CCminer(SP-MOD) Modded GPU kernels.

[Cryptozillah]

I got some 750Ti cards going last night. First time mining Nvidia ever so i feel a bit like a noob.

I have some trouble to get my system to detect the gpu:s.
With AMD i usually install my gpu driver and then add one card at the time without problems.

Does the Nvidia drivers work the same way these days ? Or do i have to have all cards installed in the rig before installing the driver ?
My system had problems finding more then one or two gpu:s when adding more cards to the rig.

I will continue working with the problem after work today.

Anyway. The 750Ti:s i have running now are mining at 3.6Mhs @ X11. I will tweak it a bit more when i get more cards added in the rigs.
I am really happy with the speed so far !

[1714697337]

1714697337

[1714697337]

1714697337

[1714697337]

1714697337

[1714697337]

1714697337

[rednoW]

nvidia is famous for its non-problem drivers )) If it is not some strategy, then x11 mining is not a best for today. Quark and lyra2v2 are more profitable (6.5 mhs and 4.5 mhs on 750ti).

[sp_]

I have corrected the hashrate bug in quark. and build a new exe. The version number is still 61

[crsminer]

--ethminer and 750ti on windows--

In a rig with one 970 and one 750ti:

Loading only the 970 with -t 1:
- when running -M (benchmark mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;
- when running without -M (mining mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;

Loading 970+750ti with -t 2:
- when running -M, memory controller load for both cards is at 100%, rig is hashing at 27.2Mh/s thus giving 8.8Mh/s for the 750ti;
- when running without -M, memory controller load is at 100% for 970 card and 11%! for 750ti, and rig is hashing at 19.4Mh/s, thus giving only 0.968Mh/s for 750ti (exactly 11% from 100% memory controller load).

I have to say that I have no coding skills, I just noticed this while playing with the switches and looking at Afterburner & GPU-Z readings.

Maybe some of the great coders from here can make something out of this so we can hash at full speed with the 750ti on windows.


EDIT: when adding another 960 memory controller load (for 750ti) drops one bit to 10%, with two 960 stays at 10% and when adding the third 960 (with -t 5), memory controller load for the 750ti drops to 5%.
 

[scryptr]

--ethminer and 750ti on windows--

In a rig with one 970 and one 750ti:

Loading only the 970 with -t 1:
- when running -M (benchmark mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;
- when running without -M (mining mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;

Loading 970+750ti with -t 2:
- when running -M, memory controller load for both cards is at 100%, rig is hashing at 27.2Mh/s thus giving 8.8Mh/s for the 750ti;
- when running without -M, memory controller load is at 100% for 970 card and 11%! for 750ti, and rig is hashing at 19.4Mh/s, thus giving only 0.968Mh/s for 750ti (exactly 11% from 100% memory controller load).

I have to say that I have no coding skills, I just noticed this while playing with the switches and looking at Afterburner & GPU-Z readings.

Maybe some of the great coders from here can make something out of this so we can hash at full speed with the 750ti on windows.

WINDOWS BUG--

This is a known Windows bug in Ethminer, and Genoil (the dev) knows about it.  In Linux, the 750ti mines at 8+ Mh/s.  I am all for collaboration, I'd gain 40-50 Mh/s if it were fixed.  If you are running the Cuda version of Ethminer, use the "-U --gpu-devices 0 1" string in the command line.  Does "-t  2" do the same thing?  I thought it was for OpenCL.

--scryptr

[crsminer]

--ethminer and 750ti on windows--

In a rig with one 970 and one 750ti:

Loading only the 970 with -t 1:
- when running -M (benchmark mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;
- when running without -M (mining mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;

Loading 970+750ti with -t 2:
- when running -M, memory controller load for both cards is at 100%, rig is hashing at 27.2Mh/s thus giving 8.8Mh/s for the 750ti;
- when running without -M, memory controller load is at 100% for 970 card and 11%! for 750ti, and rig is hashing at 19.4Mh/s, thus giving only 0.968Mh/s for 750ti (exactly 11% from 100% memory controller load).

I have to say that I have no coding skills, I just noticed this while playing with the switches and looking at Afterburner & GPU-Z readings.

Maybe some of the great coders from here can make something out of this so we can hash at full speed with the 750ti on windows.

WINDOWS BUG--

This is a known Windows bug in Ethminer, and Genoil (the dev) knows about it.  In Linux, the 750ti mines at 8+ Mh/s.  I am all for collaboration, I'd gain 40-50 Mh/s if it were fixed.  If you are running the Cuda version of Ethminer, use the "-U --gpu-devices 0 1" string in the command line.  Does "-t  2" do the same thing?  I thought it was for OpenCL.

--scryptr

Yes, -t <n> does the same thing as --gpu-devices 0 1 ... when running -U. The only difference is that you can't choose exactly what cards to use with -t as with --gpu-devices.

About the 750ti windows bug, have you noticed also that when using benchmark mode (-M) the 750ti hashes at full speed (memory controller at 100%)? It is very odd...

[djm34]

--ethminer and 750ti on windows--

In a rig with one 970 and one 750ti:

Loading only the 970 with -t 1:
- when running -M (benchmark mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;
- when running without -M (mining mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;

Loading 970+750ti with -t 2:
- when running -M, memory controller load for both cards is at 100%, rig is hashing at 27.2Mh/s thus giving 8.8Mh/s for the 750ti;
- when running without -M, memory controller load is at 100% for 970 card and 11%! for 750ti, and rig is hashing at 19.4Mh/s, thus giving only 0.968Mh/s for 750ti (exactly 11% from 100% memory controller load).

I have to say that I have no coding skills, I just noticed this while playing with the switches and looking at Afterburner & GPU-Z readings.

Maybe some of the great coders from here can make something out of this so we can hash at full speed with the 750ti on windows.

WINDOWS BUG--

This is a known Windows bug in Ethminer, and Genoil (the dev) knows about it.  In Linux, the 750ti mines at 8+ Mh/s.  I am all for collaboration, I'd gain 40-50 Mh/s if it were fixed.  If you are running the Cuda version of Ethminer, use the "-U --gpu-devices 0 1" string in the command line.  Does "-t  2" do the same thing?  I thought it was for OpenCL.

--scryptr

Yes, -t <n> does the same thing as --gpu-devices 0 1 ... when running -U. The only difference is that you can't choose exactly what cards to use with -t as with --gpu-devices.

About the 750ti windows bug, have you noticed also that when using benchmark mode (-M) the 750ti hashes at full speed (memory controller at 100%)? It is very odd...

the dag filewhich is used for benchmark is less random that's the reason why it performs better

[scryptr]

100% MCU--

Honestly, I haven't bothered to run in benchmark mode.  What I notice, is that my GTX 960 (card 0) mines with the MCU at 100% and all of the 750ti cards mine with the MCU at 10%-20%.  Ten to twenty percent of 8.8 Mh/s is about 1-2 Mh/s.

A 750ti under Windows mines at about 1-2Mh/s with Ethminer.  Ethminer does not report individual card rates like CudaMiner or CCminer.  It reports a total hashing rate.  How is the hash distributed across the rig?  Is this a clue?

Under Linux the 750ti is reported to mine at 8+ Mh/s.  The operating environment may be handling instructions differently.

--scryptr

EDIT:  Genoil says that the 750ti will run at the full hash rate under Win 7.       --scryptr

[lawrencelyl]

I have corrected the hashrate bug in quark. and build a new exe. The version number is still 61

Do I have to compile it my own or grab the exe from the github's latest release?

[crsminer]

--ethminer and 750ti on windows--

In a rig with one 970 and one 750ti:

Loading only the 970 with -t 1:
- when running -M (benchmark mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;
- when running without -M (mining mode), memory controller load is at 100%, rig is hashing at 18.4Mh/s;

Loading 970+750ti with -t 2:
- when running -M, memory controller load for both cards is at 100%, rig is hashing at 27.2Mh/s thus giving 8.8Mh/s for the 750ti;
- when running without -M, memory controller load is at 100% for 970 card and 11%! for 750ti, and rig is hashing at 19.4Mh/s, thus giving only 0.968Mh/s for 750ti (exactly 11% from 100% memory controller load).

I have to say that I have no coding skills, I just noticed this while playing with the switches and looking at Afterburner & GPU-Z readings.

Maybe some of the great coders from here can make something out of this so we can hash at full speed with the 750ti on windows.

WINDOWS BUG--

This is a known Windows bug in Ethminer, and Genoil (the dev) knows about it.  In Linux, the 750ti mines at 8+ Mh/s.  I am all for collaboration, I'd gain 40-50 Mh/s if it were fixed.  If you are running the Cuda version of Ethminer, use the "-U --gpu-devices 0 1" string in the command line.  Does "-t  2" do the same thing?  I thought it was for OpenCL.

--scryptr

Yes, -t <n> does the same thing as --gpu-devices 0 1 ... when running -U. The only difference is that you can't choose exactly what cards to use with -t as with --gpu-devices.

About the 750ti windows bug, have you noticed also that when using benchmark mode (-M) the 750ti hashes at full speed (memory controller at 100%)? It is very odd...

the dag filewhich is used for benchmark is less random that's the reason why it performs better

More findings when running only 750ti's:

Running only one 750ti:
- benchmark mode, MCU is at 100% giving expected 8.8Mh/s
- mining mode, MCU is at 63-65% giving 5.1Mh/s

Running two 750ti:
- benchmark mode, MCU is at 100% for both cards, giving 17.4Mh/s
- mining mode, MCU is at 56-63% for card 0 and 3% for card 1, giving 5.1-5.7Mh/s total.

So it seems that 750ti can do at least 5Mh/s on mining mode but only when it's the single card running.


EDIT: weird thing is that when I add the third card (also 750ti) in benechmark mode I only get 10.5Mh/s

[scryptr]

I have corrected the hashrate bug in quark. and build a new exe. The version number is still 61

Do I have to compile it my own or grab the exe from the github's latest release?

If you are running Windows, you can download the new zip package under the releases tab.  He built the executable file, and reposted Release dot 61.       --scryptr

[lawrencelyl]

I have corrected the hashrate bug in quark. and build a new exe. The version number is still 61

Do I have to compile it my own or grab the exe from the github's latest release?

If you are running Windows, you can download the new zip package under the releases tab.  He built the executable file, and reposted Release dot 61.       --scryptr

Thanks, trying it out now

[scryptr]

RE-RELEASE dot 61--

After mining on NiceHash for about 90 minutes, the re-release of dot 61 apparently performs with a slightly higher hash rate and a correct poolside report of the hash rate.  My six-card 750ti FTW rig now hashes Quark at 39.5Mh/s, whereas before it was hashing at about 38.8 Mh/s.  Thank you for the hard work!       --scryptr 

[Cryptozillah]

RE-RELEASE dot 61--

After mining on NiceHash for about 90 minutes, the re-release of dot 61 apparently performs with a slightly higher hash rate and a correct poolside report of the hash rate.  My six-card 750ti FTW rig now hashes Quark at 39.5Mh/s, whereas before it was hashing at about 38.8 Mh/s.  Thank you for the hard work!       --scryptr 

What driver are you using for your 750Ti:s ?
I am trying to get five cards running on my win8 systems but i only manage to get four of them going at the same time.

[scryptr]

RE-RELEASE dot 61--

After mining on NiceHash for about 90 minutes, the re-release of dot 61 apparently performs with a slightly higher hash rate and a correct poolside report of the hash rate.  My six-card 750ti FTW rig now hashes Quark at 39.5Mh/s, whereas before it was hashing at about 38.8 Mh/s.  Thank you for the hard work!       --scryptr  

What driver are you using for your 750Ti:s ?
I am trying to get five cards running on my win8 systems but i only manage to get four of them going at the same time.

LINUX DRIVERS--

I am running Linux for this rig.  My driver is older, 340.x, can't remember.  Hash rate might be higher with the latest driver...       --scryptr

EDIT:  Windows?  I have one rig on Win 8 with the very latest 355.60 drivers.  Windows had no problem recognizing all of my cards on installation, that was several driver versions ago.  I have 6 cards on the rig.  If you are running CCminer, use the "--devices 0,1,2,3,4" command line switch.  See the CCminer help file; if you run "ccminer -h >> help.txt" you get a copy of the help file that you can open in notepad.

One other thing!  Is your power supply big enough?       --scryptr

EDIT:  One other thing, EVGA PrecisionX 16 will only allow you to overclock 4 cards on Windows.  Look for nVidia Inspector, an overclocking utility that will allow you to monitor/adjust 6 cards.  Should pop up in a Google search.

--scryptr

[zTheWolfz]

EDIT: ccminer-lyra, the djm34 fork, should work with older cards but, afaik, only compute 3.5 and up.

Ya works with quark but no luck getting to run lyra2rev2 on the 760.
I'm working quark on the 660 as well with it.
Heck one 660 or 760 out does 2x R270's time and half with the latest sgminer for those.
I have 4xR270. and a few 5850 and 6870 setting doing nothing. Pair of GTX480 collecting dust on the shelf.

@ --scryptr I've already looked around and already have the most updated ones I could find. downloaded, thanks.

it wouldn't be too difficult to get it to work with compute 3.0 and lower, just need to get rid of shuffle instruction and ldg ones...

Good to hear I'll keep a watch out for your post of when that happens and will upgrade accordingly.
Thanks!

[zTheWolfz]

EDIT: ccminer-lyra, the djm34 fork, should work with older cards but, afaik, only compute 3.5 and up.

Ya works with quark but no luck getting to run lyra2rev2 on the 760.
I'm working quark on the 660 as well with it.
Heck one 660 or 760 out does 2x R270's time and half with the latest sgminer for those.
I have 4xR270. and a few 5850 and 6870 setting doing nothing. Pair of GTX480 collecting dust on the shelf.

@ --scryptr I've already looked around and already have the most updated ones I could find. downloaded, thanks.

In my fork Lyra2REv2 works with compute 3.0 cards, but it's not speed optimized:

If you want to try it:
https://dl.dropboxusercontent.com/u/34410241/ccminer.exe

Testing with your version, ran into a few config problems, I have to use lyra2v2 on instead of lyra2rev2.
Will there be problems if running that at server lyra2rev2.usa.nicehash.com:3347
seen a few boo's right off, but also some accepts?

Edit: Maybe a post of your config line flags would help. This is what I'm using to get it to work for me at the moment.

ccminer.exe -a lyra2v2 -o stratum+tcp://lyra2rev2.usa.nicehash.com:3347 -u 18tvS3deKZK5q4eTtPRWYeEMWmGmuErHgz -p x -d 1 --api-bind 0 --cpu-priority 2 -i 19

Edit2: Dropping -19.4 to -i 19 seems to have fix any boo's.
1237 kh/s on stock clocks with the 760... thanks for posting...!

[Cryptozillah]

RE-RELEASE dot 61--

After mining on NiceHash for about 90 minutes, the re-release of dot 61 apparently performs with a slightly higher hash rate and a correct poolside report of the hash rate.  My six-card 750ti FTW rig now hashes Quark at 39.5Mh/s, whereas before it was hashing at about 38.8 Mh/s.  Thank you for the hard work!       --scryptr  

What driver are you using for your 750Ti:s ?
I am trying to get five cards running on my win8 systems but i only manage to get four of them going at the same time.

LINUX DRIVERS--

I am running Linux for this rig.  My driver is older, 340.x, can't remember.  Hash rate might be higher with the latest driver...       --scryptr

EDIT:  Windows?  I have one rig on Win 8 with the very latest 355.60 drivers.  Windows had no problem recognizing all of my cards on installation, that was several driver versions ago.  I have 6 cards on the rig.  If you are running CCminer, use the "--devices 0,1,2,3,4" command line switch.  See the CCminer help file; if you run "ccminer -h >> help.txt" you get a copy of the help file that you can open in notepad.

One other thing!  Is your power supply big enough?       --scryptr

EDIT:  One other thing, EVGA PrecisionX 16 will only allow you to overclock 4 cards on Windows.  Look for nVidia Inspector, an overclocking utility that will allow you to monitor/adjust 6 cards.  Should pop up in a Google search.

--scryptr

I run win 8.1 on a couple of rigs. I really think it is a great os for mining, great speeds.
I use the latest driver from Nvidia.
I never had any issues putting five AMD gpu:s in these rigs before but this was a real hazzle to get working.
Now one rig is mining with five gpus and the other one with four of five gpu:s. I will give that rig another go tomorrow.

I had to move cards around in loads of configurations to finally get five cards found in windows. I spent like five hours straight this evening to get them up and running.
Now they are running smooth and cool anyway

I have Corsair RM850 psu:s in these 750Ti rigs. It should be enough, the rigs are pulling like 320w each @ X11

I use Afterburner to overclock the gpu:s, i think it works pretty good. I will check out Inspector tomorrow, thanks for the tip.

[djm34]

EDIT: ccminer-lyra, the djm34 fork, should work with older cards but, afaik, only compute 3.5 and up.

Ya works with quark but no luck getting to run lyra2rev2 on the 760.
I'm working quark on the 660 as well with it.
Heck one 660 or 760 out does 2x R270's time and half with the latest sgminer for those.
I have 4xR270. and a few 5850 and 6870 setting doing nothing. Pair of GTX480 collecting dust on the shelf.

@ --scryptr I've already looked around and already have the most updated ones I could find. downloaded, thanks.

In my fork Lyra2REv2 works with compute 3.0 cards, but it's not speed optimized:

If you want to try it:
https://dl.dropboxusercontent.com/u/34410241/ccminer.exe

Testing with your version, ran into a few config problems, I have to use lyra2v2 on instead of lyra2rev2.

it never been lira2rev2 (except for sgminer)

[sambiohazard]

Today's share sent

djm34: 374808cfd3135fc1e9b302a0c6480ca79e7f8cc7ed81169addc01d1d75f38567
pallas: e42ef59736a03b87dfd1946de7911d4d5bf6f36e3cfdabb180a1733579aa9589
sp_: 66c08edeccebc6f028801dd6aafe6de761dc7fbc1911cbce42917052533015b0

This will be last donation for now. I will keep sending more whenever possible. Thanks to all the devs for awesome work!!