Author

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 824. (Read 2347664 times)

full member
Activity: 201
Merit: 100
No gains with p0 state on 960, 970 and 980. Only ETH gains some % with memory oc.
Anyway I managed to put all the cards to p0 state also in windows. It is the same "trick" as in linux using nvidia-smi tool.

I'm not responsible if You kill Your GPU. The risk is Yours.
The procedure is:
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 -ac 3505,1455 -i 0

-i 0 means display adapter 0. -i 4 would mean display adapter 4.
Step 3 is used to show the possible clock settings combination for each power state. The numbers will change if You overclock Your gpu before running step 3.
The result after typing step 3 would be looking like that:
Code:
C:\Windows\system32>cd C:\Program Files\NVIDIA Corporation\NVSMI

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


==============NVSMI LOG==============

Timestamp                           : Sat Jan 02 19:09:48 2016
Driver Version                      : 359.06

Attached GPUs                       : 5
GPU 0000:01:00.0
    Supported Clocks
        Memory                      : 3505 MHz
            Graphics                : 1635 MHz
            Graphics                : 1623 MHz
            Graphics                : 1610 MHz
            Graphics                : 1598 MHz
            Graphics                : 1585 MHz
            Graphics                : 1572 MHz
            Graphics                : 1560 MHz
            Graphics                : 1547 MHz
            Graphics                : 1534 MHz
            Graphics                : 1522 MHz
            Graphics                : 1509 MHz
            Graphics                : 1496 MHz
            Graphics                : 1484 MHz
            Graphics                : 1471 MHz
            Graphics                : 1458 MHz
            Graphics                : 1446 MHz
            Graphics                : 1433 MHz
            Graphics                : 1420 MHz
            Graphics                : 1408 MHz
            Graphics                : 1395 MHz
It always starts with the clocks of the p0 state. In my example it shows memory 3505 after that You can find listed all the possible graphics settings. You must chose one and type it exactly the same number in the step 4.

Example:
nvidia-smi -ac 3505,1446 -i 0
The nvidia inspector shows Your card in p0 and You can overclock also the memory with msi afterburner or any other oc tool.

If You type nvidia-smi -ac 3505,1440 -i 0 it won't be accepted because 1440 is not in a list of matching Mhz for memory clock 3505.

I'm not responsible if You kill Your GPU. The risk is Yours.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
I am failing to  p0 state on my gtx 960 and gtx 970.

nvidiaInspector.exe -setPstateLimit:0,0
nvidiaInspector.exe -setPstateLimit:1,0
nvidiaInspector.exe -setPstateLimit:2,0
nvidiaInspector.exe -setPstateLimit:3,0

nvidiaInspector.exe -forcepstate:0,3
nvidiaInspector.exe -forcepstate:1,3
nvidiaInspector.exe -forcepstate:2,3
nvidiaInspector.exe -forcepstate:3,3


how can I do it with nvidia inspector?
legendary
Activity: 1176
Merit: 1015
@sp  It would be nice if somehow in your ccminer ..you could get the 980ti  to stay at the p0 state..it hashs at the p2 state...which is about 20% less hashs.....my guess. So the hashrates we see on the 980ti is at only 80% efficiency.

My 970s have always hashed at p2. Maybe cause of the Linux drivers.

Powerstate only affects memory hard algos on my 970 windows rig, at p2 state it is impossible to overclock memory. Nvidia Inspector works for me, just choose overclocking - P2 - slider to the right - save.

Resets at reboot.
Thx  I'll try it.......but 980ti at the p2 state clocks are at 1/2  so imagine if we got them to p0  100% and then overclocked. Its got to add some hashrate...a lot Cheesy

Are you sure about that? Under ccminer load your core clocks should be somewhere around 1400-1500MHz area.
legendary
Activity: 3164
Merit: 1003
@sp  It would be nice if somehow in your ccminer ..you could get the 980ti  to stay at the p0 state..it hashs at the p2 state...which is about 20% less hashs.....my guess. So the hashrates we see on the 980ti is at only 80% efficiency.

My 970s have always hashed at p2. Maybe cause of the Linux drivers.

Powerstate only affects memory hard algos on my 970 windows rig, at p2 state it is impossible to overclock memory. Nvidia Inspector works for me, just choose overclocking - P2 - slider to the right - save.

Resets at reboot.
Thx  I'll try it.......but 980ti at the p2 state clocks are at 1/2  so imagine if we got them to p0  100% and then overclocked. Its got to add some hashrate...a lot Cheesy
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
I'm having trouble compiling with CUDA 7.5.  I installed 7.5.18 from the nVidia web site but when I run ccminer it still says compiled with CUDA 6.5?  I'm stumped.  Undecided

Do you have the latest version of the sourcecode?

release 78 has this in the heading:

Compiled with Visual C++ 18 using Nvidia CUDA Toolkit 7.5
legendary
Activity: 1176
Merit: 1015
@sp  It would be nice if somehow in your ccminer ..you could get the 980ti  to stay at the p0 state..it hashs at the p2 state...which is about 20% less hashs.....my guess. So the hashrates we see on the 980ti is at only 80% efficiency.

My 970s have always hashed at p2. Maybe cause of the Linux drivers.

Powerstate only affects memory hard algos on my 970 windows rig, at p2 state it is impossible to overclock memory. Nvidia Inspector works for me, just choose overclocking - P2 - slider to the right - save.

Resets at reboot.
sr. member
Activity: 346
Merit: 250
I'm having trouble compiling with CUDA 7.5.  I installed 7.5.18 from the nVidia web site but when I run ccminer it still says compiled with CUDA 6.5?  I'm stumped.  Undecided
legendary
Activity: 2716
Merit: 1094
Black Belt Developer
@sp  It would be nice if somehow in your ccminer ..you could get the 980ti  to stay at the p0 state..it hashs at the p2 state...which is about 20% less hashs.....my guess. So the hashrates we see on the 980ti is at only 80% efficiency.

My 970s have always hashed at p2. Maybe cause of the Linux drivers.
legendary
Activity: 3164
Merit: 1003
@sp  It would be nice if somehow in your ccminer ..you could get the 980ti  to stay at the p0 state..it hashs at the p2 state...which is about 20% less hashs.....my guess. So the hashrates we see on the 980ti is at only 80% efficiency.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
I linked with 6.5, it was the git HEAD of yesterday.

I am tweaking to run faster with the cuda 7.5 compiler. Cuda 6.5 is untested since release 74.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
-faster qubit (+5%)
-faster blake/blakecoin (+6.5% since release 74)
-faster doom(luffa 512) (+10%) (compute 5.2)
-faster x11
-faster lyra2v2

1.5.78(sp-MOD) is available here: (02-jan-2015)

https://github.com/sp-hash/ccminer/releases/

The sourcecode is available here:

https://github.com/sp-hash/ccminer

(Cuda 7.5 only. No merged kernals)

Please donate to get more hashrate.
legendary
Activity: 2716
Merit: 1094
Black Belt Developer
I linked with 6.5, it was the git HEAD of yesterday.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
I added 0.5MHASH to the qubit algo and 20MHASH in the doom algo(10% faster). on the gtx 970 @github

I compared the latest qubit(release 77+++) with the release 74.

It's faster on the 750ti. (+50Khash)
slower on gtx 960 and gtx 970. (-200khash)
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
I'm not a big fan of the hybrid option, seems simpler to manage multiple executables, but I'll throw in my 2 satoshis.

I am not a fan eighter. So I try to make small adjustments to the code to aid the cuda7.5 compiler.
sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
I tried making a quick and dirty hybrid ccminer according to this table, to mine X11.  Unfortunately it is slower. Maybe it's because I'm on Linux...

I am profiling the release 74 exe file. (If you want to build you need to extract the sourcecode from the 6.5 version wich is release 74)

the command line can be something like:

nvprof ccminer74.exe -a x11 --benchmark -d 0
and compare with:
nvprof ccminer77.exe -a x11 --benchmark -d 0

sp_
legendary
Activity: 2954
Merit: 1087
Team Black developer
I added 0.5MHASH to the qubit algo and 20MHASH in the doom algo(10% faster). on the gtx 970 @github
legendary
Activity: 1470
Merit: 1114
legendary
Activity: 2716
Merit: 1094
Black Belt Developer
legendary
Activity: 2002
Merit: 1051
ICO? Not even once.
Happy New Year!
full member
Activity: 173
Merit: 100
Happy New Year
Jump to: