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

[see360]

cbuchner1, did you create cudaminer.com? Looks maybe legit (no direct download or donate buttons), but then it might be worth keeping an eye on if it's not your handiwork.

[masher]

Just wanted to report that with my 750ti Asus I can get a marginally higher hashrate when mining (scrypt) with x86 Cudaminer as opposed to x64.  On Windows 8.1 64bit.  Running bios mod, overclocked 130 clock and 577 mem. Hashrate now hitting 320khs as opposed to a 315khs average before. Also, before the bios mod I was hitting around 280khs.

Using -H 1 -T15x24

[cbuchner1]

cbuchner1, did you create cudaminer.com? Looks maybe legit (no direct download or donate buttons), but then it might be worth keeping an eye on if it's not your handiwork.

looks like a fan site to me. No download links provided yet, but I will keep an eye out for changes on this site.

Christian

[Decagrog]

Have you tried to look into the task manager just to see what process was causing that ? (can't be cudaminer if it is closed... could be windows explorer though... had something really strange yesterday while plugging my phone into my computer...)

Yeah, there isn't anything visible at all (this is *with* all processes shown for all users).

However I might be making some headway:
Previous settings:
GPU Clock Offset: 0
Memory Clock Offset: +500

I did this because I read somewhere that the GPU clock offset didn't seem to have much of an impact on hashing speed.

Anyway I just reset the memory clock offset from +500 to 0, and increased the GPU clock offset from 0 to 20. It was stable (but hashing slower). I cancelled Cudminer and it was responsive in closing. The system CPU usage wasn't through the roof after closing either. I repeated this with a GPU Clock Offset of 40. Same result. System fine. CPU usage looking healthy / normal.

I'm going to try to confirm this next, but this appears to suggest increasing the Memory Clock Offset puts a lot more pressure on the system both before and after(?!) mining.

Look at my post 2 pages behind, (I've posted some screeshot) and check the "system interrupts" process in the windows resource monitor. You can access to resource monitor from the task manager panel

[ManIkWeet]

Are you guys using any remote desktop-ish tool?
I believe that gives major issues.

[ltcnim]

i'd love to see a few lines of static text in the cmd (like cgminer for example). So, each card should be listed with temp, last khash, valid/invalid shares etc. is this maybe possible? that way you would have a short overview of your gpu's.

I'm not familiar with cgminer's output, but I gave Christian the beginnings of a patch to enable cudaminer to print GPU temps, fan speed, and power consumption data as reported by the NVIDIA drivers.  Not all GPUs are capable of reporting all of those measurements, but all recent GPUs will report temperature and fan speed.  Here's an example log from my patched version of cudaminer.  I'm sure that it can be improved a lot beyond the code I sent Christian, but this is at least a starting point:

[2014-03-09 23:07:24] Stratum detected new block
[2014-03-09 23:07:24] GPU #0: Tesla K20c, 537.70 khash/s
[2014-03-09 23:07:24]         Temp:  68C  Fan speed:  68%  Power: 186W
[2014-03-09 23:07:24] GPU #2: GeForce GTX 680, 321.40 khash/s
[2014-03-09 23:07:24]         Temp:  88C  Fan speed:  72%  Power:  N/A
[2014-03-09 23:07:24] GPU #1: GeForce GTX 750 Ti, 266.37 khash/s
[2014-03-09 23:07:24]         Temp:  66C  Fan speed:  53%  Power:  N/A
[2014-03-09 23:07:24] GPU #3: Tesla K20c, 541.51 khash/s
[2014-03-09 23:07:24]         Temp:  72C  Fan speed:  78%  Power: 188W
[2014-03-09 23:07:42] DEBUG: hash <= target
Hash:   0000006504c7edb989ead541e02089f081a84d44157c58ae617f37e79503987f
Target: 0000007fff800000000000000000000000000000000000000000000000000000
[2014-03-09 23:07:42] GPU #1: GeForce GTX 750 Ti, 266.31 khash/s
[2014-03-09 23:07:42]         Temp:  66C  Fan speed:  54%  Power:  N/A
[2014-03-09 23:07:42] accepted: 17/17 (100.00%), 1667 khash/s (yay!!!)

this is already pretty nice, and in my eyes a big improvement. however, i'd really like to have a static part where you get all the info at one, and then the lower part could be the way its currently. in the end it could look like this (just a dirty 5min job):

[Decagrog]

this is already pretty nice, and in my eyes a big improvement. however, i'd really like to have a static part where you get all the info at one, and then the lower part could be the way its currently. in the end it could look like this (just a dirty 5min job):

+1
A clearer informative header would be very nice

[liomojo1]

Use cuda manager for now , it gives you all that info.

[gfgrimm]

Are you guys using any remote desktop-ish tool?
I believe that gives major issues.

(Assuming Windows environment) I use VNC to start/re-start cudaminer. For all other purposes, I use RDP for other managerial tasks. RDP does not allow for CUDA to run because of the way it works.

[jsamul]

My 780 gets 220kh on n-scrypt, any way to boost a bit more out of it?
cudaminer.exe --algo=scrypt:2048 -H 1 -C 0 -m 1 -d 0 -i 0 -l T12x24 -o blah blah

I'm using The
T10x20
I'm passing -H 2
No C, no -m

Get close to 300khash/s constant on my 780

cudaminer.exe --algo=scrypt:2048 -H 2 -d 0 -i 0 -l T10x20 got me only like 109.

The best I could do was:

 cudaminer.exe --algo=scrypt:2048 -H 2 -d 0 -C 2 -i 0 -m 0 -l T64x2

If you do -H 1, you could get 5-10kh/s more. It helped 3/5 of my 780's.

man, those 5-10khps are from CPU
watch your CPU load when changing the -H switch...

[jsamul]

would there be any benefit from implementing DirectCompute 5.0 for new Maxwell cards?

[cbuchner1]

would there be any benefit from implementing DirectCompute 5.0 for new Maxwell cards?

I am not qualified to answer this. For sure it would kill cross platform support.

[antonio8]

Use cuda manager for now , it gives you all that info.

Can you run more than 1 card in Cuda Manager?

I have 3 cards by "adding miner to manager" but only 1 card is running.

[djm34]

Use cuda manager for now , it gives you all that info.

Can you run more than 1 card in Cuda Manager?

I have 3 cards by "adding miner to manager" but only 1 card is running.

you proceed in the same way as for the standalone cudaminer -d card1,card2 etc...

[tachyon_john]

i'd love to see a few lines of static text in the cmd (like cgminer for example). So, each card should be listed with temp, last khash, valid/invalid shares etc. is this maybe possible? that way you would have a short overview of your gpu's.

I'm not familiar with cgminer's output, but I gave Christian the beginnings of a patch to enable cudaminer to print GPU temps, fan speed, and power consumption data as reported by the NVIDIA drivers.  Not all GPUs are capable of reporting all of those measurements, but all recent GPUs will report temperature and fan speed.  Here's an example log from my patched version of cudaminer.  I'm sure that it can be improved a lot beyond the code I sent Christian, but this is at least a starting point:
(chopped log...)

this is already pretty nice, and in my eyes a big improvement. however, i'd really like to have a static part where you get all the info at one, and then the lower part could be the way its currently. in the end it could look like this (just a dirty 5min job):
(chopped image...)

The NVIDIA APIs don't report fan RPM, but rather than commanded fan speed as a percentage of full speed.  And the NVIDIA APIs only report the commanded speed, not the actual fan speed -- if you stop the fan with your fingers, it can't tell that this has happened.   To draw a header like the mock up you showed would likely be nothing more than just printing a full page when the log scrollls, starting with the header.  The only thing I wouldn't like about that is that if you wanted to keep a log in a file, heaving the header show up in the log file wouldn't be very desirable.  Perhaps the logging code could be setup so that it logs the header to stdout, but not to stderr, or something similar, so that it would be easy to exclude from I/O redirection, at least on Linux.  One could also just add new flags that enable a variety of optional logging behaviors.

[djm34]

i'd love to see a few lines of static text in the cmd (like cgminer for example). So, each card should be listed with temp, last khash, valid/invalid shares etc. is this maybe possible? that way you would have a short overview of your gpu's.

I'm not familiar with cgminer's output, but I gave Christian the beginnings of a patch to enable cudaminer to print GPU temps, fan speed, and power consumption data as reported by the NVIDIA drivers.  Not all GPUs are capable of reporting all of those measurements, but all recent GPUs will report temperature and fan speed.  Here's an example log from my patched version of cudaminer.  I'm sure that it can be improved a lot beyond the code I sent Christian, but this is at least a starting point:
(chopped log...)

this is already pretty nice, and in my eyes a big improvement. however, i'd really like to have a static part where you get all the info at one, and then the lower part could be the way its currently. in the end it could look like this (just a dirty 5min job):
(chopped image...)

The NVIDIA APIs don't report fan RPM, but rather than commanded fan speed as a percentage of full speed.  And the NVIDIA APIs only report the commanded speed, not the actual fan speed -- if you stop the fan with your fingers, it can't tell that this has happened.   To draw a header like the mock up you showed would likely be nothing more than just printing a full page when the log scrollls, starting with the header.  The only thing I wouldn't like about that is that if you wanted to keep a log in a file, heaving the header show up in the log file wouldn't be very desirable.  Perhaps the logging code could be setup so that it logs the header to stdout, but not to stderr, or something similar, so that it would be easy to exclude from I/O redirection, at least on Linux.  One could also just add new flags that enable a variety of optional logging behaviors.

Is this really needed ?
I mean, on windows MSI AB gives already these infos (and some others more important) and on linux there is nv_smi (or something like that) which can give you info about the card temp, fan, clocks speed)
As long as it won't be possible to overclock the card from cudaminer (obviously this is the goal), you will still need to have MSI AB opened to be able to overclock the card, so you already see the temps/fan speed etc... so...

edit: Must admit that MSI AB is quite messy when there are several cards on the system

[ManiacMiner]

Today NVIDIA released new GeForce 335.23 WHQL driver.

•Performance
 •Enhanced GPU clock offset options for GeForce GTX 750Ti / GTX 750

[cbuchner1]

Today NVIDIA released new GeForce 335.23 WHQL driver.

•Performance
 •Enhanced GPU clock offset options for GeForce GTX 750Ti / GTX 750

yes, especially in conjunction with the BIOS TDP unlock, you can now get higher clocks than before
(assuming your card doesn't crash trying....)

Christian

[gazzamc]

this is already pretty nice, and in my eyes a big improvement. however, i'd really like to have a static part where you get all the info at one, and then the lower part could be the way its currently. in the end it could look like this (just a dirty 5min job):

+1
A clearer informative header would be very nice

+1 Much cleaner and nicer on the eye...

Today NVIDIA released new GeForce 335.23 WHQL driver.

•Performance
 •Enhanced GPU clock offset options for GeForce GTX 750Ti / GTX 750

Will wait for others to try it before i do hopefully it improves speed even more..

[jack80]

Sorry I have installed 2 msi 750ti

how should I write in .bat cudaminer ?

im use now string for 1 750ti and is :

cudaminer.exe -o stratumblabla -u blabla - p bla bla    -l T10x24 -i 0

.