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

[Lacan82]

This might just be my crappy hotel internet, but I am using stratum mining. Every so often I'm losing connection on the miner.

[1713592878]

1713592878

[1713592878]

1713592878

[1713592878]

1713592878

[cbuchner1]

This might just be my crappy hotel internet, but I am using stratum mining. Every so often I'm losing connection on the miner.

Let me guess, you're in a hotel for the "free" electricity, and you hide a truckload of miners in the closet.

Christian

[mg27341]

It is probably safe to over clock to 822. My card is standard at 822 mhz factory OC

I've seen factory overclocks from 770 MHz to 786 MHz, but 822 MHz does sound high. But, hey if that's safe, maybe I'll give it a go for a while at 1.038 v and see what that nets me.

Michael

[mg27341]

This might just be my crappy hotel internet, but I am using stratum mining. Every so often I'm losing connection on the miner.

Let me guess, you're in a hotel for the "free" electricity, and you hide a truckload of miners in the closet.

Christian

LOL, it probably went something like this:

Hotel attendant: Sir, you sure have many suitcases.
Lacan: Oh those... They're just my clothes for the trip.

Ten minutes later, as the hotel lights start dimming...

Lacan: Two outlets? That's all these hotel rooms have is two outlets? OK, where are my octopus splitters? Ahh, there we go... Wait, WTF? The internet is petering out and I don't know why? Could it have anything to do with the lights dimming?

Ten additional minutes pass and Lacan calls the hotel electrician.

Electrician: Sir, why are you splicing the electrical cables connected to the fan?

...

[Lacan82]

This might just be my crappy hotel internet, but I am using stratum mining. Every so often I'm losing connection on the miner.

Let me guess, you're in a hotel for the "free" electricity, and you hide a truckload of miners in the closet.

Christian

Nah, I switched jobs, and my house isn't ready for closing yet so I have to stay in a hotel for a couple weeks.

[gchil0]

Here's a 32-bit linux binary of the 4/17 release.
https://www.dropbox.com/s/twjhw65hfwfcbca/cudaminer_417.gz

If you have library problems with the one above, here's the binary linked slightly differently:
https://www.dropbox.com/s/nymd9z6eklfzb7p/cudaminer_417_2.gz

On my old C1060's, using the texture cache has increased the speed from 85 to 90 khash/s.

[Limie]

I tried this latest release, upon every start of it, it takes a long time to autotune, after auto tune is finished, if you restart the app, it auto tunes again. Shouldn't it auto tune only once?

Second, all it is reporting back after autotune is, new block found, no hash rate reported.

Reverted back to 04-04-2013 version for now.

If you are satisfied with what autotune does for you, then you can set that rate instead of autotune somehow. op help with this?

[Lacan82]

I tried this latest release, upon every start of it, it takes a long time to autotune, after auto tune is finished, if you restart the app, it auto tunes again. Shouldn't it auto tune only once?

Second, all it is reporting back after autotune is, new block found, no hash rate reported.

Reverted back to 04-04-2013 version for now.

If you are satisfied with what autotune does for you, then you can set that rate instead of autotune somehow. op help with this?

yes, you can use -l  type in the config it used.  so like -l 60x4 for example

[SubNoize]

@cbuchner1

For some reason -Interactive is now enabled by default?? I didn't realise and it was causing massive drama with my gpu usage and temps. Is this a bug or?


make sure you have "-i 0" in launch options if you don't want it enabled.

Also i found on my 580's and 570's that -C 1 gives a massive drop. -C 2 gives a stable around 230ish mark for the 580. but -C 0 gives 220 to 245KH/s

I'll test -C 2 and -C 0 some more when i have some time.

[shivansps]

20.20 khash/s on the little GT520 with -C2 this keep getting better and better.

[haderach]

With regards to my Titan, the 4.17 build appears to either fixed or introduced bugs in the validation of hashes using the CPU. The perspective just depends on if I want to believe I've possibly damaged my Titan.

The 4.17 build reports a bad GPU hash about once every 30-45 seconds while the 4.14 build never reported a single one. This is using the exact same startup parameters (-l 294x2 -C 0 -i 0). The rejected rate from 4.14 also matches with p2pool. I also briefly tried -C 1, -C 2 incase they might have any performance impact but I didn't see anything noticeable.

[cbuchner1]

With regards to my Titan, the 4.17 build appears to either fixed or introduced bugs in the validation of hashes using the CPU. The perspective just depends on if I want to believe I've possibly damaged my Titan.

The 4.17 build reports a bad GPU hash about once every 30-45 seconds while the 4.14 build never reported a single one. This is using the exact same startup parameters (-l 294x2 -C 0 -i 0).

What's the hash rate on Titan now?

I am programming blindly with no testing opportunity. It is way more likely I introduced a programming bug!

Try reducing your launch parameters to half the block count, maybe. 147x2. It could be that the const __restrict__ pointers have some size limitations similar to the 1D texture size restrictions on the affordable consumer devices (Titan is not affordable IMO). Another thing to try is to set the -m 1 flag (single memory allocation mode). The -C flag does not currently affect the Titan.

[Mikanoshi]

GTX 670 @1201/7008:

-i 0 -l 210x2
183 kH/s, 1880 MB

-i 0 -m 1
174 kH/s, 1264 MB, Autotuned 70x4

-i 0 -C 1 and -i 0 -C 2
184 kH/s, 1264 MB, Autotuned 56x5

-i 0 -m 1 -C 2
184 kH/s, 1264 MB, Autotuned 70x4

What values to use for 2 cards? Tried those with the lowest VRAM usage:
-d 0,1 -m 1,1 -C 2,2 -i 1,1 -l 56x5,56x5
but computer becomes very slow and eventually freezes or videodriver crashes.
It seems cudaMiner starts to use shared system memory, because VRAM is not enough.

[FalconFour]

I uncovered a bit of a performance problem that's probably worth looking into.

Playing with many test-bed PCs, I've found that Windows XP runs circles around Windows 7 CUDA performance. There's a pretty huge issue.

XP with a 9600 GT: 20-24khps
Win7 64 on exactly the same system with the same launch configuration: 11khps - Aero or not. Autotune makes it worse - 8khps.

Check this thread out: https://devtalk.nvidia.com/default/topic/484590/windows-7-vs-windows-xp-cudamalloc-performance-difference/

Maybe try allocating a single block of memory all at once, then prevent any further allocate/release functions in the hashing cycle...?

[cbuchner1]

Maybe try allocating a single block of memory all at once, then prevent any further allocate/release functions in the hashing cycle...?

there's no further alloc/release of GPU memory during hashing cycle. The only mallocs are done on the CPU for system memory in the scrypt.cpp code - and they happen infrequently.

The problem we're seeing could be one of dynamic GPU clock speeds, maybe?

[FalconFour]

Nope. The 8000- and 9000-series (8000 I know for certain from much debugging; 9000 I haven't owned long enough - sad, huh?) don't have dynamic clocks. They just shut down idle parts of the core, but run 100% clock all the time, hence, heat-monsters and lots of early failures

Strange though. Both installs were using exactly the same driver version - 314.x (forgot the decimal ATM, not in front of it). Exactly the same hardware configuration as well - I'd just installed that GPU and fresh-installed the nVidia drivers after being an unknown device in XP. Only difference was the operating system. :/

I'm restoring XP back to a second partition on the now-Win7 x64 system so I can dual-boot it and compare/debug... (it's a work-in-progress PC for a friend, so I took a WIM image of it before wiping - so I can restore it back exactly)

Could it be that the malloc, done once, was placed in a segment of GPU memory that didn't land in a high-performance configuration - and since it was only done once, it varies on each launch?

[nakedman]

I'm getting an error on Debian Wheezy 32 bit. I have a 560TI. Driver is 304.64. I download the file you have linked here somewhere, the cudaminer_414.gz. I can run it but with the configuration of cudaminer -o URL -O credentials it does nothing but start and wait while with

Code:

mads@Adria:~/cudaminer$ ./cudaminer_414 -m 1,1 -l auto,auto -o http://netcodepool.org:8337 -O nakedman.1:xxx 
	   *** CudaMiner for nVidia GPUs by Christian Buchner ***
	             This is version 2013-04-14 (alpha)
	based on pooler-cpuminer 2.2.3 (c) 2010 Jeff Garzik, 2012 pooler
	       Cuda additions Copyright 2013 Christian Buchner
	   My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-04-18 16:59:31] 2 miner threads started, using 'scrypt' algorithm.
[2013-04-18 16:59:31] Long-polling activated for http://netcodepool.org:8337/LP
[2013-04-18 16:59:31] GPU #0:  with compute capability -1244551188.-1244550632
[2013-04-18 16:59:31] GPU #0: interactive: 0, tex-cache: 0, single-alloc: 1
[2013-04-18 16:59:31] GPU #1:  with compute capability -1252943892.-1252943336
[2013-04-18 16:59:31] GPU #1: interactive: 0, tex-cache: 0, single-alloc: 1
[2013-04-18 16:59:31] GPU #0: Performing auto-tuning (Patience...)
[2013-04-18 16:59:31] GPU #0:    0.00 khash/s with configuration  0x0
[2013-04-18 16:59:31] GPU #0: using launch configuration  0x0
[2013-04-18 16:59:31] GPU #1: Performing auto-tuning (Patience...)
Floating point exception

I am not sure where the floating point exception comes from. I have only downloaded and installed 32bit stuff. Unless the cudaminer you linked under the guise of being 32 bit is actually 64bit. You know why I'm getting this error?

I just noticed there was a new release here, the cudaminer 417. I am getting an error trying to run this:

Code:

mads@Adria:~/cudaminer$ ./cudaminer_417 -m 1,1 -l auto,auto -o http://netcodepool.org:8337 -O nakedman.1:xxx 
./cudaminer_417: error while loading shared libraries: libcrypto.so.6: cannot open shared object file: No such file or directory

I am not sure where I can get the libcrypto.so.6 on Debian. I have tried install openssl and libssl, but it doesn't seem to do the trick.

[TiuraZ]

Thank you very much for this miner. Will definately donate something for further development. My GTX 460 performs about 50% better with this than with Reaper AND my computer is still usable.



Do you know if this 90-95 khash/s is a good performance for this card or could I squeese little more out of it?

[cbuchner1]

My GTX 460 does slightly more than 100kHash when overclocked - I think I've reached 110 kH.

[cbuchner1]

I'm getting an error on Debian Wheezy 32 bit. I have a 560TI. Driver is 304.64. I download the file you have linked here somewhere, the cudaminer_414.gz. I can run it but with the configuration of cudaminer -o URL -O credentials it does nothing but start and wait while with

Update nVidia driver to release 304.56 or later please.