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

[djm34]

CCMiner is released.

No -d flag? Does -t flag use first GPU or first available GPU?

uses all GPUs by default. Uses less if you give a smaller -t. No GPU selection, sorry.
Source code is available to tinker.

All of this was hacked in 9 days, and Fugue within approximately 24 hours. Only
minimal cleanup afterwards.

Christian

Interesting, on fugue256, the gtx780ti gives 232Mhash/s and clearly beats the R9290x which does only 157Mhash/s on x64 it is somewhat slower on x32

[cbuchner1]

Interesting, on fugue256, the gtx780ti gives 232Mhash/s and clearly beats the R9290x which does only 157Mhash/s

yes, we've done a midstate optimization, the OpenCL code does the full hashing on GPU. If they optimize, your GTX 780Ti is toast.

BTW I am only getting 175 MHash/s per 780Ti on Linux (550 MHash/s on my rig of 3) - but I have no OC options there.

Christian

[ltcnim]

what khash can i expect from a gtx660 or a 750Ti @ HVC?

[jack80]

Im go 9000 kh/s on 2 Nvidia 750 ti Overclocked at 180 gpu and 570 mem ( 4500 kh/s at one ) , It's ok ? .

Im use :

ccminer.exe -a heavy -o stratum+tcp://stratum01.heavycoinpool.com:5333 -u username.worker -p workerpassword -v 512

On X86 Directory .

[Chris84]

Im go 9000 kh/s on 2 Nvidia 750 ti Overclocked at 180 gpu and 570 mem ( 4500 kh/s at one ) , It's ok ? .

Im use :

ccminer.exe -a heavy -o stratum+tcp://stratum01.heavycoinpool.com:5333 -u username.worker -p workerpassword -v 512

On X86 Directory .

Sounds good, my 750Ti did ~4000...
If you want, you could try to compile it with compute 3.5, maybe this could speedup the things a Little bit...

[JimmyD]

All of this was hacked in 9 days, and Fugue within approximately 24 hours. Only
minimal cleanup afterwards.

Christian

I just got a yay!!! on FC after such a long time mining, finally! Thanks Christian! :-)

[sin242]

Initial CCminer results on fugue:

780 (very mild oc) - ~170mh

750ti (+100core, 500mem) ~ 47mh (seems a bit on the low side.  might need > comp 2.0 )

670gtx (200 core, 400mem) ~52mh-95mh  (2 of the 670s are doing low 50s (2 pnys) however a stock clocked evga 670 gtx in another system is doing 95mh.  Don't have time atm
to explore this. )

650ti (stock) ~52mh (rawr.)

[cbuchner1]

Initial CCminer results on fugue:
780 (very mild oc) - ~170mh
750ti (+100core, 500mem) ~ 47mh (seems a bit on the low side.  might need > comp 2.0 )

yes, the performance really scales with number of SMX. So the 750 Ti loses bad against a 780 Ti. Such is life.

[zelante]

660Ti (msi afterburner: +125 core, +580 memory, 114 power limit, temperature 57C) ~6000 khash/s (on heavycoinpool)

Can anyone compile with compute 3.0, plz?

[djm34]

Ok, I tried to compile with compute 3.5, I don't see any difference. Running now at 240Mhash/s at 1270MHz (was running 230MHash/s at 1250MHash/s)
The GPU usage is rather low 93% and the power doesn't go further than 99% (even if it allows to).

Still didn't find a block with it. (2 blocks were found by the other cards since I started the gtx780, had only one boo with it)

[debido666]

[sin242]

Initial CCminer results on fugue:
780 (very mild oc) - ~170mh
750ti (+100core, 500mem) ~ 47mh (seems a bit on the low side.  might need > comp 2.0 )

yes, the performance really scales with number of SMX. So the 750 Ti loses bad against a 780 Ti. Such is life.

Still crushes my 5870+6970 combined lol

[cbuchner1]

660Ti (msi afterburner: +125 core, +580 memory, 114 power limit, temperature 57C) ~6000 khash/s (on heavycoinpool)

Can anyone compile with compute 3.0, plz?

When building for Compute 3.5 with the intention to mine Heavycoin, try a max reg. count of 128.
Some kernels may need so much register space. I found it may make a 10-20% difference to target
Compute 3.5 on platforms like the GTX 780 (Ti)...

Compute 3.0 only supports a max reg. count 63 unfortunately.

I will investigate how future Windows builds can include kernels for all platforms simultaneously.
There's got to be a way.

Christian

[cbuchner1]

Ok, I tried to compile with compute 3.5, I don't see any difference. Running now at 240Mhash/s at 1270MHz (was running 230MHash/s at 1250MHash/s)
The GPU usage is rather low 93% and the power doesn't go further than 99% (even if it allows to

if you built on Linux, try putting sm_35 instead of compute_35 into Makefile.am (and rerun ./autogen.sh). The latter case only generates PTX code from the nvcc compiler which then later the driver translates into binary form - whereas in the sm_35 option nvcc generates the binary kernels (with full register assignments and optimizations). This may be more performant, as nvcc does smarter optimization.

[djm34]

Ok, I tried to compile with compute 3.5, I don't see any difference. Running now at 240Mhash/s at 1270MHz (was running 230MHash/s at 1250MHash/s)
The GPU usage is rather low 93% and the power doesn't go further than 99% (even if it allows to

if you built on Linux, try putting sm_35 instead of compute_35 into Makefile.am (and rerun ./autogen.sh). The latter case only generates PTX code from the nvcc compiler which then later the driver translates into binary form - whereas in the sm_35 option nvcc generates the binary kernels (with full register assignments and optimizations). This may be more performant, as nvcc does smarter optimization.

I am on windows, I replaced both compute_10 and sm_10 with compute_35 and sm_35

[cbuchner1]

All of this was hacked in 9 days, and Fugue within approximately 24 hours. Only
minimal cleanup afterwards.

Christian

I just got a yay!!! on FC after such a long time mining, finally! Thanks Christian! :-)

I am still getting about 10 Yays per hour on each rig of ca. 500 MHash/s.
It felt like one per minute per Yay on each rig when I was having half the network hash rate.  ;-)

[fruitsdemers]

Question:  Can I expect better hashrates compiling for 3.0/3.5 on a 780ti ?  Using this release I am getting about 22000 khash/s  on a 2x 780ti & 3930K system.

what coin?

Heavy.

Thanks

Same situation.
It doesn't seem to use more than 77% of the gpu though.

[cbuchner1]

Same situation.
It doesn't seem to use more than 77% of the gpu though.

Google Chrome Trick maybe?

Also consider running 2 instances side by side. It helped to get some extra 10-20% sometimes.

[jack80]

I dont find anything HVC in 30 minutes on heavycoinpool.com and 556650 Shares submitted , and 15000 kh/s ( 3 Nvidia 750 ti ) .

[cbuchner1]

I am on windows, I replaced both compute_10 and sm_10 with compute_35 and sm_35

It seems I set compute_10,sm_10 on the cuda_hefty.cu module in the Visual Studio
project. I believe it was done because we had some stack overruns in the compiler with
some earlier versions of cuda_hefty.cu leading to a crash during compilation.

if the compiler doesn't crash for you by all means set it back or remove the options from
the vcxproj entirely, so the project wide settings take over. ;-)

Christian