CCminer(SP-MOD) Modded GPU kernels.

[sambiohazard]

Does this mean anything for mining PC? As far as i understood, our problem is other way round where we need to use 2 PSUs in one PC.

http://www.guru3d.com/news-story/phanteks-shows-power-splitter-with-isolated-dual-system-technology.html

[CoffeeCat]

The Lyra2v2 timings are depended on a good coreclock/memclock and boostclock (no trottle'ing)
The problem is when I mod a kernal to run fast on the 960, it might hurt the 970 performance.
When I mod a kernal to use more power, cards with little power will start to trottle.

The cryptomining blog has compared the 20 last sp-mod versions on the gtx 970:

http://cryptomining-blog.com/wp-content/uploads/2015/11/ccminer-spmod-gtx-970.jpg

What clocks would you recommend for 750 ti and a 980? I thought I remember someone here saying that memory clocks didn't make that much of a difference. Thanks.

[scryptr]

Does this mean anything for mining PC? As far as i understood, our problem is other way round where we need to use 2 PSUs in one PC.

http://www.guru3d.com/news-story/phanteks-shows-power-splitter-with-isolated-dual-system-technology.html

PHANTEKS SYSTEM --

Conceivably, you could run two low-power systems with one 1200W PSU.  Perhaps 2 x motherboards with 4-6 low end 750ti cards each.       --scrypter

[kangjooe]

hello guys... i'm new to this and seek for some advice for mining BTC.

currently i'm using ccminer-1.5.74-git-spmod with bat like this:

Code:

ccminer.exe --algo=x11 -o stratum+tcp://us-west.multipool.us:11111 -u username -p password
pause

i'm joining multipool. but with that setting, i'm not mining BTC but DRK instead. (Dask pools Stats)...   


my spec: i7 2600K @ 4.4 Ghz / GTX 970 jetstream (Default) / 12 GB RAM / Win 10 x64
pool: multipool
try to mining: BTC

so do i have to change pool? or ccminer version? or simply change the setting.bat? i need some advice...

thanks before 

[chrysophylax]

anyway - as for the overclocking side ... i think i will take wolfs advice and take another look into it next week week ... the next 3days for me is incredibly busy ...

Linux vs Windows, cuda 6.5 vs 7.5, x86 vs x86_64. Holy crap hasn't this been hashed and rehashed over and over again?

From my experience mining on Windows or Linux produces the same hash rates everything else being equal. However,
I have noticed that Windows hashing slows more due to other activity, in particular logging in remotely with ssh.
I haven't tried cuda 7.5 on either OS but the hashrates you reported on cuda 7.5 is significantly slower that I get with 6.5
on Linux or Windows. I conclude 7.5 is equally slow on both OS's.

32 vs 64 bit is not about the OS but about how ccminer is compiled. Compiling for 32 bit tends to be a little more efficient
when dealing with mostly 32 bit arithmetic.

Overclocking on Linux has been discussed at length in this thread, including getting it to work headless. Search for
"cooolbits" for everything you need to know.

id like to say that i totally agree with you joblo - but i dont ... as the systems i have running are showing the complete opposite of what you are stating ...

coolbits and oc are NOT working as they should on the gigabyte 750ti oc lp cards i have running ... investigation into this lead me to conclude that ( and i did read this on the nvidia site ) that the support is not currently implemented in the drivers ( cuda and video drivers ) as of yet - and is only available for the higher end cards ...

the cuda 7.5 implementation i have compiling ccminer-spmod IS causing hashing as fast ( and slightly faster ) than cuda 6.5 compiles ...

im not trying to be obnoxious - its a fact ...

sp has these results of slower hashrate using c7.5 as you do - i dont ...

go figure ...

#crysx

[sp_]

What clocks would you recommend for 750 ti and a 980? I thought I remember someone here saying that memory clocks didn't make that much of a difference. Thanks.

You need to find out yourself. But I can give you a hint. The lyra2v2 can do 5.2MHASH@750ti with the sp-mod opensource kernal. But it won't do it with a gpu core of 1500MHASH, and intensity to the max. To mine with a profit you need to know your cards well. Changing the sourcecode in the kernals is not enough.

The opensource kernals kan perform 15-20% bether with a right config and overclocks.

[joblo]

anyway - as for the overclocking side ... i think i will take wolfs advice and take another look into it next week week ... the next 3days for me is incredibly busy ...

Linux vs Windows, cuda 6.5 vs 7.5, x86 vs x86_64. Holy crap hasn't this been hashed and rehashed over and over again?

From my experience mining on Windows or Linux produces the same hash rates everything else being equal. However,
I have noticed that Windows hashing slows more due to other activity, in particular logging in remotely with ssh.
I haven't tried cuda 7.5 on either OS but the hashrates you reported on cuda 7.5 is significantly slower that I get with 6.5
on Linux or Windows. I conclude 7.5 is equally slow on both OS's.

32 vs 64 bit is not about the OS but about how ccminer is compiled. Compiling for 32 bit tends to be a little more efficient
when dealing with mostly 32 bit arithmetic.

Overclocking on Linux has been discussed at length in this thread, including getting it to work headless. Search for
"cooolbits" for everything you need to know.

id like to say that i totally agree with you joblo - but i dont ... as the systems i have running are showing the complete opposite of what you are stating ...

coolbits and oc are NOT working as they should on the gigabyte 750ti oc lp cards i have running ... investigation into this lead me to conclude that ( and i did read this on the nvidia site ) that the support is not currently implemented in the drivers ( cuda and video drivers ) as of yet - and is only available for the higher end cards ...

the cuda 7.5 implementation i have compiling ccminer-spmod IS causing hashing as fast ( and slightly faster ) than cuda 6.5 compiles ...

im not trying to be obnoxious - its a fact ...

sp has these results of slower hashrate using c7.5 as you do - i dont ...

go figure ...

#crysx

I know coolbits doesn't work. Flash your GPUs.

I'm not willing to go as far as flashing FW but coolbits worked for me.

Crysx, I don't think you are being obnoxious, just maybe a little stubborn. Your results are counter to others in
two ways that are contradictory. First you say Cuda 7.5 is faster than 6.5, while it's the reverse for everyone else.
But then your 7.5 hash rate (2800) is lower than others' 6.5 hash rate (3100). Putting both of those facts together
and your 6.5 hash rate would be even lower than 2800.

In order to accept your facts I would have to believe three things:

1. your particular 750ti's can't perform to the same level as any others,
2. for some unique reason you don't suffer the 10% loss in performance between 6.5 and 7.5 that anyone else experiences,
3. it's just a coincidence that your hash rates are lower by the same amount of degradation expected between 6.5 and 7.5.

If it was happening to me and potentially costing me 300 KH per card I'd want to get to the bottom of it.

[hashbrown9000]

getting coolbits to work on linux does require some legwork but it does work headless at least on driver version 346.59 + Cuda 6.5

I just ran 2x Gigabyte WF GTX 750ti cards to their max on lyra2v2. I can't get much past 1358 MHZ core clock and +100 MHz mem clock without them crashing on Win 10: 32 bit, Release 66.  Best I could do was 4900 kH/s.  Once I start actively using the computer, the card that displays on the monitor crashes.

[sp_]

getting coolbits to work on linux does require some legwork but it does work headless at least on driver version 346.59 + Cuda 6.5

I just ran 2x Gigabyte WF GTX 750ti cards to their max on lyra2v2. I can't get much past 1358 MHZ core clock and +100 MHz mem clock without them crashing on Win 10: 32 bit, Release 66.  Best I could do was 4900 kH/s.  Once I start actively using the computer, the card that displays on the monitor crashes.

If you want to clock higher you need to modify the tdp in the bios from 38W to 63W. And you probobly need powered risers/or cards with a 6pin power / or motherboards with extra power.

[antantti]

getting coolbits to work on linux does require some legwork but it does work headless at least on driver version 346.59 + Cuda 6.5

I just ran 2x Gigabyte WF GTX 750ti cards to their max on lyra2v2. I can't get much past 1358 MHZ core clock and +100 MHz mem clock without them crashing on Win 10: 32 bit, Release 66.  Best I could do was 4900 kH/s.  Once I start actively using the computer, the card that displays on the monitor crashes.

If you want to clock higher you need to modify the tdp in the bios from 38W to 63W. And you probobly need powered risers/or cards with a 6pin power / or motherboards with extra power.

I had one ETH mining rig with 3x750ti. With ubuntu 14.04, cuda 7.0 or 7.5 and 35?. drivers overclocking was easily done. Genoil cudaminer thread where I found all the info I needed.

Hashbrown, you have 32-bit W10? Or 32-bit ccminer?

[impulse2000]

getting coolbits to work on linux does require some legwork but it does work headless at least on driver version 346.59 + Cuda 6.5

I just ran 2x Gigabyte WF GTX 750ti cards to their max on lyra2v2. I can't get much past 1358 MHZ core clock and +100 MHz mem clock without them crashing on Win 10: 32 bit, Release 66.  Best I could do was 4900 kH/s.  Once I start actively using the computer, the card that displays on the monitor crashes.

http://gpuz.techpowerup.com/15/11/27/prx.png
http://gpuz.techpowerup.com/15/11/27/csq.png

[hashbrown9000]

yes, all of my gigabyte windforce models have the power-limiter mod to 64 watts instead of 38.

@antantti, yes two of the gigabyte cards are in my daily win 10: 32 bit machine, running release 66 of ccminer.  the rest are running under kopiemtu (Lubuntu 14.10 - 64 bit) USB install.

[antantti]

yes, all of my gigabyte windforce models have the power-limiter mod to 64 watts instead of 38.

@antantti, yes two of the gigabyte cards are in my daily win 10: 32 bit machine, running release 66 of ccminer.  the rest are running under kopiemtu (Lubuntu 14.10 - 64 bit) USB install.

Ok, you running 32-bit operating system. There must be a good reason for that?

[hammer24p]

can you flash 750 non ti to 750 ti and what do you change

[hashbrown9000]

only reason for win 10- 32 bit was that i already had win 7- 32 bit and the upgrade to 10 was free. the 3GB RAM limitation is really killing me though. firefox and even chrome are such memory hogs.

as for flashing from non-ti to ti, i'm not sure you could do that.  i think non-ti has 512 cuda cores, and ti has 640 cores.

[CoffeeCat]

What clocks would you recommend for 750 ti and a 980? I thought I remember someone here saying that memory clocks didn't make that much of a difference. Thanks.

You need to find out yourself. But I can give you a hint. The lyra2v2 can do 5.2MHASH@750ti with the sp-mod opensource kernal. But it won't do it with a gpu core of 1500MHASH, and intensity to the max. To mine with a profit you need to know your cards well. Changing the sourcecode in the kernals is not enough.

The opensource kernals kan perform 15-20% bether with a right config and overclocks.

Um... thanks?

[bathrobehero]

I'm preaching to the choir once again but there are no magical 38 watt 750 cards exist, period. Even the non-Ti 750 cards can pull up to 60 watts with the stock BIOS regardless of having a 6 pin or not. That of course doesn't mean they will use 60 watt with every algo but they will use close to that much with more power hungry algos  (eg. groestl).
Look into their BIOS' or any reliable reviews/stress tests or better yet, look at the wall with a watt meter.

There are no huge efficiency gaps (hash per watt) within Maxwell cards.

[scryptr]

can you flash 750 non ti to 750 ti and what do you change

NO YOU CAN'T--

The 750 has 512 CUDA cores and the 750ti has 640 CUDA cores.  The number of CUDA cores is the basis for the processing capacity.  It is a significant physical difference.       --scryptr

[chrysophylax]

anyway - as for the overclocking side ... i think i will take wolfs advice and take another look into it next week week ... the next 3days for me is incredibly busy ...

Linux vs Windows, cuda 6.5 vs 7.5, x86 vs x86_64. Holy crap hasn't this been hashed and rehashed over and over again?

From my experience mining on Windows or Linux produces the same hash rates everything else being equal. However,
I have noticed that Windows hashing slows more due to other activity, in particular logging in remotely with ssh.
I haven't tried cuda 7.5 on either OS but the hashrates you reported on cuda 7.5 is significantly slower that I get with 6.5
on Linux or Windows. I conclude 7.5 is equally slow on both OS's.

32 vs 64 bit is not about the OS but about how ccminer is compiled. Compiling for 32 bit tends to be a little more efficient
when dealing with mostly 32 bit arithmetic.

Overclocking on Linux has been discussed at length in this thread, including getting it to work headless. Search for
"cooolbits" for everything you need to know.

id like to say that i totally agree with you joblo - but i dont ... as the systems i have running are showing the complete opposite of what you are stating ...

coolbits and oc are NOT working as they should on the gigabyte 750ti oc lp cards i have running ... investigation into this lead me to conclude that ( and i did read this on the nvidia site ) that the support is not currently implemented in the drivers ( cuda and video drivers ) as of yet - and is only available for the higher end cards ...

the cuda 7.5 implementation i have compiling ccminer-spmod IS causing hashing as fast ( and slightly faster ) than cuda 6.5 compiles ...

im not trying to be obnoxious - its a fact ...

sp has these results of slower hashrate using c7.5 as you do - i dont ...

go figure ...

#crysx

I know coolbits doesn't work. Flash your GPUs.

I'm not willing to go as far as flashing FW but coolbits worked for me.

Crysx, I don't think you are being obnoxious, just maybe a little stubborn. Your results are counter to others in
two ways that are contradictory. First you say Cuda 7.5 is faster than 6.5, while it's the reverse for everyone else.
But then your 7.5 hash rate (2800) is lower than others' 6.5 hash rate (3100). Putting both of those facts together
and your 6.5 hash rate would be even lower than 2800.

In order to accept your facts I would have to believe three things:

1. your particular 750ti's can't perform to the same level as any others,
2. for some unique reason you don't suffer the 10% loss in performance between 6.5 and 7.5 that anyone else experiences,
3. it's just a coincidence that your hash rates are lower by the same amount of degradation expected between 6.5 and 7.5.

If it was happening to me and potentially costing me 300 KH per card I'd want to get to the bottom of it.

that is what im saying joblo ...

its not contradictory at all ...

my cards did just below 2800KH on c6.5 compile on f20x64 - now they do a little over 2800KH on c7.5 compiled on f22x64 ... these cards NEVER did 3100KH on ANY of my systems ... period ...

so contradictory to OTHERS or not - MY systems are doing this ... thats it ...

i can allow YOU access to prove it mate - as i trust YOU ( and only a very few others on here ) if you would like to try and compile for yourself in MY systems ... i can allocate one machine and you can have a go ... THEN tell me how contradictory it is when you see the SAME results as i do ...

my offer is open mate ...

#crysx

[joblo]

In order to accept your facts I would have to believe three things:

1. your particular 750ti's can't perform to the same level as any others,
2. for some unique reason you don't suffer the 10% loss in performance between 6.5 and 7.5 that anyone else experiences,
3. it's just a coincidence that your hash rates are lower by the same amount of degradation expected between 6.5 and 7.5.

If it was happening to me and potentially costing me 300 KH per card I'd want to get to the bottom of it.

that is what im saying joblo ...

its not contradictory at all ...

my cards did just below 2800KH on c6.5 compile on f20x64 - now they do a little over 2800KH on c7.5 compiled on f22x64 ... these cards NEVER did 3100KH on ANY of my systems ... period ...

so contradictory to OTHERS or not - MY systems are doing this ... thats it ...

i can allow YOU access to prove it mate - as i trust YOU ( and only a very few others on here ) if you would like to try and compile for yourself in MY systems ... i can allocate one machine and you can have a go ... THEN tell me how contradictory it is when you see the SAME results as i do ...

my offer is open mate ...

#crysx

Thanks for the offer but I wouldn't accept it unless it would help solve the dilema, not just to prove it. Also you're
getting the expected performance of cuda 7.5 on a 750ti so there may be little for you to gain by pursuing this issue.

I looked at the clock specs for the GB LP card and they are about 10% lower than my EVGAs so that can account
for your lower baseline.

With that variable eliminated it leads me to my real pet peeve, the lower performance with cuda 7.5 vs 6.5. It still
baffles me that Nvidia would do this with apparently no effort to either fix it, provide a compatibility mode, or continue
to support 6.5.

I think I need to do some testing of cuda 7.5, maybre things are improving.

EDIT:

I have tested cuda 6.5 vs 7.5 on Centos 6.7, driver 352.63, with 2 EVGA GTX 750ti SC standard clocks, mining x11.
Each card produces different hash rates:

GPU0: 6.5 3085
           7.5 3015

GPU 1: 6.5 3155
            7.5 3090

Cuda 7.5 deficit is still about 70k, about 2%.