Author

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 694. (Read 2347641 times)

sr. member
Activity: 445
Merit: 255
SP, just a FYI, your version is still not getting as good hash as DJMs Neoscrypt kernel.

Also another FYI, after DJM updated his kernels, it increased power usage by about 15%.
On the 970 I'm getting 700 kh/s. sp 5 private.  Shocked  = to 3.5 750ti. The best one you have so far sp for ALL algo's.

Hi

have you made some watt-meter measurements at 700KH/s ? What clock speed and batch command line?

Thank you in advance
legendary
Activity: 3164
Merit: 1003
check that the sata for the riser is separate, maybe try installing just the one card and see then if it will change anything. maybe check the asic quality with gpu-z to see what comes up.
I just ran it then activated nvidia inspector ...crash!  So I updated it a week ago but I also had trouble getting the drivers to install ... as soon as I added that card..... everything started crashing.
Gpu-z how do I check the asic quality with that?
I also had all memory clocks show 4505 on my 2x 980ti and the one 970..then I added the second one and the memory clocks showed 4505 on the 980ti and 5109 on both 970's. Like they linked up.

Looks good....before crash.

that hashrate is impressive are you using sp mod or what? i can barely scratch 13.5 mega with 1500 core
sp private #5 all algo and that is stock clocks... it can do 1 mh/s more on quark.
And I think it was in p2 state.
EDIT: I thought that the 970 fans where suppose to be off when idle mine are not.



THANKS everyone for helping me.. very much appreciated Smiley  That hashrate on the 970 was at stock clocks.. it was doing 19 mh/s but unstable .. needed to set the intensity of course... it was the 970 causing the problem I think and so far it hashed overnight without crashing ..

PS anyone know how to open up a command window separate from an app using ccminer like MinerControl .. I need to see it like opening a reg ccminer window.  thx
legendary
Activity: 3164
Merit: 1003
SP, just a FYI, your version is still not getting as good hash as DJMs Neoscrypt kernel.

Also another FYI, after DJM updated his kernels, it increased power usage by about 15%.
On the 970 I'm getting 700 kh/s. sp 5 private.  Shocked  = to 3.5 750ti. The best one you have so far sp for ALL algo's.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
SP, just a FYI, your version is still not getting as good hash as DJMs Neoscrypt kernel.
Also another FYI, after DJM updated his kernels, it increased power usage by about 15%.

What cards are you testing on?. How much faster? My kernal does around 1MHASH on the 980ti.
15% more power is a waste.
legendary
Activity: 3164
Merit: 1003
yes you're right, does setting them in p0 has some benefit outside of ethereum mining?
I don't think so... I do have them set at p0 for any algo but thinking of just letting them stay at the p2 state.
full member
Activity: 187
Merit: 100
1CTiNJyoUmbdMRACtteRWXhGqtSETYd6Vd


TXID:becf99c164222d671f5d0c098e6f66626a2c6dc129f5b55c00b857144da24504 Roll Eyes Roll Eyes

0.1BTC
legendary
Activity: 1764
Merit: 1024
SP, just a FYI, your version is still not getting as good hash as DJMs Neoscrypt kernel.

Also another FYI, after DJM updated his kernels, it increased power usage by about 15%.
legendary
Activity: 3248
Merit: 1072
yes you're right, does setting them in p0 has some benefit outside of ethereum mining?
legendary
Activity: 3164
Merit: 1003
can you check for me one thing on your 970, are your memory on 3000M without oc? or even with oc? mine are sitting there and no matter what oc i put they remain there....

something strange happened
Yours are maybe in the p2 state. Mine are 3505,1455 in the p0 state.
I set my clocks to nvidia-smi -ac 3505,1455
When I have just one 970 it shows those clocks... but with 2 cards it showed 3505,1519  strange.
But I still set them as above even though one could be set higher.
Below is something someone gave me.
-----------------------------------------------------------------------

It is possible to oc the mem clock beyond nominal value, I use this command nvidia-smi.exe -i 0,1 -ac 3506,1506 (where i is the card number and ac the highest clock possible) this force the card to run in state 0

1) open cmd.exe with the admin privileges
2) cd C:\Program Files\NVIDIA Corporation\NVSMI
3) nvidia-smi -q -d SUPPORTED_CLOCKS -i 0
4) nvidia-smi -q -d SUPPORTED_CLOCKS -i 1
3) nvidia-smi -q -d SUPPORTED_CLOCKS -i 2
5) nvidia-smi -q -d SUPPORTED_CLOCKS -i 4
6) nvidia-smi -ac 350X,14XX -i 0 (where 350X is that very first number printed after supported clocks-memory and 14XX one of those numbers printed after graphics

cd C:\Program Files\NVIDIA Corporation\NVSMI
nvidia-smi -q -d SUPPORTED_CLOCKS -i 0 | more

cd C:\Program Files\NVIDIA Corporation\NVSMI
nvidia-smi -q -d SUPPORTED_CLOCKS -i 1 | more
---------------------------------------------------------------------

Now that I know my clocks , oc'ing must be off to set them, I use this.
Open administrative cmd window and I paste and enter this. Remember, look up your clocks first.
Two of those cards are 980ti  same clocks.

cd C:\Program Files\NVIDIA Corporation\NVSMI
nvidia-smi -ac 3505,1455 -i 0,1,2,4
legendary
Activity: 3248
Merit: 1072
can you check for me one thing on your 970, are your memory on 3000M without oc? or even with oc? mine are sitting there and no matter what oc i put they remain there....

something strange happened
legendary
Activity: 3164
Merit: 1003
SP  what is the default setting on sp 5 private for quark 970 g1?
EDIT: Looks like 23.
So far lowering the intensity is helping.
EDIT: That may have been the problem....I never set intensity on my regular rigs.
Will see be tomorrow.
This is the way I have it... but its working on all cards on some algo's. downing all the hashrates...and some algo's ..just on the card specified.
-d 0,1,2,3,4  -i 0,0,14,14,0
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
submitted a tiny speedup in jh512.

The groestl + jh change push quark 10-15KHASH on the gtx960M
legendary
Activity: 3164
Merit: 1003
I still get gpu 0 found nonce plus the readout of the card I'm benchmarking. I don't know why.
That's because sp is using the thread id for the "nonce found" message, not the card number.
Thx Schleicher... that is one thing out of the way... trying intensity settings now.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
Submitted a tiny speedup in echo-512. X11 +2KHASH on the gtx960M. @github
 
Hehe I know It is nothing, but I just need to do it 100 more times.
legendary
Activity: 3164
Merit: 1003
legendary
Activity: 3164
Merit: 1003
check that the sata for the riser is separate, maybe try installing just the one card and see then if it will change anything. maybe check the asic quality with gpu-z to see what comes up.
I just ran it then activated nvidia inspector ...crash!  So I updated it a week ago but I also had trouble getting the drivers to install ... as soon as I added that card..... everything started crashing.
Gpu-z how do I check the asic quality with that?
I also had all memory clocks show 4505 on my 2x 980ti and the one 970..then I added the second one and the memory clocks showed 4505 on the 980ti and 5109 on both 970's. Like they linked up.

Looks good....before crash.

that hashrate is impressive are you using sp mod or what? i can barely scratch 13.5 mega with 1500 core
sp private #5 all algo and that is stock clocks... it can do 1 mh/s more on quark.
And I think it was in p2 state.
EDIT: I thought that the 970 fans where suppose to be off when idle mine are not.


legendary
Activity: 3164
Merit: 1003
Try swapping the suspect card with card 0 (in the x16 slot I presume) to see if the probkem moves.

Swapping cards... but gpu0 is not the 16x slot.


I think you're going in the wrong direction. You installed a new card and your system destabilized.
You know the system was stable before but you don't know if the card is defective. That is
your first objective. You need to isolate the problem.

Swap the suspect card with whatever card just to see if the problem follows it. Try removing the suspect card to
get back to a stable system. Try removing all but the suspect card. Look for a pattern.

If the card proves ok then you can look at system problems, starting with power. Do you have enough power?
Do you have too many risers on the same rail?

There appears to be nothing in the software that has changed that could have caused it.



@tbearhere - I was thinking the same thing.  If you don't have enough power feeding the cards the system will act wonky!  What size power supply and how many video cards in total on the board?  Only (2) 970's?
I got multiple problems now... will try that -i setting.  thx  psu 1200 continuous up to 1300 but drawing 1050 watts. It cuts off sometimes too.. it should not...but the fan is too slow.. can't find anything to override the psu fan speed.

The humidity is 18%  Shocked  it is part of 3 problems.
It was hard getting the drivers installed on that card too but worked.
EDIT: Humidity 16% I don't believe it.... Cheesy Can't go near the rig.
legendary
Activity: 3164
Merit: 1003
check that the sata for the riser is separate, maybe try installing just the one card and see then if it will change anything. maybe check the asic quality with gpu-z to see what comes up.
I just ran it then activated nvidia inspector ...crash!  So I updated it a week ago but I also had trouble getting the drivers to install ... as soon as I added that card..... everything started crashing.
Gpu-z how do I check the asic quality with that?
I also had all memory clocks show 4505 on my 2x 980ti and the one 970..then I added the second one and the memory clocks showed 4505 on the 980ti and 5109 on both 970's. Like they linked up.

Looks good....before crash.

Do you have SLI enabled by any chance?
That maybe the problem but I don't know how to do it or disabled it.
Now it only last a minute then crashes.
How do I check to disabled it?  thx
EDIT: I'm checking now  but I didn't enable it unless it does it automatically.
I can't even access nvidia control panel.

tbearhere ...

the sli connection is a physical connector for the cards ... if the cards are separate ( ie - you dont have a phisical connecter ont he top of both cards connecting them to one another ) - then you dont have sli connected ...

unless there is a bios settign as well that you changed - then sli will be disabled there as well - though i dont now of any bios settigns that exist for sli anyway ... but i thought id mention that just in case ...

i also noticed that you are running ccminer with NO -i settings ... have you considered that the default intensity is way too high for the cards? ...

sp doesnt publish the default intensities ( actually none of the devs do - and ive requested this many times before in this thread - but no one has ) - so there may be an intensity issue here ...

try using -i 16 as a lower intensity - which wont mine much - but may just show you that the miner itself is stable ... if it still crashes while intensity is as low as 16 - then im out of ideas ... unless i had the machine infront of me for a full rebuild ...

#crysx
I got multiple problems now... will try that -i setting.  thx  psu 1200 continuous up to 1300 but drawing 1050 watts.
The sli connection yes connection.. I don't have them connected.
full member
Activity: 210
Merit: 100
when i have hardware miner ..good job all
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
I submitted a tiny speedup in the groestl--512 algo @ github.
Jump to: