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

[sambiohazard]

I ran autotune again in debug.  And got this
cudaminer -d gtx780ti -H 1 -l T59x4 -C 2 -i 0

This gives me what I got with the december release around 630 to 650.

Would you try 60x4 as 780ti has 15 SMX thus 15*4 = 60. Maybe it will help.

EDIT: Also should it not be t instead of T = Y kernel from nvidia.

[1714028172]

1714028172

[bathrobehero]

Could you give me some advices how to finetune Kepler and Titan cards for mining low-factor scrypt-jane coins? For example with Vertcoin

VertCoin is an exception, it's not scrypt-jane, it's scrypt with an adaptive N factor.

[Galvin]

I ran autotune again in debug.  And got this
cudaminer -d gtx780ti -H 1 -l T59x4 -C 2 -i 0

This gives me what I got with the december release around 630 to 650.

Would you try 60x4 as 780ti has 15 SMX thus 15*4 = 60. Maybe it will help.

EDIT: Also should it not be t instead of T = Y kernel from nvidia.

I tried t60x4 get around 370.  and T60x4 gives me 570.
So T59x4 is the best so far.

[morbooo]

Multi-GPU on Kepler is working great now, thanks!

[micryon]

any support for SHA-3 upcoming MAXCoin?

[cbuchner1]

any support for SHA-3 upcoming MAXCoin?

depends. I tend to add new hashing algorithms only if there are enough coins to mine (there's more than a dozen Yacoin clones, for example).

If someone posts a reasonable bounty, I might add it faster. Or someone else takes cudaminer and adds it provided the bounty is reasonably high (it's open source, and forking it on github is easy as pie).

Anyway.... 9 more hours before the wallet and miners are released, so I cannot currently see how it is working - in particular how the PBKDF2 is applied and whether they use Keccak256 or Keccak512.

Christian

[Morgahl]

I tried t60x4 get around 370.  and T60x4 gives me 570.
So T59x4 is the best so far.

There seems to be a lot of confusion over the different kernels since the release so I just want to repost directly from the README:

Code:

Prefix	Alias	Compute Req.	Registers	use for
F	L	1.0		64		scrypt & low N-factor scrypt-jane
K	Y	3.0		63		scrypt & low N-factor scrypt-jane
T	Z	3.5		80		scrypt & low N-factor scrypt-jane
				
f	X	1.0		32		high N-factor scrypt-jane
k		3.0		32		high N-factor scrypt-jane
t		3.5		32		high N-factor scrypt-jane
				
the old "Legacy" kernel has been replaced with the F kernel, which will also be faster on Compute 1.0 legacy devices in many cases.
Therefore the F kernel has been compiled to require only Compute 1.0 capability.	

      

T, K, F, t, k, and f are different kernels with different uses. Please keep this in mind when you are making suggestions to others on what kernel to use.

For those who were using interim releases (special thanks to those who provided their self compiled binaries!)

• If you were using T, K, or X kernels for scrypt-jane , you will want to use t, k, or f respectively.
• If you were using Z, Y for scrypt in the interim releases, you will want to use T, K not t, k.

SMX counts(source http://en.wikipedia.org/wiki/CUDA):

Code:

Compute ver.	1.0 	1.1 	1.2 	1.3 	2.0 	2.1 	3.0 	3.5
Cores per SMX	8	8	8	8	32 	48 	192 	192

To find the compute level of your card please review this site:
http://www.techpowerup.com/gpudb/

For scrypt using the K Kernel generally the best basic configuration is going to be K(SMX)x32, you can find the SMX number by finding the number of Cuda cores on your card and dividing by 192. For example a GTX 680 has 1536 Cuda cores this means the SMX for your card is 8 giving you a basic config of:

Code:

-l K8x32

For scrypt using the T Kernel generally the best basic configuration is going to be T(SMX)x24, you can find the SMX number by finding the number of Cuda cores on your card and dividing by 192. For example a GTX Titan has 2688 Cuda cores this means the SMX for your card is 14 giving you a basic config of:

Code:

-l T14x24

Special note here, some oddities have been noted for some makes and models of cards in this kernel and the T(SMX)x24 is not always the best, Once you have establish a baseline hash rate at T(SMX)x24 try to reduce the "warps" section (ex. T(SMX)x20 or T(SMX)x16) and see if there is improvement.

For scrypt using the F Kernel generally the best basic configuration is going to be F(SMX/2)x32, you can find the SMX number by finding the number of Cuda cores on your card and dividing by the appropriate core count for your Compute capability. For example a GTX 580 has 512 Cuda cores and is compute 2.0 this means the SMX for your card is 16 giving you a basic config of:

Code:

-l F8x32

Note: If anyone has a different observation for the Fermi cards please let me know! I have only compiled this part from observation I have no direct experience with this GPU Class and Cudaminer.


This only helps with basic Kernel selection and configuration, it is up to the individual to identify what additional -H, -C, -m, and -L parameters work best on your configuration.

[bigjme]

ok so my microcoin on my 780 is sitting at just over 1million coins every 24 hours.
i hope these coins go up, im still using my build from 1/2/14 as i havent been getting emails for replies to this post AGAIN and i cant remmeber where i last read from, something to do with upper and lower case kernels doing different things now.

how are you finding things christian? and you bathrobe?
switched over to something else or are you riding out the microcoin silence?

[SystmHash]

Vertcoin on GTX780 with core overclocked to 1254Mhz = ~335kh/s
cudaminer.exe --algo=scrypt:2048 -d 0 -H 2 -C 1 -l T12x20 -i 0

GTX 670 overclocked to 1200mhz (using CPU for sha256) = ~155kh/s
cudaminer.exe --algo=scrypt:2048 -d 1 -H 1 -l K35x6 -i 0

Use these!

Donations welcome!
VTC = VouDyTwF5QNvmGukvdHgsZNwtJKjcY6aRD

I have a GTX 780 and with those arguments I get around 25kh/s. Adding -L 2 gets me to 200kh/s.
What model is your 780?

It's GIGABYTE GV-N780GHZ-3GD GeForce GTX 780 3GB 384-bit GDDR5 PCI Express 3.0 HDCP Ready Video Card

[djm34]

Can someone re-explain what does 24 represents. I didn't get it the first time and it is quite difficult to find it in the thread..

[djm34]

ok so my microcoin on my 780 is sitting at just over 1million coins every 24 hours.
i hope these coins go up, im still using my build from 1/2/14 as i havent been getting emails for replies to this post AGAIN and i cant remmeber where i last read from, something to do with upper and lower case kernels doing different things now.

how are you finding things christian? and you bathrobe?
switched over to something else or are you riding out the microcoin silence?

Funny, but it seems easier to get blocks in solo mining (less miner ?).
Yesterday I I was on a pool for a two hours MRC mining session, then got bored after 1 hours and switched to solo got 2 block in 10 minutes intervals...

[cbuchner1]

ok so my microcoin on my 780 is sitting at just over 1million coins every 24 hours. how are you finding things christian? and you bathrobe? switched over to something else or are you riding out the microcoin silence?

Currently microcoining with just above 2MHash/s. Having 20+ mio. in the wallet just feels right. Always mining in a pool though.

Christian

[bigjme]

Ahh, following my luck with solo mining, 550khash/s gets just over 1 million a day. 2 mhash/s following that should be getting about 4million a day. Obviously its never that simple but the principle is there

[Stanr010]

I posted a new release 2014-02-04 fixing two important bugs.

- autotune underreporting kHash/s values if the kernel finished in under 50ms (forgot to divide time elapsed by number of measurements, doh!)
- Multi-GPU support was not working - it is now.

OMG at last!!!! Thank you, thank you, thank you!!!

For the longest time I couldn't understand why I got cpu validation errors. I finally understood after reading the readme that I had to run it individually for each GPU. Glad that's finally fixed.

[eduncan911]

@cbuchner1:

Awesome fracking work with scrypt-jane man!

May I make a developer suggestion?  Could you "branch" in Github, away from the master, the next time you want to experiment and upgrade code?  It was quite difficult to find stable scrypt versions during this last endeavor.  Ping me in PM and I can help ya out on github branch management, and how to submit pull requests to merge it back into the master.  It's actually quite simple.

In short, leave the master alone: calling it your "stable" releases.  Branch for any possible changes, upgrades, or even hotfixes with "git checkout -b".  Nothing to download, and you are now on a new branch.  Once it's fixed, patched, etc and you are read, submit a Pull request to the master, and then go approve the Pull request - BAM now you have the latest in master.

It would keep those of us that like to tinker with your code, to only tinker with the "stable" master branch - while you merge in hotfixes and big new algorithms.  

Thanks again man!


Now, to get me some Jane!

[supaflyx3]

Could someone with a 780ti chime in with their settings for vert? I've tried everything I've found in this thread so far, and I've only been able to get 175KH/s max out of my 780ti. My config is:
--algo=scrypt:2048 -L 1 -d 0 -H 2 -l t15x16 -i 0

[bigjme]

anyone else's micro coin wallet saying the checkpoint is too old? and to wait for the block chain to update or contact the developer?
it popped up a few minutes ago for me

[ak84]

Can someone upload a copy of cudaminer.zip (february 04 2014)?

Mega download is not working for me.

[DeltaQuebec]

I would like to say a thank you to the dev for 2.3 & 2.4. I was on the December beta running my GTX 650TI and thought I had all I could get out of her with that. HOWEVER, 2.2 & 2.4 x64 versions have givin me another 11-12%$ hash!!! I have documented this the best I could on https://www.weminecryptos.com/forum/topic/516-nvidia-geforce-gtx-650-ti/ with photos. For those of you whom are running a ASUS GTX 650TI (OC)  and you get sub 130 kh/s performance, you may want to check out the link above. Your mileage may vary.

Thanks again to the dev & I am VERY VERY excited to hear fail-overs are in the plans! YAY!!!

[supaflyx3]

Use cudamanager for fail-overs, its great.