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

[cbuchner1]

Pulling 150-170 kHash/s on a 560 Ti, this is pretty awesome, I was pulling around 210-220 on a HD 5770.

This number looks unreasonably high (not incomparison to cgminer though).

It's unreasonable compared to my 660Ti which pulls 130 kHash only. I want my money back!

How much did the 560Ti cost? This could easily be the nVidia GPU with the best price/performance ratio for mining.

Is this returning valid shares to the pool? What's the launch configuration you're running?
For example S1x4 launch configuration is currently borked (i.e. non functioning)

Christian

[1714246853]

1714246853

[1714246853]

1714246853

[1714246853]

1714246853

[Wole]

I want my money back too, my 560Ti with the latest version is "only" (compared to cgminer its a lot) getting 100-107 kh/s!
With the latest version there are some annoying CPU spikes which freezes everything though, maybe thats just because I'm forced to use the no autotune?

[GForce]

Is this normal?

http://puu.sh/2wVmH

If so, thats my 560 Ti

[cbuchner1]

It can't be normal that it finds so many shares so quickly.

This pool seems to accept anything you throw at it with a "yay".

Must be a fault. Use --no-autotune, for a change.

Christian

[GForce]

It can't be normal that it finds so many shares so quickly.

This pool seems to accept anything you throw at it with a "yay".

Must be a fault. Use --no-autotune, for a change.

Christian

http://puu.sh/2wW2A

This is P2Pool btw.

[chanzilla]

Thanks for this. The tweaks are great. I'm getting around 77kh/s from my 460 though I had to manually set 28x4 since autotune was choosing something else that wasn't working at all.

[Wole]

Looks like P2Pool returns yay for whatever. But do you actually get paid for them? Do you show up on "Payouts if a block were found NOW:"?

[cbuchner1]

!!! The April 09 update is ready !!! Fixes the autotune bug I introduced yesterday, adds back support for the oldest nVidia cards (compute 1.0 cards like the 8800GTX), enables compilation on Linux through autotools.

Donations welcome

Still need somebody.... anybody.... to test this with the Geforce Titan.

Christian

[Neverest]

Is this just for Nvida cards to catch up, or will it improve Radeon's performance as well? Did anybody test?

[Sparks]

!!! The April 09 update is ready !!! Fixes the autotune bug I introduced yesterday, adds back support for the oldest nVidia cards (compute 1.0 cards like the 8800GTX), enables compilation on Linux through autotools.

Donations welcome

Still need somebody.... anybody.... to test this with the Geforce Titan.

Christian

With Geforce 560Ti now it's OK with autotune, made 108 khash/s to me.
Great job, thanks!

[cbuchner1]

Can anyone port it to Linux, please?

Done! Back to Linux, where it came from.

I tested this on Ubuntu 9.04 today as well as Ubuntu 12.04

needs libssl-dev (openssl-dev?) and perhaps some other dependencies installed (configure will complain if it cannot find the required dependencies). I recommend you use the ./configure.sh helper shell script and have CUDA 5.0 toolkit installed in the default location. This calls configure with the right arguments (for Linux).

Christian

[cbuchner1]

Is this just for Nvida cards to catch up, or will it improve Radeon's performance as well? Did anybody test?

For Radeons, use OpenCL miners. This CUDA stuff is nVidia only.

[cbuchner1]

Now this is awkward. I was expecting that log messages with LOG_DEBUG level don't get shown unless the --debug option was turned on. I guess I was wrong.  That means that autotune output will always be shown, even without the -D switch.

Well if there wasn't something to fix, there wouldn't be a next release to look forward to.

[Number6]

Note:
Autotuned for ten minutes, it kept auto tuning and not mining, 209 rounds of autotuning. Went back to using the no-autotune switch. GT520

Yeah, I tried this too for a bit on a GTX660TI box I have. Gave up after 15 minutes of still auto tuning.

Also, would it be possible to get some type of graceful exit/close option? I had to Control-C to break out and my graphic card driver crashed and then auto recovered before I could see the hit "Y" to cancel message.

[cbuchner1]

I just fixed the issue that it printed autotune debug messages even when the -D switch was not given. For very slow GPUs I added some feature that it does not always average 3 measurements to determine the khash value. If the measurements have taken more than 50ms, it will bail out early. On slow GPUs that will make autotuning faster by a factor of 3.

On a different note - for people having trouble joining a pool:

Check out the mining pool litepool.eu - they require no sign up. You just pass an inbound wallet address as your username, and the password is litepool.eu.  URL is http://www.litepool.eu:80 Fees are 1% I think. Just search these forums for litepool.eu and you will find their announcement thread.

The the arguments for cudaminer look like this (use your own wallet address please, unless you want to mine for me ;-))

-o http://www.litepool.eu:80 -O LWKZoNHb11P57LAmRx6MSkzftCH4Sp9oZe:litepool.eu

Looks like this pool is sporadically up or down, just like any other pool. Whoever is doing DDOS attacks on my favourite mining pools - GO TO HELL. I will be solo mining if this problem gets worse.

UH, check out launch configuration 56x2 for GTX 460 devices. Occasionally autotune gets funky and tries something new. Suddenly this does 74 kHash instead of 66. This is a CUDA 2.1 device (GF104 chip) and has 7 multiprocessors. So multiply that by 8 and use x2

Christian

[TiWu]

Hi,

Just tried out this miner on my Club 3D GTX 570 with great results!
Autotune detects the card correctly, runs the test and selects launch config 60x4. Then it starts mining at around 150K immediately!
I was only able to get it around 100K with reaper, untill now.  Very happy!

Not sure if it might be useful for the dev, but I posted the autotune output: http://pastebin.com/ZAem3MB9

Running Win8 x64, with driver 314.22.
Stock core speed (732), but increased memory clock (now 2100, stock:1900) with MSI afterburner. It seems to run a few degrees cooler than with reaper too, but that might be my imagination.

I also 'suffer' from the driver reset when quitting the miner with ctrl+c, but that is not really an issue for me. Reaper suffers from it too, and often even causes a bsod (although these are looking a lot nicer in win8  ).

[Number6]

I ran the new code and the autotuner seems to be much improved! No output spam and about 3 minutes to tune, which is good.

My GTX 660 TI settled on 148x2 for startup and it is getting anywhere from 130-145 Khash/sec. I am running at 1306 MHz core clock on my GPU for reference.

Would be nice to see some other speeds posted for comparison, I am thinking I should be able to squeeze a bit more out of this thing?

[chanzilla]

UH, check out launch configuration 56x2 for GTX 460 devices. Occasionally autotune gets funky and tries something new. Suddenly this does 74 kHash instead of 66. This is a CUDA 2.1 device (GF104 chip) and has 7 multiprocessors. So multiply that by 8 and use x2

I tried 56x2 for my 460 but it didn't seem to be any different to 28x4 which I guess makes sense. Any particular reason to use 56x2 over 28x4?

[TiWu]

Just moved my desktop rendering to the IGP, so I have the GPU dedicated for mining. Not that it will make much difference in speed, but it does in desktop user experience. 
I did a few restarts of the miner, and it seems to take another config each time, but they seem to result is about the same speeds of around 145-150K.

I've reset the clocks to stock now, only cranked up the fan a bit. Once it took 119x2, 120x2 (not that much difference I guess?). With increased memory speed to 2100 it chose 60x4?
What is the autotune using to determine the outcome? Should I try to find an suitable config and then fix than in the startup params? Or is it better if I let it autotune on each startup?

[wndrbr3d]

Is it the SP's available or card memory available that is a limiting factor for the CUDA miner? I ask because I have a 670 with 4GB (vs. 2GB standard) and was curious if I should pop that in and give it a try.