs for 1 750Ti. That's crazy. Still don't understand why 5x750Tis rig is slower? Because they are on risers??

To be short, you are terrible Christian  

Thanks for this new beast.

PS: Still don't understand why Kepler is so bad
 

[jk_14]

.

Hi,
There is someone raporting such errors:

Code:

[2014-03-21 17:52:08] GPU #0: Performing auto-tuning (Patience...)
[2014-03-21 17:52:08] GPU #2: cudaError 2 (out of memory) calling 'cudaFree(d_idata)' (salsa_kernel.cu line 728)

[2014-03-21 17:52:08] GPU #2:    0.00 hash/s with configuration T0x0
[2014-03-21 17:52:08] GPU #2: using launch configuration T0x0
[2014-03-21 17:52:08] GPU #0: maximum total warps (BxW): 480
[2014-03-21 17:52:08] GPU #2: cudaError 2 (out of memory) calling 'cudaMalloc((void **) &tmp, mem_size)' (salsa_kernel.cu line 251)

[2014-03-21 17:52:09] GPU #2: cudaError 2 (out of memory) calling 'cudaMalloc((void **) &tmp, mem_size)' (salsa_kernel.cu line 252)

[2014-03-21 17:52:09] GPU #2: cudaError 2 (out of memory) calling 'cudaMalloc((void **) &tmp, mem_size)' (salsa_kernel.cu line 253)

[2014-03-21 17:52:09] GPU #1: maximum total warps (BxW): 0
[2014-03-21 17:52:09] GPU #1:    0.00 hash/s with configuration T0x0
[2014-03-21 17:52:09] GPU #1: using launch configuration T0x0
Floating point exception (core dumped)

and this:

"I occasionally get some memory error on the startup as well."

https://litecointalk.org/index.php?action=profile;area=showposts;u=28846


Is it already known problem? Hardware related?
(after booting 4GB RAM system, usually: ~1GB is used, ~3GB is free...)

[ltcnim]

is the CudaMiner master branch broken atm.? I'm getting an error when compiling under ubuntu 13.10:

Code:

g++  -g -O2 -pthread -L/usr/local/cuda/lib64  -o cudaminer cudaminer-cpu-miner.o cudaminer-util.o wrapnvml.o cudaminer-sha2.o cudaminer-scrypt.o cudaminer-maxcoin.o cudaminer-blakecoin.o cudaminer-sha3.o cudaminer-scrypt-jane.o salsa_kernel.o sha256.o keccak.o blake.o cudaminer-blake.o fermi_kernel.o kepler_kernel.o test_kernel.o nv_kernel.o nv_kernel2.o titan_kernel.o -L/usr/lib/x86_64-linux-gnu -lcurl compat/jansson/libjansson.a -lpthread  -lcudart -fopenmp -lcrypto -lssl  -lcrypto -lssl
/usr/bin/ld: wrapnvml.o: undefined reference to symbol 'dlclose@@GLIBC_2.2.5'
/lib/x86_64-linux-gnu/libdl.so.2: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status

A -l dl or -ldl (??)  linker option in the Makefile.am might help  (don't forget ./autogen.sh afterwards)

cudaminer_LDFLAGS       = $(PTHREAD_FLAGS) @CUDA_LDFLAGS@ -l
cudaminer_LDFLAGS       = $(PTHREAD_FLAGS) @CUDA_LDFLAGS@ -ld
cudaminer_LDFLAGS       = $(PTHREAD_FLAGS) @CUDA_LDFLAGS@ -ldl

none of them worked

[cbuchner1]

PS: Still don't understand why Kepler is so bad

Kepler's shared memory bandwidth is underdimensioned for the size of the SMX (192 CUDA cores).
It has essentially the same bandwidth as Fermi with 32/48 CUDA cores.

So per core, shared is much faster on Fermi. This is why Fermi wins here.

We put the lookup tables for Fugue256 into shared memory (it was actually slower reading
this data from texture memory in the v0.1 version - despite being cached).

[cvax]

Anyone recently mined Micro Coin with a 750ti, does 37kh/s sound about right?

I believe the current N-Factor is 11.

Micro should be about 60kh/s on 750tis.

[djm34]

Just a few more detail regarding the problem I have:
As I said after a while, the power which was around 84%, drop to 37% and at the same time the cpu usage which was oscillating around 80% climb to 100% and obviously the card stop working.

The card is not plugged into a monitor and does not use the boost feature (skynet mod), so that strange.

This is with the x86 version.
With the x64 version, the program does not start. It displays the pool address, and crash after 10 seconds or so.

[AizenSou]

PS: Still don't understand why Kepler is so bad

Kepler's shared memory bandwidth is underdimensioned for the size of the SMX (192 CUDA cores).
It has essentially the same bandwidth as Fermi with 32/48 CUDA cores.

So per core, shared is much faster on Fermi. This is why Fermi wins here.

We put the lookup tables for Fugue256 into shared memory (it was actually slower reading
this data from texture memory in the v0.1 version - despite being cached).

Thanks for the explanation Christian.
I still can't reproduce the strange phenomenon with one of my 750Ti rig. I pretty sure that with 4x750tis i got 500Mhs for at least 1/2 hour. Then I restart the miner and it goes down to 350Mhs. I have restarted the rig but it is still 350Mhs. Strange.

[RavenX]

To be short, you are terrible Christian  

Thanks for this new beast.

+1

[cbuchner1]

I have restarted the rig but it is still 350Mhs. Strange.

that's what you get for touching it while it was running perfectly

[ivanlabrie]

Yeah! Close to 300mh/s mining fugue (780+750ti) 

You're baws Christian!

[becker]

\ccminer-master\miner.h(34): fatal error C1083: Cannot open include file: 'alloca.h': No such file or directory

Tried to compile it on windows VS Studio 2010 Prof. Says missing that. . I compiled Version 1 fine.