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

[Legionnairez]

Scrypt-jane is supposed to be memory hard? with the current settings, I'm using just 1302MB.
which algo benefits the most from GPU ram?
Cheers people! I hope I can return some help to you peeps at some point.

scrypt-jane is memory hard, but microcoin is at a low N-factor meaning that you still have enough memory on your video card to make use of all computational resources.

That is why you do not need all the 4GB of RAM in order to get best performance and that is also why the lookup-gap does not help you at the moment.

Try mining some Yacoin and then you willl notice how memory-starved you get.

Christian

Actually, tried to mine it, but get an error message spam when trying to run, missing some file I think. I'll try to solve it myself first, you don't have to do everything
Sent a wee bit of MAX for you, the next time the cuda group hits the jackpot and I'm in, you'll get a more decent amount for your efforts!

[bathrobehero]

@Legionnairez
Easiest way is to do a .bat and put this in and run:
cudaminer.exe -a scrypt-jane:9 --benchmark -i 0 -H 2 -C 0 -m 0 -l Y --time-limit 20
cudaminer.exe -a scrypt-jane:9 --benchmark -i 0 -H 2 -C 0 -m 1 -l Y --time-limit 20
cudaminer.exe -a scrypt-jane:9 --benchmark -i 0 -H 2 -C 1 -m 1 -l Y --time-limit 20
cudaminer.exe -a scrypt-jane:9 --benchmark -i 0 -H 2 -C 2 -m 1 -l Y --time-limit 20
pause

Than read the values. You'll probably end up with Y8x32.

[cbuchner1]

One more tip for the latest maxcoin build.

For the kernel launch configuration: Take the total number of CUDA cores on your card, multiply by 2, and append x32 for the K kernel and x16 for the F kernel. Don't bother with the T kernel for the moment.

This should give somewhat higher performance than the original 1000x32 or 1024x32 configurations suggested in the README, especially on the biger cards like GTX 660Ti, GTX 780, etc...

Use -L 512 with these huge launch configurations or out of memory is guaranteed. But on the other hand, these out of memory errors don't prevent keccak hashing (the program just continues...). Cleaning up these unnecessary memory allocations would be the next item on my "finish the keccak feature" TODO list.

Christian

[boomerzz]

if you're not doing any maxcoin, maybe stick with the 2014-02-04 release for the time being.

I am not sure why CUDA would spit out an "unknown error". Too much of a GPU overclock maybe?

Christian

Thanks, it's 50Mhz GPU over default, it's been that way since I put it in.  I'll actually flip over to maxcoin on this system for now.

[cbuchner1]

@Legionnairez
Easiest way is to do a .bat and put this in and run:
cudaminer.exe -a scrypt-jane:9 --benchmark -i 0 -H 2 -C 0 -m 0 -l Y --timel-limit 20
cudaminer.exe -a scrypt-jane:9 --benchmark -i 0 -H 2 -C 0 -m 1 -l Y --timel-limit 20
cudaminer.exe -a scrypt-jane:9 --benchmark -i 0 -H 2 -C 1 -m 1 -l Y --timel-limit 20
cudaminer.exe -a scrypt-jane:9 --benchmark -i 0 -H 2 -C 2 -m 1 -l Y --timel-limit 20
pause

Than read the values. You'll probably end up with Y8x32.

this is supposed to give you four auto-tunes in a row, right?

BTW you misspelled --time-limit . Oh, and I did not know --time-limit also worked with the benchmark feature. If so, that is nice.

Christian

[bigjme]

just got my first reject because a job could not be found. think thats a pool issue though
at 97/98 now

way way better then before

[bathrobehero]

Oh, typo.

Yeah, I love this feature, especially if you want to venture into some high lookup gap territory, which could take hours. Setup a bat with everything, go to sleep, reap the rewards in the morning

However, --benchmark has some issues so I would suggest doing it while solomining but didin't wanted to bother explaining that.

[Legionnairez]

I noticed a drop of about 50% hashing rate with todays build, which stayed on when trying other builds too. Restarting computer solved this - probably driver issue or something.

[bigjme]

which coin?

[bathrobehero]

Anybody has any idea why is running every single cudaMiner build takes like 15 seconds before it starts hashing, only for the very first time it is running?

[bigjme]

ive never noticed this

[Legionnairez]

which coin?

Everything I tried, if that was directed towards my comment. Booting solved it anyway. Currently testing the autotune and then the tips Christian gave.
Looks like another monday being super tired at work!  
Edit: Trying the autotune given above, is it normal to receive "Given launch config "Y" doesn't validate"?

[apst14]

One more tip for the latest maxcoin build.

For the kernel launch configuration: Take the total number of CUDA cores on your card, multiply by 2, and append x32 for the K kernel and x16 for the F kernel. Don't bother with the T kernel for the moment.

This should give somewhat higher performance than the original 1000x32 or 1024x32 configurations suggested in the README, especially on the biger cards like GTX 660Ti, GTX 780, etc...

Use -L 512 with these huge launch configurations or out of memory is guaranteed. But on the other hand, these out of memory errors don't prevent keccak hashing (the program just continues...). Cleaning up these unnecessary memory allocations would be the next item on my "finish the keccak feature" TODO list.

Christian

having this error GTX780 card

Code:

[2014-02-09 23:30:32] GPU #0: using launch configuration K4608x32
[2014-02-09 23:30:32] GPU #0: cudaError 2 (out of memory) calling 'cudaMalloc((void **) &tmp, mem_size)' (D:/Christian/Documents/Visual Studio 2010/Projects/CudaMiner/salsa_kernel.
cu line 254)

[2014-02-09 23:30:32] GPU #0: cudaError 2 (out of memory) calling 'cudaMalloc((void **) &tmp, state_size)' (D:/Christian/Documents/Visual Studio 2010/Projects/CudaMiner/salsa_kerne
l.cu line 290)

[2014-02-09 23:30:32] GPU #0: cudaError 2 (out of memory) calling 'cudaMalloc((void **) &tmp, state_size)' (D:/Christian/Documents/Visual Studio 2010/Projects/CudaMiner/salsa_kerne
l.cu line 291)

[ManIkWeet]

One more tip for the latest maxcoin build.

For the kernel launch configuration: Take the total number of CUDA cores on your card, multiply by 2, and append x32 for the K kernel and x16 for the F kernel. Don't bother with the T kernel for the moment.

This should give somewhat higher performance than the original 1000x32 or 1024x32 configurations suggested in the README, especially on the biger cards like GTX 660Ti, GTX 780, etc...

Use -L 512 with these huge launch configurations or out of memory is guaranteed. But on the other hand, these out of memory errors don't prevent keccak hashing (the program just continues...). Cleaning up these unnecessary memory allocations would be the next item on my "finish the keccak feature" TODO list.

Christian

I get a 2mhash increase with that at the cost of desktop usability, I'll stick with K1024x32

[bathrobehero]

Nothing beats Y1024x32 on my 660.

If I would have more time I'd figure out to setup autotune so that it starts testing from like Y700x

[MexiMelt]

One more tip for the latest maxcoin build.

For the kernel launch configuration: Take the total number of CUDA cores on your card, multiply by 2, and append x32 for the K kernel and x16 for the F kernel. Don't bother with the T kernel for the moment.

This should give somewhat higher performance than the original 1000x32 or 1024x32 configurations suggested in the README, especially on the biger cards like GTX 660Ti, GTX 780, etc...

Use -L 512 with these huge launch configurations or out of memory is guaranteed. But on the other hand, these out of memory errors don't prevent keccak hashing (the program just continues...). Cleaning up these unnecessary memory allocations would be the next item on my "finish the keccak feature" TODO list.

Christian

I tried this. but got out of memory errors when doubling. However, not doubling worked for about a 5mh increase. (-l 1344x32 -L 512)

GTX 670 if anyone else wants to use it.

[djm34]

Running on igh, I get 98.58% accepted (over a total of 353) and it seems I get larger %tage of the block (however, on their hand they have also adjusted the number of share per block which was pretty low this afternoon)

[bigjme]

Increase -L to 1024 and doubling works :-)

[peat]

Use -L 512 with these huge launch configurations or out of memory is guaranteed. But on the other hand, these out of memory errors don't prevent keccak hashing (the program just continues...). Cleaning up these unnecessary memory allocations would be the next item on my "finish the keccak feature" TODO list.

Christian

yeh I was getting the out of memory but it didn't stop it from mining successfully. the -L 512 parameter seems to sort this error from appearing even if it was inconsequential ;+)

Cheers for that

[DirtyWilly]

Working good here!  98%, mostly from initial start up.

780ti SC non/ACX - +51 offset - 220,000 KHash.
Opening Chrome and leaving it running gives more hash?  Opening maxcoind wallet and setting affinity to two less cores increases hash?

Settings:

cudaminer.exe -a keccak -q -l K1024x32 -i 1 -C 0 -m 1 -H 2 -L 128 -o stratum+tcp://stratum1.max.hashfaster.com:4004 -O {username}.1:{pass}

https://i.imgur.com/zFoa601.png