Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1074. (Read 2347601 times)

TDP limit raised from 100% to 106% with slider in nVidia Inspector/MSI Afterburner.

Mining quark, it bumps right up against the limit and goes over it.

I don't know how to NOT overclock.

OC or die tryin!
-----
SP_, I am trying the build you recommended (http://cryptomining-blog.com/4857-updated-windows-binary-of-the-ccminer-1-5-51-git-fork-by-sp-for-maxwell/)

It is good.

Thanks!

The 980-ti is released  
http://videocardz.com/nvidia/geforce-900/geforce-gtx-980-ti

You can also  try this build

http://cryptomining-blog.com/4857-updated-windows-binary-of-the-ccminer-1-5-51-git-fork-by-sp-for-maxwell/

What is your TDP limit? I have Gigabyte 970 cards and they max out at 112%, but I'm no where close to it without boosting the core. Mining quark, TDP is around 70%.

Is there anyway to see what the max TDP is on cards without purchasing them? Is there a database somewhere?

Use the 32 bit verson. build 51 has some pointer hacks that works best on 32 bit.
Also make sure that you have compute5.2 in the projectfile/makefile when building.

Use nVidia Inspector or MSI Afterburner or GPU-Z or whatever program you choose to graph the GPU power usage.

What I have seen is the power limit being reached by the V45 and later sp_mod ccminer programs, which results in the card throttling itself.
This results in a square wave looking graph of the GPU power usage, and much lower reported and actual hashrate.

These later versions, especially V51 as it is the fastest and most intense, seem to be begging for a higher TDP limit BIOS flash.


V44 is the last version that is a little less intense. It might result in slightly less hashrate than V51 is capable of, but the sustained and steady GPU power usage results in more stable mining performance.

I am dragging my feet on flashing my GTX 970, but I see it as necessary to get the utmost performance out of sp_mod ccminer.

i think the issue with all of this is that very few people are grading the performance on 'stock' cards ...

yes - even when you dont mess with the tweaking of the clocks and voltages - each card is different - thats fine ...

but if we all want a baseline comparison - ALL cards should be run with NO tweaking of any sort ( as sp already includes higher intensity levels in the code anyway ) THEN we can have a baseline dataset to compare to for the various cards out there ...

adjust intensities and voltages and other tweaks as you will after that - and record another reading which gives you a comparison of 'base' to 'tweaked' ...

most of the algos dont require much settings apart from intensity - where other algos require parameters to tweak them further ...

so if there is anything that really needs to be done - its a baseline measurement ( for all the cards ) that the tweaks and settings can be compared to ...

i would do it - but my time is so limited at the moment (and my spreadsheet skills leave very little to be desired ) with the rebuild and organization of the farm and all things associated ...

stock to tweaked to maxed ... thats a table that i would like to see ...

#crysx

thats right ... my apologies ...

i keep forgetting you are tprovot ...

much appreciated ...

#crysx

My quark intensity is set at 24 in startup batch

FYI I ran the same test on X11 at solo.nicehash I am getting 10575 per card for a total of 21150 k/h reported out of ccminer using release 50 setting intensity at 21

however when I run release 51 I am only seeing 10050 per card out of ccminer.

also nicehash is only reporting 13 to 15 M/h total no matter what release I use

they are both NVIDIA brand GTX980 Cards


it may be the card brand that is having the issue.. I have my GPU at +253 and I am seeing temperatures of 65ºC avg.
I can try other algos and see if the same holds true if you want
 

Blake was working the last time I tried. Run -a blake --benchmark. It will return found nounce xxx after a while.. but there are different versions of blake. Blake,pentablake,blake256 etc. Make sure your coin is using then same version Of the hashingn algorithm.

Version 51 has a lower memory usage and higher default intensity. Less memory usage and less power.

@nubminer, joblo,

OK....OK....I get it ppl. I know that these algos can be similar, and tricky, and that they are, as was put to me, 'added features'.......etc. We were mildly drifting off the thread subject (ccminer),.....again, etc. So......back off bitches!!   Oh, comon.....don't get your feelin's hurt. I'm joking here, just kidding on stuff. Really  

Now...we are all just workin' together to help get the most we can from this code (ccminer). That's all this is about. As for the difference in speed between 50 vs, 51. I've been running the latest compile for 51 via Miner Control, which I did a re-write on for the nVidia version, am have got to say that 51 runs much more stable than 50. It's been running for well over 28hrs. now @ 6.5+ Mh mining quark on my 750Ti with these settings from the default...
 
ASUS GTX750Ti DFseries 2GB /w 6pin ribbon using GPUTweak, GPU +172, mem -60, fans auto, the temp never over 72C, voltage never over 1143mV, these .bat switches -q -i 22.9 -r 3 -R 10 --cpu-priority 2 -a quark. I have no control /w voltage as it is hard-coded and NONE of the OC programs will unlock. I did try to re-flash but that isn't allowed either. So this card is maxed.

That works for MY card. There are a bunch of other GTX750Ti cards out there, and they are all different. You have to find what works for yours. Plain and simple. But, I may have given you guys a place to start. I just know what works for me and I'm retired so I have time to play.

It was put to this group for someone to try the blake algo. So....I did, and was only relaying the results I got (not knocking anything/anyone). I could have posted a screen shot, but that's a pain in the ... never mind, the error was.. H-not-zero...I have no clue what that refers to, don't care, just know it didn't work in pool or solo. I test things, I push to the max just to see what it does. I couldn't get '-a blake' to function. That's all, so have fun, let me know what needs to be done, and I'll try it. For now....I'M LOVIN' 51.

Happy mining    

in my signature...

Are they built the same? IIRC 32 bit builds perform better than 64, and cuda 6.5 performs better
than 7. Also are they generating code for the same compute versions? Do you get the same results
with the pre-built binaries?

I couldn't even fathom trying to improve what they have done... I am just curious as to why v51 that should be faster for quark is considerably slower on my cards.
And I don't expect anybody to do it for me so I figure I would at least make a vain attempt at it..

CapnBDL,

I think you quoted the wrong conversation, but no matter.

The point I was trying to make to you is that you should moderate your expectations regarding new
algos that have similar names as existing ones. They may not be as similar as they may seem and may
require significant development to implement. It's more like a new feature than a bug fix.

just out from left field here ...

siacoin is going to be released in a few days - and works off blake2b algo ...

any chance we could get that in also to mine sia? ...

#crysx

As far as trying to reverse engineer this....ya really can't improve much on what sp_ has done. He, and the gang, are the best. I just test the product and let them figure it out. Just sayin....   Current code for blake mining returns the mining error...  H-not-zero.  Work on that.

That's curious. AFAIK changing the target architecture only affects the CPU code, not the Cuda code. If 64 works
but 32 doesn't there must be something messed up with your 32 bit environment, and it seems it's been that way
for a while.

You did AWESOME..!!!

 I tried this a couple of months ago and was getting the same errors and I thought I was told that I needed the Full version of VS and the community version would not be compatible... I don't have that kind of money so I gave up until I saw your post. (the last time I bought VS was when VB6 came out (yea i know I am older than dirt)

anyway I ended up changing to release x64 and got it to compile just fine.. it is running and stable.

I do have exactly the same results with v50 being 400+ h/s faster than 51 and I am now trying to side by side compare the differences just to reverse engineer this and find out why it is slower...