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

[nadadenada]

My GPU Cock is 925MHz Fan speed 78% and GPU Temperature is 68ºC  (air temp is about 24ºC) and I am using MSi afterburner for controlling it.

How the hell did you manage to overclock your card to 925 MHz? (Your default clock should be 782 MHz, right?)
Are you sure that this is stable and won't damage your card in the long run?

About temperatures: I have the HAF 932 Tower Case from Coolermaster, which is fitted with 3 big fans and 2 small fans.
Cool Air is getting sucked in and hot air will get out - so there is a good airflow already.
One of the big fans is right next to the GPU and blasting cool air in. No need for a extern fan.


Did anyone of you guys try to change some driver settings?
I still want to know if there is an option to improve the performance for the mining task.

My card with stock settings is at 782MHz and I am using cudaminer released on 10th December, using it since it was released and I've been mining with this card for two months. I don't know how it is going to affect with long use, but just before start mining I opened my card and changed thermal paste (Arctic Silver 5).
IMO you should open your PC case, and control temps with lower fan speed just for testing.

[blade87]

Does anyone know how I can limit or slowdown the miner? It makes my 570 GTX sound like it will fly to the moon rather quickly, would love to be able to use my computer normally while it mines at like a 20% dip in performance.

Download MSI Afterburner (latest beta version works best)

Fermi users : Undervolt and underclock the card manually.
Kepler users : Decrease the power target slider from 100% to 90%, 80%, etc and the card will lower its voltage and clocks automatically.

[cainy393]

Just tried this on my new GTX 760. I get around 250 KH/s although when I came back after a few hours it had dropped to 200 KH/s and was feeling rather hot to the touch, although i didn't think to actually check its temp. Anyone else got a GTX 760 to compare to?

[Nullu]

Very nice, seeing about a 50-60khash/s improvement on this new version on a GTX 670. Nice one OP!

Same here. Sometimes my hashrate drops, and at max I can push it close to 300 khash/sec, but 250 khash/sec gives me a good balance of temperature and speed.

[Valnurat]

How do you use cudaminer through a proxy?

http://www.lpshowboat0099.com/Blog/how-to-mining-ltc-with-cudaminer-on-a-stratum-server/


this should help ya out


but with newest releases of cudaminer , you do not need a proxy to run.

Yes, I have tried that but it will not Work.



Anyone have an idea why it will not Work?

[ak84]

How do you use cudaminer through a proxy?

http://www.lpshowboat0099.com/Blog/how-to-mining-ltc-with-cudaminer-on-a-stratum-server/


this should help ya out


but with newest releases of cudaminer , you do not need a proxy to run.

Yes, I have tried that but it will not Work.



Anyone have an idea why it will not Work?

'

probably because the TCP server address is wrong or the server is down. Try connecting to another server or checking your syntax

[cbuchner1]

Did some testing on the X-perimental kernel (test_kernel.cu) on my GTX 780 Ti, slightly overclocked.

compiled for Compute 3.0: 510 kHash (no funnel shifter and no texture cache used)
compiled for Compute 3.5: 620 kHash (standard Titan kernel with funnel shifter and __ldg() based scratchpad access)

doubled the ALU workload (salsa_core called two times in each iteration): 340 kHash
removed the ALU workload (salsa_core replaced by an empty function): 694 kHash

Hence we're nearly bandwidth limited and we fully saturate the ALUs, it seems. So it doesn't appear that we can implement a LOOKUP_GAP with any speed-up. There's no way we can trade memory bandwidth for more computation to gain speed.

Also I ran some tests on Linux where overclocking of Kepler devices isn't an option.

With the standard BIOS I get 515 kHash
With the skyn3t-Ti vBios I get 585 kHash, but intolerable heat emission and power draw (apparently the default voltage is set very high)
With svl7's GTX 780Ti "OC edition" rev 04 vBIOS I get 560 kHash with only a slight increase in power draw.

So I guess for building a Linux based mining rig using the 780 Ti, it's currently the best option to use svl7's BIOS. Still, a GTX 780 overclocked on Windows will provide more kHash than a 780 Ti using the modded vBIOS.

[dga]

Did some testing on the X-perimental kernel (test_kernel.cu) on my GTX 780 Ti, slightly overclocked.

compiled for Compute 3.0: 510 kHash (no funnel shifter and no texture cache used)
compiled for Compute 3.5: 620 kHash (standard Titan kernel with funnel shifter and __ldg() based scratchpad access)

doubled the ALU workload (salsa_core called two times in each iteration): 340 kHash
removed the ALU workload (salsa_core replaced by an empty function): 694 kHash

Hence we're nearly bandwidth limited and we fully saturate the ALUs, it seems. So it doesn't appear that we can implement a LOOKUP_GAP with any speed-up. There's no way we can trade memory bandwidth for more computation to gain speed.

Also I ran some tests on Linux where overclocking of Kepler devices isn't an option.

With the standard BIOS I get 515 kHash
With the skyn3t-Ti vBios I get 585 kHash, but intolerable heat emission and power draw (apparently the default voltage is set very high)
With svl7's GTX 780Ti "OC edition" rev 04 vBIOS I get 560 kHash with only a slight increase in power draw.

So I guess for building a Linux based mining rig using the 780 Ti, it's currently the best option to use svl7's BIOS. Still, a GTX 780 overclocked on Windows will provide more kHash than a 780 Ti using the modded vBIOS.

Concur.  I tried something slightly different today - instead of just doubling salsa_core, I tweaked it so that every thread handled two columns of the 4x4 key matrix.  My thought behind this was that it would let the inner loop process a single key with only 8 total __shfl()s per iteration instead of 4*3=12.  I targeted this because __shfl() only has an instruction throughput of 32 per cycle (vs 32 for shift and 160 for add).  Made things a lot slower.  I think we're basically done other than micro-optimization.

[Valnurat]

How do you use cudaminer through a proxy?

http://www.lpshowboat0099.com/Blog/how-to-mining-ltc-with-cudaminer-on-a-stratum-server/


this should help ya out

but with newest releases of cudaminer , you do not need a proxy to run.

a couple of pages back I tried to explain the -x proxy option. It supports http: https: and socks: proxies, as far as I determined. I did not write that code though, as it is inherited from the cpuminer project.

Christian

I'm sorry have mist that.

[ak84]

Christian, thanks for all your excellent work on cudaMiner. I have a question that hopefully you can shed some light on.

With all the optimization you've done for cudaMiner mining scrypt, what % would you say is optimized already? IE. How much more % increase in hash rate can we expect to come from new versions/improvements?

It seems that the latest version of cudaminer puts 660Ti hashrate for scrypt at around 80% of that of price-comparable AMD card, hd 7850.

eg. 660Ti ($180, 280 khash/s)  vs.   7850 ($180, 350 khash/s)

280 / 350 = 80%

We might be nearing a point where Nvidia cards can act as decent substitutes for AMD cards.

edit:: list i've compiled so far
Card name, price, hash rate/s stock (Overclock)

GTX 560Ti, $150, 280
GTX 580, $, 282-320
GTX 650, $120,
GTX 660, $180, 200 (224)
GTX 660Ti $, 238-257 [EVGA]
GTX 660Ti, $180, 270 (300) [ASUS]
GTX 670, $270, 310
GTX 680, $360, 310, (325-slight, 355)
GTX 690, $ , 400
GTX 760, $260, 250-280
GTX 780, $500, 435-480 (550 oc)
GTX 780Ti, $700, 585-700

[see360]

Thank you for cudaminer! I got a 41% increase in speed from 2013-12-10 to 2013-12-18.

[UNOE]

What a great job you have done with this miner.  Its working so much better each month I check this thread.  Good Job bro

[nuclearstar]

Very happy with cudaminer at the moment.
with the latest release and playing about with settings I have got my EVGA A/C Superclock 780 to run at 550Khash/s

This is my config cudaminer.exe -i 0 -H 1 -m 1 -l T12x16

my overclock settings  too http://imgur.com/WoxJGJU

Gotta give a shoutout for my case too silverstone raven rv02
It is an amazing case and really impressed with the airflow and cooling.

Would the cuda 6.0 tools have any benefit to us when thats release? are you planning to update cudaminer with 6.0 tools?

[Nullu]

I have a GTX 670 that can run at 250khash/sec - 300khash/sec, but it's running my card too hot.

Is there a configuration so I can run it at 200khash/sec?

At this rate my card just throttles after a few hours and hashrate plummets.

[Banin]

Just loving you man, hash speed increased a lot

thanks again, i'll send you 1 LTC once i have !

[cbuchner1]

Just loving you man, hash speed increased a lot

thanks again, i'll send you 1 LTC once i have !

a quarter LTC will do. First post was written when 1 LTC was like 2 dollars and easy to mine

[cbuchner1]

I have a GTX 670 that can run at 250khash/sec - 300khash/sec, but it's running my card too hot.

Is there a configuration so I can run it at 200khash/sec?

At this rate my card just throttles after a few hours and hashrate plummets.

I plan to add a percentage scaling for mining intensity as well as temperature throttling in software...

[Banin]

LTC: 0.02

Currently it's not possible LOL, but dont worry, u'll get ur money for your work bro'

Currently 245 KH/S with GTX 760

[cbuchner1]

____
1050 khash/s

1050/740 = 41% increase in hashrate from 12-10 to 12-18!!!!

Windows 7 Home Premium
intel i7-3820 @ 3.6GHz liquid cooled but no overclock
MoBo Gigabyte GA-X79-UP4
nvidia driver 331.82 (2013-11-19)
1000W PS

great to see other people hitting a megahash with a single nVidia equipped PC...

[M.Winter]

I have done some undervolting and overclocking to see what happens with temp and hashrate.

core i7 920 @ 3.0 GHz, 6 GB RAM @ 1600 MHz
GTX 580, 3 GB 384 bit, Core: 844, Mem: 2055, Shader: 1688
-i 0 -H 1 -C 1 -l F16x16
307.11 khash/s | 74 °C

Would it be helpfull if I try to compile CudaMiner myself for Fermi Kernel?