Topic: [ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX] - page 1050. (Read 3426947 times)

2 x EVGA 780 Classified overclocked to 1300MHz
cudaminer settings: -i 0 -H 1 -m 1 -l T12x16

https://i.imgur.com/O2dXdTf.jpg

WOOOT!!!!!  kernels10 you have been awarded 5k doge.  My conclusion about RDP was 100% accurate and I was able to verify that via chrome remote desktop.

I used RDP to install chrome remote desktop, exited RDP, entered through chrome remote desktop and the scripts started up perfectly.  What this verified is that at least on the GTX 560 Ti RDP does indeed kill the CUDA nVidia drivers upon connection; therefore making it impossible to restart cudaminer.

I'm curious if this is the case with all Microsoft RDP sessions.

Semi-idle question:  Is there community interest in sponsoring some more optimization of the cudaminer code for Kepler GK104 and GK110-based cards?

(I ask about sponsoring, because part of what I have in mind is picking up a new card or two to use for benchmarking.  I'm thinking in particular of trying to optimize for the GK110-based 780 and 780Ti, but I'd also like to make sure my results generalize.  Unfortunately, I'm not a gamer, so outside of using them for research in my day job - where I already have some - I don't have any real use for shiny video cards. :-)

In honesty, I think I threw my best optimization ideas at the version most of you are already running.  *grins*  But there's probably another 5% here and there, which could translate into, say, at least a 10-20kh/sec boost on faster cards.

  -Dave

I completely agree that it was indeed a driver crash that caused the initial hardware errors.  I had them happen several times during testing to get the perfect khash on this card.

However the "unable to query CUDA driver" still happens after reboot.

http://imgur.com/AW36Ct0

Ok, tried overclocking for the first time in my life. GTX 660 Ti

I've cracked the 300 khash/s barrier! with

-H 1 -i 0 -C 1 -D -l K7x32


Thanks everyone for the help.

weird. I got that from an auto run (-H 1 -i 0 -C 2 -D), not sure which version of cudaminer I was using at the time. Also, I am using afterburner to oc.

That's weird, I get 255 khash/s with those settings.
How or why did you choose K31x6? I'm not sure what the numbers mean beyond K = Kepler

Is there some command to find optimal settings for my comp?

Driver crashed? happens to me if I try to push my oc to high, does it still happen after reboot?

Haven't used RDP but I am using chrome remote desktop and haven't had issues.


I am able to get up to about 322kH/s (jumps around 315-326 I'm logged in via chromeRD) under win8 pro using  -H 1 -i 0 -C 2 -D -l K31x6 if you (or anyone else with a 660Ti) wanna try that.

I don't think the -C option does anything for kepler right now though.

Edit: I'm slow

Asus GTX660 TI-DC2OC-3GD5 GeForce GTX 660 Ti Graphic Card


My settings. hash rate is at the end. Card is not overclocked.



-H 1 -i 0 -C 1 -d 0 -l K14x16 (266 khash/s)
-H 1 -i 0 -C 1 -D -l K7x32     (269 khash/s)
-H 1 -i 0 -C 1 -D -l K14x8     (258 khash/s)
-H 1 -i 0 -C 1 -D -l K14x16   (265 khash/s)
-H 1 -i 0 -t 1 -C 2 -l K12x16 (256 khash/s)
-H 2 -i 0 -C 0 -m 1 -D -l K7x32  (261 khash/s)

oh hey, we have the same video card. Mine is 3GB version.

what are your other settings? did you overclock? I'm on stock and can't pass 270 khash/s

K14x16

http://www.asus.com/Graphics_Cards/GTX660_TIDC2O2GD5/

Thanks! Yea, running at -i 0 makes my comp a little slow but its ok

I'm on GTX 660Ti with settings:
-H 1 -i 0 -C 1 -D -l K7x32

With K14x16 I get ~265 khash/sec
With K7x32 I get ~269 khash/sec


Might I ask your settings?

Awesome update! Now my 660ti have 295khash/sec.

So after running my GTX 560 Ti for 2days staight at a solid 280khash on the new 12/18 software, cudaminer out of nowhere spiked the hashrate up to about 450 and the card started giving hardware errors as it can't run that high.

I got the notification from my.mining pool that a worker was down, so I RDP to the machine close out cudaminer and restart my script, no changes made at all.

Now all of a sudden cudaminer is saying, "unable to query CUDA driver version.  Is an nVidia driver installed."
This of course isn't true.

Seeing as how this happened the very first time I ran cudaminer I simply tried to reinstall the driver.  When that didn't work I tried downgrading the driver and still no luck.  I even installed the CUDA development kit and that didn't work either.  I can no longer get cudaminer to launch any of the 3 versions that I have previously used.

I'm very confused at the moment.  The only thing crossing my mind is that maybe when I RDP to the machine the graphic settings are changing for remote desktop and the CUDA driver is being disabled and therefore cannot relaunch.

Anyone ever tried to restart cudaminer via RDP before?
Bigger question is why did cudaminer decide to randomly jump to 450khash after 2 straight days mining at 280?

Thoughts, comments, help, all appreciated.  5k doge to anyone that can help me find a solution.

Lots doge you rich coins wow cudaminer wow doge happy coin.

sort of new to this.. below is my low gpu usuage bat file. I want my gpu to not run so hot and high any thing else i need to add?
it still gives me about 246 kh/s on a gtx 670 with 67c. anyway to make the miner use less gpu when im home so i can play some LoL =]
color 0A
taskkill /f /im "cudaminer.exe"
setx GPU_MAX_ALLOC_PERCENT 40
setx GPU_USE_SYNC_OBJECTS 1
CD "G:\doge\Cudaminer\x64"
cudaminer -o stratum+tcp://server1.cryptovalley.com:3333 -u x -p y

One instance per card, specify which one it should run on with the -d operator, -d 0, -d 1 for the next cudaminer etc, I guess you'd have to make 4 folders for that too.
Also, do you have double precision disabled? Supposedly that gives a small boost

Indeed, I was looking at resources and the RAM is so underutilized. Currently, fans are running at 40% speed. Occasionally rises up but pretty much stable near there. Yeah I am running just a single instance for all four.

So as per your suggestion, I should run 4 instance for one card each or 4 instance running on 4 cards each? I believe the later will fry the GPU...lol

I actually did get near about same boost. Previously was getting 1400'ish, now getting 1600'ish hashing rates. So ~200 kh/s increased breaking down to 50 kh/s per card. Before running with manual config, one card would do at much faster and others were doing pretty slow. Now, all are working at pretty much same pace.

-l T15x32 is the right start.  Try T15x64 and see if it runs, but it will probably either crash or give you completely outlandish results.  I've been unable to get that config running on a Tesla, though I may still poke at it a little more.

What you'll find, though, is that the Quadro is a terrible ROI for mining:  -T15x32 will only use about 2GB of the DRAM on your card (and using more provides no additional benefits).  The ratio of registers and FLOPS to DRAM on the Quadro isn't optimal for mining.  Lower-end cards with less memory perform similarly (or better because they often run at a slightly higher clock rate).  The 780 Ti, for example, has roughly the same compute horsepower but with only 3GB of DRAM, and it's a pretty speedy beast.  Well, at least, in the range of nvidia cards.

I'm a little surprised you're not seeing about 40-50kh/s more on each of your cards, but that may be in the noise.  Are you running one instance of CudaMiner across all four?  You might try running four independent instances if your drivers handle it.

  -Dave

I just created the bat file in the folder like I did in previous release (and still mining with it), never needed to specify path.

Although I did try with path too, still doesn't start. I think it's error in the code, since miner does fire up for a millisecond when I hit the bat file.

Thanks bro. The -l with T15x32 boosted the performance. Now getting about 1600 kh/s and all the GPUs are hashing in near about same rate. nVidia GPU's surely suck for mining. This one is such a high-end GPU and such poor performance. Not sure I'll continue this. The trial run's result will decide if its going any further. Thanks again for your help though.