CCminer(SP-MOD) Modded GPU kernels.

[bensam1231]

Down and down everything goes, this is the worst I've seen algos in a long time...

Thats why you need to mine algos that aren't supported by nicehash / multipools..

I have continued to mod pentablake now. 90MHASH on the gtx970..

the opensource does 15MHASH on the 750ti.

So with my modded kernal 1 gtx970 is producing the same as 6 750ti's running the opensource kernal.

@200 MHASH i get around 82 Joincoins/day (0.02BTC / day)

Pentablake is X5. (5 rounds of (blake-512 16 rounds))

Most coins not supported by Multipools or Nicehash have really low market caps and difficulty. The moment you hop on them they get squashed. Looks like there is a total of 1.5BTC buys for Joincoin.

[coinut]

Sp
 
what hashrate on your Pentablake mod for the 750 ti?

is the miner available?

edit

all good noticed the info above 

[sp_]

Most coins not supported by Multipools or Nicehash have really low market caps and difficulty. The moment you hop on them they get squashed. Looks like there is a total of 1.5BTC buys for Joincoin.

5,5BTC worth of joincoins are mined everyday. The 1.5BTC support seems to hold. The coin will go POS soon, and people are investing.

[ZenFr]

since i was curious about the evolution of the miner, i've performed some benchmarks on a 960 with various popular building with both cuda 6.5 and 7.5...

Code:

cuda 6.5.19

version    x11       x13   neoscrypt   qubit     quark    lyra2v2
-----------------------------------------------------------------
1.5.61   4896.75   3817.49   307.50   7370.82   9901.39   4975.50
1.5.62   4830.94   3824.48   308.17   7367.76   9832.79   4892.54
1.5.63   4829.33   3827.89   307.98   7329.33   9942.74   5064.79
1.5.64   4830.37   3828.82   305.08   7364.21   9939.97   5022.05
1.5.65   4833.80   3829.02   308.03   7316.46   9935.27   5128.57
1.5.66   4841.17   3828.98   308.19   7309.84   9945.59   4964.04
1.5.67   4843.54   3842.94   304.59   7307.92   9895.11   5311.58
1.5.68   4871.72   3849.88   304.81   7333.10   9897.92   5284.52
1.5.69   4871.32   3846.09   305.23   7316.51   9926.38   5291.48
1.5.70   4850.56   3840.83   300.97   7325.78   9928.29   5274.59
1.5.71   4864.87   3848.43   302.00   7311.15   9938.50   5434.95
HEAD     4859.92   3846.95   279.25   7309.49   9943.24   5442.70


cuda 7.5.18

version    x11       x13   neoscrypt   qubit     quark    lyra2v2
-----------------------------------------------------------------
1.5.61   4545.64   3557.33    na      6833.08   9649.00   5137.34
1.5.62   4548.16   3536.73    na      6765.00   9741.35   5082.66
1.5.63   4549.85   3552.70    na      6766.62   9595.26   5254.05
1.5.64   4469.91   3544.88    na      6766.10   9772.69   5268.54
1.5.65   4537.04   3558.38    na      6760.28   9845.07   5329.07
1.5.66   4379.26   3547.62    na      6816.86   9779.27   5404.03
1.5.67   4572.08   3575.58    na      6764.94   9649.66   5782.35
1.5.68   4568.98   3576.28    na      6763.20  10016.10   5752.82
1.5.69   4564.96   3576.59    na      6760.18   9971.33   5747.59
1.5.70   4566.48   3564.19    na      6775.10   9936.68   5737.31
1.5.71   4529.04   3568.78    na      6756.01   9961.33   5505.06
HEAD     4564.34   3580.72    na      6748.33  10013.80   5506.37

edit: this is under linux

I am OK with these results (I work witk quark and lyra2rev2, Ubuntu and standard kernel/clock, for lower TdP) :
 Cuda 7.5 is faster rhan Cuda 6.5
 release 1.5.71/70/69 are lower than the previous 1.5.68 (the faster release for me).

Thanks for the table, skunk :-).
 

[sp_]

I am OK with these results (I work witk quark and lyra2rev2, Ubuntu and standard kernel/clock, for lower TdP) :
 Cuda 7.5 is faster rhan Cuda 6.5
 release 1.5.71/70/69 are lower than the previous 1.5.68 (the faster release for me).
Thanks for the table, skunk :-).
 

What hashrates do you get? You can get 10% more coins by using x86 cuda6.5 and windows.

[pallas]

since i was curious about the evolution of the miner, i've performed some benchmarks on a 960 with various popular building with both cuda 6.5 and 7.5...

Code:

cuda 6.5.19

version    x11       x13   neoscrypt   qubit     quark    lyra2v2
-----------------------------------------------------------------
1.5.61   4896.75   3817.49   307.50   7370.82   9901.39   4975.50
1.5.62   4830.94   3824.48   308.17   7367.76   9832.79   4892.54
1.5.63   4829.33   3827.89   307.98   7329.33   9942.74   5064.79
1.5.64   4830.37   3828.82   305.08   7364.21   9939.97   5022.05
1.5.65   4833.80   3829.02   308.03   7316.46   9935.27   5128.57
1.5.66   4841.17   3828.98   308.19   7309.84   9945.59   4964.04
1.5.67   4843.54   3842.94   304.59   7307.92   9895.11   5311.58
1.5.68   4871.72   3849.88   304.81   7333.10   9897.92   5284.52
1.5.69   4871.32   3846.09   305.23   7316.51   9926.38   5291.48
1.5.70   4850.56   3840.83   300.97   7325.78   9928.29   5274.59
1.5.71   4864.87   3848.43   302.00   7311.15   9938.50   5434.95
HEAD     4859.92   3846.95   279.25   7309.49   9943.24   5442.70


cuda 7.5.18

version    x11       x13   neoscrypt   qubit     quark    lyra2v2
-----------------------------------------------------------------
1.5.61   4545.64   3557.33    na      6833.08   9649.00   5137.34
1.5.62   4548.16   3536.73    na      6765.00   9741.35   5082.66
1.5.63   4549.85   3552.70    na      6766.62   9595.26   5254.05
1.5.64   4469.91   3544.88    na      6766.10   9772.69   5268.54
1.5.65   4537.04   3558.38    na      6760.28   9845.07   5329.07
1.5.66   4379.26   3547.62    na      6816.86   9779.27   5404.03
1.5.67   4572.08   3575.58    na      6764.94   9649.66   5782.35
1.5.68   4568.98   3576.28    na      6763.20  10016.10   5752.82
1.5.69   4564.96   3576.59    na      6760.18   9971.33   5747.59
1.5.70   4566.48   3564.19    na      6775.10   9936.68   5737.31
1.5.71   4529.04   3568.78    na      6756.01   9961.33   5505.06
HEAD     4564.34   3580.72    na      6748.33  10013.80   5506.37

edit: this is under linux

So neoscrypt isn't working on cuda 7.5? Last time I checked it was fine, just 40% slower :-)

[ZenFr]

I am OK with these results (I work witk quark and lyra2rev2, Ubuntu and standard kernel/clock, for lower TdP) :
 Cuda 7.5 is faster rhan Cuda 6.5
 release 1.5.71/70/69 are lower than the previous 1.5.68 (the faster release for me).
Thanks for the table, skunk :-).
 

What hashrates do you get? You can get 10% more coins by using x86 cuda6.5 and windows.

No windows for me, sorry :-).

[sp_]

I am OK with these results (I work witk quark and lyra2rev2, Ubuntu and standard kernel/clock, for lower TdP) :
 Cuda 7.5 is faster rhan Cuda 6.5
 release 1.5.71/70/69 are lower than the previous 1.5.68 (the faster release for me).
Thanks for the table, skunk :-).
 

What hashrates do you get? You can get 10% more coins by using x86 cuda6.5 and windows.

No windows for me, sorry :-).

So the conclusion is:

1. Don't buy the gtx 960 for mining.(100% more power. but only 10% more hashgain than the 750ti in the lyra2v2 algo)
2. Don't use linux. (10% loss)
3. Don't use cuda 7.5(40% loss in the hashrate (neoscrypt)
4. Don't use 64 bit builds. (10% loss)

My modded kernals are made for windows, x86 and cuda 6.5 and they are always the fastest.

[sp_]

@sp_, I mined Trinity at coinspool last night at around 1.13 GH for about two hours, but it didn't seem like the pool was going to credit me any coins even though i was getting accepts in the miner.  Only diff .0039xxxx was working for me to avoid "above target share" errors.  Should I just leave all my gigahashes there and see what happens?  b/t/w I was using commit number 1124.

The gui of the pool is broken, but you should have coins in your wallet.

[NiceHashSupport]

I am OK with these results (I work witk quark and lyra2rev2, Ubuntu and standard kernel/clock, for lower TdP) :
 Cuda 7.5 is faster rhan Cuda 6.5
 release 1.5.71/70/69 are lower than the previous 1.5.68 (the faster release for me).
Thanks for the table, skunk :-).
 

What hashrates do you get? You can get 10% more coins by using x86 cuda6.5 and windows.

No windows for me, sorry :-).

So the conclusion is:

1. Don't buy the gtx 960 for mining.(100% more power. but only 10% more hashgain than the 750ti in the lyra2v2 algo)
2. Don't use linux. (10% loss)
3. Don't use cuda 7.5(40% loss in the hashrate (neoscrypt)
4. Don't use 64 bit builds. (10% loss)

My modded kernals are made for windows, x86 and cuda 6.5 and they are always the fastest.

And additionally to this, use NiceHash Miner so you always mine most profitable algorithm.

[ZenFr]

I am OK with these results (I work witk quark and lyra2rev2, Ubuntu and standard kernel/clock, for lower TdP) :
 Cuda 7.5 is faster rhan Cuda 6.5
 release 1.5.71/70/69 are lower than the previous 1.5.68 (the faster release for me).
Thanks for the table, skunk :-).
 

What hashrates do you get? You can get 10% more coins by using x86 cuda6.5 and windows.

No windows for me, sorry :-).

So the conclusion is:

1. Don't buy the gtx 960 for mining.(100% more power. but only 10% more hashgain than the 750ti in the lyra2v2 algo)
2. Don't use linux. (10% loss)
3. Don't use cuda 7.5(40% loss in the hashrate (neoscrypt)
4. Don't use 64 bit builds. (10% loss)

My modded kernals are made for windows, x86 and cuda 6.5 and they are always the fastest.

That's your conclusion.
And I'm not sure that you are honest in your calculation : windows is not free, your modded kernals are not free and the extra electricity consumption (for more TdP) is not free : what is the best now ?

[sp_]

That's your conclusion.
And I'm not sure that you are honest in your calculation : windows is not free, your modded kernals are not free and the extra electricity consumption (for more TdP) is not free : what is the best now ?

Most of my modded kernals are opensource and free. Take the quark kernal f.ex.. The latest linux optimized version from tvpruvot (1.6.6) is producing 21MHASH on a gtx980ti (cuda 7.5).

My opensource mod is doing 28.5MHASH on the same card and clock on windows. This is a 35% free gain.

10% loss cuda 7.5
10% loss linux
10% loss 64bit
5% loss Made in France

[ZenFr]

5% loss Made in France

?

[skunk]

What card is this (memclock/coreclock)? And how long did you run the benchmark

it's a gigabyte on default clocks (3505/1278)
each test ran for one minute...
since profits are very low, i'll maybe repeat the tests today on a 5 minutes timeframe while hashing on nicehash

You get bether rates under windows x86 builds. Asus strix 960 should do above 6MHASH on lyra2v2 and 10.7 in quark. (cuda 6.5)

no windows here...

For more accurate benchmarking you should use the --timelimit option.
ccminer -a quark --benchmark --timelimit 500  (will mine for 500 seconds and then exit)

i'm using the api for getting the hash rate just before killing the process as it's more convenient than parsing the output

If your card trottle, (hashrate is decreasing over time) The gpu-clock is downlocked automaticly and hashrate is lost. To avoid trottling reduce the intenstiy, increase the tdp in the bios, increase the temp limits of the card. (overclock) make sure that you always run in p0 mode.

no, it doesn't throttle as the hashrate always increase after starting and then stays constant without decreases

Interesting that my junior asus strix 750ti with +200 on the mem-clock/gpu-clock is almost running faster than your gtx 960 in lyra. I get 5.4MHASH.
(640 shaders vs 1024)

yes, the card performance is disappointing but was cheap...

[pallas]

5% loss Made in France

You could have avoided that... tpruvot did so much (and is still doing) to cpuminer and ccminer!

[sp_]

5% loss Made in France

You could have avoided that... tpruvot did so much (and is still doing) to cpuminer and ccminer!

I see no real optimalizations. All he does is to copy from others.

[myagui]

5% loss Made in France

You could have avoided that... tpruvot did so much (and is still doing) to cpuminer and ccminer!

I see no real optimalizations. All he does is to copy from others.

As someone using from both forks at different times, it's been funny to experience that plenty of times, the made in France is actually faster as reported poolside, for the same compilation/execution environment, and as long as launched with the same intensity (I hear that in France, they are more conservative about default intensities)...

Tpruvot has done - a F^#@#%ng lot - more than copy from others.
From extending miner functionality (api stuff, backup pool support, conditional mining) to adding new algorithms (some public, some private), he's done quite a lot for this community. No wonder that it gets hard to have any meaningful collaboration going between developers.  

Edit: There will be no followup from me, I'm done with the pissing contest.

[sp_]

5% loss Made in France

You could have avoided that... tpruvot did so much (and is still doing) to cpuminer and ccminer!

I see no real optimalizations. All he does is to copy from others.

As someone using from both forks at different times, it's been funny to experience that plenty of times, the made in France is actually faster as reported poolside, for the same compilation/execution environment, and as long as launched with the same intensity (I hear that in France, they are more conservative about default intensities)...
Tpruvot has done - a F^#@#%ng lot - more than copy from others.
From extending miner functionality (api stuff, backup pool support, conditional mining) to adding new algorithms (some public, some private), he's done quite a lot for this community. No wonder that it gets hard to have any meaningful collaboration going between developers.  
Edit: There will be no followup from me, I'm done with the pissing contest.

I still don't understand why people are using a miner that is 30% slower in most algos.
The purpose of a miner is to MINE. There are no friends in crypto. Only enemies. Less developers is bether, because then my private kernals are more profitable.
Greed is driving you.
That's why you are here..

Go get those COINS and stop complaining.

[pallas]

5% loss Made in France

You could have avoided that... tpruvot did so much (and is still doing) to cpuminer and ccminer!

I see no real optimalizations. All he does is to copy from others.

That's nonsense.
You copied from him much more than he did from you. You initially copied the whole miner.... Then some fixes, some enhancements, etc.
Tpruvot is improving ccminer every day and doesn't ask for a penny.

[sp_]

5% loss Made in France

You could have avoided that... tpruvot did so much (and is still doing) to cpuminer and ccminer!

I see no real optimalizations. All he does is to copy from others.

That's nonsense.
You copied from him much more than he did from you. You initially copied the whole miner.... Then some fixes, some enhancements, etc.
Tpruvot is improving ccminer every day and doesn't ask for a penny.

Tpruvot has forked CCminer. Please read the credits. It's based on poolers work and the two Christians (CC).. 
Tpruvot has added a remote control feature and buffer overflows, and a few new algos copied from djm34.