Topic: NVIDIA Kepler (K20) from 134MHash/s to 330MHash/s with CUDA - page 6. (Read 73331 times)

@InqBit: is 137694 khash/s the first shown hash rate? Please run the miner 5 min and than look to the hash rate. The first hash rate is always lower because the first kernel call takes very long.
I see you run the code with without -aggression, please add -aggression=8  or 7 to you parameter to get more performance. -gputhreads is not needed because it is ignored and hard set to 256.

theoretic calculation rate for GTX480: 1401*15/(3733/32+1194/16)=109 MHash/s   (magic numbers are the count of operations from the binary for this implementation)

Mhh in my theoretic calculation  I assumed 109 MHash/s for a GTX480. I think I must check if my operation count for fermi is right.
Can someone with a GTX480 run the code under linux without -gpugrid.

I've just added the 64-bit build, in a separate branch https://github.com/cdmackie/rpcminer-mod/tree/x64.

Grab the whole repo (or just bin) and run it from within bin as before.

I don't imagine you'll see much difference, but if anyone could try it out so I can make it the default.

Now must sleep...

Here was my output fwiw. GTX 480

C:\miners\rpcminer nvopt>rpcminer-mod-cuda.exe -url=api.bitcoin.cz:8332 -user=user -password=password -gpugrid=512 -gputhreads=480 -gpu=0
Client will start 1 miner threads
Work will be refreshed every 4000 ms
2 GPU GUDA device(s) found
CUDA initialized
Search Configuration for gpu named: GeForce GTX 480
Your device:
   - 15 streaming multiprocessors (SM)
   - can run sm version 200
   - bitcoin miner is optimized for 3 gpugrids per SM
   - max value for gpugrid parameter is 65535
max nonce 3FFFFFFF
Done allocating CUDA resources for (512,256)
iterations on gpu=32 blocks=512 threads=256
Iterations: 255,hashs per iteration 15994212185991544832
Target = 00000000ffffffffffffffffffffffffffffffffffffffffffffffffffffff
Found nonce 364CFF6C
2013-04-08 07:11:01 Found Hash!
Hashs found: 1
Sending to server: {"method":"getwork","params":["000000029bddf49cc3f72f75bf4552
d589df0fb9ec95184c247082e00000009200000000ffb19deae961efc905812ff5b1fcf69ec34f2 7
c9d0ba4194749c538e3ceb2ca751626d7e1a022fbe6cff4c3600000080000000000000000000000 0
000000000000000000000000000000000000000000000000000000000080020000"],"id":1}
Server sent: {"result": true, "id": "1", "error": null}
Found nonce AE9CEA4
2013-04-08 07:11:03 Found Hash!
Hashs found: 2
Sending to server: {"method":"getwork","params":["000000029bddf49cc3f72f75bf4552
d589df0fb9ec95184c247082e000000092000000000c048fe68ab3317714aaee783c242a630feb0 1
86961ef159df4949318f8688e051626d851a022fbea4cee90a00000080000000000000000000000 0
000000000000000000000000000000000000000000000000000000000080020000"],"id":1}
Server sent: {"result": true, "id": "1", "error": null}
137694 khash/s

I did a full repull of the files from charlie's github and it is working now, using only 1 of my 2 gpu's. I did uninstall the CUDA dev prog as well after seeing you did, and pulled the correct file it needed from github.

As of now, not seeing any difference in hash rate on a Fermi card, but I assume that's what was expected.

@charliemaggot

I have updated bitcoinminer.cpp in my repo. I disable the hard crash if a error came out of the device selection. Normaly if no cudaSetDevice is called the driver select the first device. I have not tested the code because I do the changes in my webbrowser, but I think it compile^^

I have deinstalled my cuda and run it under windows withe the dll from the repo. All worked fine.

Can you please post the commandline which you use to start the miner and all output from the miner.

Did you try the latest?...from about 3hrs ago, as there was a bug fix if no gpu parameter was specified.

I'm seeing the same results, yesterdays build I was able to run, today with the changes.. aka required cudart and then when having it provides the same output..

GTX 680

Mh, the selection of your device crash. I add later some more output which device is selected. In most cases there is a problem with the driver.

I sent you my rpcminer.dmp file from most recent attempt.

Trying to set the -gpu flag made no difference. The video driver stops responding if that helps.

I've added the cudart32_50_35.dll to the binaries so it can be downloaded, although you've probably got it by now.

I think there is a bug if you don't specify the gpu flag, e.g. using "-gpu" and not something like "-gpu=0". So can you try explicitly setting it 0, 1 or whatever.

If you still get an error, there should be a rpcminer.dmp file created in the same folder. Can you send it to me, charlie (at) legmail.com

I'm trying out the windows version with a GTX 670 and it doesn't seem to be working

I can confirm that I've installed CUDA dev software 5.0
I also have updated drivers including cudart32_50_35.dll

When clicking on the rpcminer-mod-cuda.exe file the black cmd screen comes up and an error window immediately says "rpcminer-mod-cuda.exe has stopped working"

On the cmd window in the background it says

Client will start 1 miner threads
Work will be refreshed every 4000 ms
1 GPU CUDA device(s) found
:54[CUDA] Error: invalid device ordinal
Error:invalid device ordinal

When I run the debugger it tells me there is a Run-Time Check Failure #0 - The value of ESP was not properly saved across a function call.  This is usually the result of calling a function declared with one calling convention with a function pointer declared with a different calling convention

To be honest I wasn't sure where exactly to extract the zip files, or how to make a file that alters how the executable runs...  Any help will be greatly appreciated.  Post your address and I'll tip you my first bitcent or so

I am 1 version behind on drivers, will download the latest ones.

Latest error with newest rpcminer-mod-cuda (1 version behind on drivers fyi)

Client will start 1 miner threads
Work will be refreshed every 4000 ms
2 GPU GUDA device(s) found
:54[CUDA] Error: invalid device ordinal
Could not retrieve work from RPC server.
CURL return value = 7

edit: Latest drivers, 314.22, same error as above.

I get the same error, no I never installed cuda since I wasnt compiling this, I have the latest nvidia drivers tho since this is my uber gaming rig..

any chance to include the dll?

EDIT: .. installed 32bit version of cuda 5 from nvidias site.. now when I run the miner I the program crashs.. its not missing the dll .. but

Client will start 1 miner threads
Work will be refreshed every 4000 ms
1 GPU GUDA device(s) found
:54[CUDA] Error: invalid device ordinal

Do you have 32-bit CUDA SDK 5.0 installed?

Ok, whilst building a debug version to try and track your issue down, I found out that there is an error in the error handler. So based on the line you had the error, it's that the CUDA device driver didn't initialise properly. Do you have the right and latest drivers?

I've fixed the error handler and uploaded a new binary so you should see a more meaningful message.

I can confirm im getting the same thing..

when running rpc a window pops up and says

the program cant start because cuda32_50_35.dll is missing from your computer..

seems something went whacko

Downloaded & installed the x64 CUDA dev software & the .exe is just crashing now. I re-downloaded all the files this morning & was getting an error about the cudart32 file, & that's when I downloaded the CUDA package.

It begins to start, then crashes after outputting:

Client will start 1 miner threads
Work will be refreshed every 4000 ms
2 GPU GUDA device(s) found
:54

Event log below:

Faulting application name: rpcminer-mod-cuda.exe, version: 0.0.0.0, time stamp: 0x5161065f
Faulting module name: rpcminer-mod-cuda.exe, version: 0.0.0.0, time stamp: 0x5161065f
Exception code: 0x40000015
Fault offset: 0x0002eadc
Faulting process id: 0x19ac
Faulting application start time: 0x01ce33c69eee3f83
Faulting application path: C:\miners\rpcminer nvopt\rpcminer-mod-cuda.exe
Faulting module path: C:\miners\rpcminer nvopt\rpcminer-mod-cuda.exe
Report Id: dd0fe4b3-9fb9-11e2-94d6-001fbc083cc7

Can't test it under linux. Thats a Laptop with 2 GPUs it was not possible to run the miner under linux :-(

I hope we get some results from a user with a GTX Titan.

@InqBit Just seems like the file is corrupted. Does it match the one on gibhub? Can you try downloading it again? You will need to 32bit CUDA SDK, or at least cudart32_50.dll if that isn't included in the 64bit version. I hadn't built 64bit, as there were 3rd party dependencies, but I might try and them them later.

@psychocoder Seems there are some other ways around the time-out issue:

1) As you say, change grid values
2) Unplug the monitor from the card... assuming you aren't using it
3) Change registry settings to increase the time-out. See http://msdn.microsoft.com/en-us/library/windows/hardware/ff569918(v=vs.85).aspx. Perhaps only for the hardcore and if it's going to make a significant difference.

Were the 22.3-22.9 values you got on your GT555M in Windows what you expected? Is it the same as Unix?