CCminer(SP-MOD) Modded GPU kernels.

[scryptr]

[quote/]
<< pictures and repetetive stuff cut >>

those are damn good hash rates ...

i get about 2700KH each on the powered gigabyte 750ti oc and about 2600KH each for the unpowered low profiles ones in x11 ... using the latest build of ccminer compiled under 64bit linux ...

this is with fedora 19 x64 and each of those machines are running 875W thermaltake bronze ...

im sure the evga cards are supposed to be better than the gigabyte ones when it comes to hashrates ... anyway - its only a few hashes diff ...

still - the hashrates seem to be progressing with ccminer ...

#crysx
[/quote]

I don't know the stock clocks on the Gigabyte cards, and I can't remember the EVGA specs.  Either of us could look them up.  The EVGA FTW is the best model 750ti they make, and the EVGA SC is their best 750ti card without 6-pins.  You could run 8 cards with those power supplies.  The EVGA FTW rig draws less than 500 watts at the wall.  You certainly could run bigger cards with those power supplies.

And, yes, I was very suprized to see real improvements in error rates, power consumption, and hash rates in recent code.  The FTW rig tops out at under 550 watts on power-hungry algorithms like scrypt.       --scryptr

[chrysophylax]

[quote/]
<< pictures and repetetive stuff cut >>

those are damn good hash rates ...

i get about 2700KH each on the powered gigabyte 750ti oc and about 2600KH each for the unpowered low profiles ones in x11 ... using the latest build of ccminer compiled under 64bit linux ...

this is with fedora 19 x64 and each of those machines are running 875W thermaltake bronze ...

im sure the evga cards are supposed to be better than the gigabyte ones when it comes to hashrates ... anyway - its only a few hashes diff ...

still - the hashrates seem to be progressing with ccminer ...

#crysx

I don't know the stock clocks on the Gigabyte cards, and I can't remember the EVGA specs.  Either of us could look them up.  The EVGA FTW is the best model 750ti they make, and the EVGA SC is their best 750ti card without 6-pins.  You could run 8 cards with those power supplies.  The EVGA FTW rig draws less than 500 watts at the wall.  You certainly could run bigger cards with those power supplies.

And, yes, I was very suprized to see real improvements in error rates, power consumption, and hash rates in recent code.  The FTW rig tops out at under 550 watts on power-hungry algorithms like scrypt.       --scryptr
[/quote]

hehehe ... images cut out well ...

yup - the power draw is so minimal ...

i just looked at the specs and they are very similar ... gigabyte make a series of cards that go to massive oc on their top level 750ti card - so i guess its a much of a muchness as the saying goes ...

for the power consumption that the farm is drawing at the moment ( i work it in amps as that is the max the property will draw before a blowout ) means that i can replace the whole farm with these cards and draw the same power yet have a lot more hashes ...

id be very interested to know how neoscrypt goes with this also - when its implemented and optimized here in ccminer ... need to test the other algo's also when the farm is redone ...

#crysx

[tbearhere]

I compiled your latest source, and get worse results than from tpruvots latest on my GTX760.
http://snag.gy/uBKmK.jpg - http://snag.gy/BfP3L.jpg

It's better on the 750tis, that's for sure
http://snag.gy/e6KjJ.jpg - http://snag.gy/FjUyD.jpg

Bombadil (KBomba) seems to be working on "ccmonitor" code just recently.

Previews are up there ^^ Anyone likes tabs as much as I do?

I love that monitor, what is it? Can it be used with windows 8? And where can I get it please.

[sp_]

Ccminer should be compiled.for x86 and not.x64 for best results.

[chrysophylax]

Ccminer should be compiled.for x86 and not.x64 for best results.

how would i be able to do that under fedora? ...

just install a x86 fedora system then compile in there - or is there an easier way inside x64 fedora? ...

im guessing you're going to say a new box with x86 native install ... i hope im wrong ...

wont be an issue to do so - i have a few spare machines here - but its just time consuming ... and im a lazy bugga ... :|

#crysx

[djm34]

Ccminer should be compiled.for x86 and not.x64 for best results.

how would i be able to do that under fedora? ...

just install a x86 fedora system then compile in there - or is there an easier way inside x64 fedora? ...

im guessing you're going to say a new box with x86 native install ... i hope im wrong ...

wont be an issue to do so - i have a few spare machines here - but its just time consuming ... and im a lazy bugga ... :|

#crysx

this comment is for windows not linux

[chrysophylax]

phew ...



#crysx

[scryptr]

This is awesome:


imag

100% accepted, 980 at +185MHz core, best setting for temperature (around 55ºC max) and no voltage increase.

Cheers

X11 HASH RATES - AMD 280X vs nVidia 750ti--

Here is my 6x 280X rig:



Here is my fast 750ti rig:



Here is my slower 750ti rig:



I think that any more improvements to the Cuda code will take the 750ti past the 280X. My GPU0 280X is turned off due to frequent crashes.  The sgminer I use is a set of Linux binaries released just before Christmas by NiceHash.  They include some Wolf code, I think.  His top level code is not in general use, so, pricewise and energy wise, Cuda wins! 

--scryptr

[HoY]

scryptr you are getting about 40-45% of expected performance from your 280x's. with even mild over clocking (1100/1500) they do 6.5mh/s per card using Wolf's recently released bin files.
To save you the searching here's a link to his post on reddit

http://www.reddit.com/r/DRKCoin/comments/2o1yoz/rewritten_x11_binaries/

[djm34]

some progress on neoscrypt:
750ti: 81kh/s
780ti: 165kh/s
980:   290kh/s
(yeah well, doesn't scale right  )

[HoY]

There were numerous other updated kernels (that you pointed me at) that were publicly available even before that, I was already at 4.5mh/s on my 280x's when your kernels hit the light of day.

[djm34]

some progress on neoscrypt:
750ti: 81kh/s 110kh/s
780ti: 165kh/s 182kh/s
980:   290kh/s 299kh/s
(still doesn't scale right   )

[sp_]

Pretty good Around 100% faster than the opencl code?

[djm34]

Pretty good Around 100% faster than the opencl code?

more or less, I haven't checked... (it is about that for the 750ti though... I asked the question...)
It is also consistent with running sgminer code on nvidia (a factor 2 reductions... this is how it should be seen...).

That version is based on Wolf0, took me a little while to implement correctly the large vectors on cuda... but it seems to really help on the scratchpad... implemented a ulonglong16 (32 uint64_t... well actually  8 ulonglong4...)   not sure if the alignement is correct though

[sp_]

I have added another 20-30KHASH to x11,x13,x15 etc. on the 750ti. The standard clocked gainward is hashing at around 2840KHASH. Will cleanup a bit and check in to github. This card has not got the 6pin powerplug and uses around 38-40watt to do x11.

[sp_]

The 22th public beta(windows executable) is available here: (28-12-2014)

http://www.filedropper.com/ccminer22

The sourcecode is available here:

https://github.com/sp-hash/ccminer


Not many changes from release 21 but.

Faster quark(200khash) and x11 (20khash). (750ti)

And more CPU usage

[scryptr]

scryptr you are getting about 40-45% of expected performance from your 280x's. with even mild over clocking (1100/1500) they do 6.5mh/s per card using Wolf's recently released bin files.
To save you the searching here's a link to his post on reddit

http://www.reddit.com/r/DRKCoin/comments/2o1yoz/rewritten_x11_binaries/

Yeah, it's been a while since then, surprised more people don't know.

X11 HASH:  CUDA vs AMD--

To be fair, I re-downloaded and substituted the Wolf Tahiti-x11 binary.  This time I took the additional step of write-protecting the *.bin (binary) file before booting sgminer.  I got the following results upon firing up the same sgminer as in my earlier post:



I am running my GPUs at 1030/1500, stock clocks for my 280x cards.  I will try faster clocks later, but I am having a real problem with stability.  GPU0 crashes within 10 minutes if I don't disable it.  I disabled it at about 9min 30sec, after I clipped the image above.  It had started to misbehave again.

Any card management tips would be appreciated.       --scryptr

[jpouza]

The 22th public beta(windows executable) is available here: (28-12-2014)

http://www.filedropper.com/ccminer22

The sourcecode is available here:

https://github.com/sp-hash/ccminer


Not many changes from release 21 but.

Faster quark(200khash) and x11 (20khash). (750ti)

And more CPU usage

Downloading!

Tks

[HoY]

scryptr you are getting about 40-45% of expected performance from your 280x's. with even mild over clocking (1100/1500) they do 6.5mh/s per card using Wolf's recently released bin files.
To save you the searching here's a link to his post on reddit

http://www.reddit.com/r/DRKCoin/comments/2o1yoz/rewritten_x11_binaries/

Yeah, it's been a while since then, surprised more people don't know.

X11 HASH:  CUDA vs AMD--

To be fair, I re-downloaded and substituted the Wolf Tahiti-x11 binary.  This time I took the additional step of write-protecting the *.bin (binary) file before booting sgminer.  I got the following results upon firing up the same sgminer as in my earlier post:

http://s23.postimg.org/jkgxqorjv/wolf_bin_x11.png

I am running my GPUs at 1030/1500, stock clocks for my 280x cards.  I will try faster clocks later, but I am having a real problem with stability.  GPU0 crashes within 10 minutes if I don't disable it.  I disabled it at about 9min 30sec, after I clipped the image above.  It had started to misbehave again.

Any card management tips would be appreciated.       --scryptr

Change your intensity setting to xintensity 128. You'll see a very significant hash rate jump. Also, are you using sgminer 5?

[jpouza]

Release 22 report in X11:

980 185MHz+ core = > 9,9MH/s

750Ti 200MHz core / 400MHz mems = > 3,2MH/s