CCminer(SP-MOD) Modded GPU kernels.

[Eliovp]

Guys , is there any optimization on ETH for Maxwell GPUs? Last releases are much slower than the AMD cards...

Try to use Genoil's chunky miner.

That could help to increase your hash rate.

here's a link: https://forum.ethereum.org/discussion/4212/chunky-opencl-miner

[1714795862]

1714795862

[1714795862]

1714795862

[1714795862]

1714795862

[sp_]

your release 78 is so crappy that I lose even 50kh/s...  most of that is on you (as usual)
but I was asked to use it so...

I have submitted some improvements @ github.

gtx 960: +20khash
gtx:970: +40khash
gtx 750ti: +5khash

I removed this call:

cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync);

I have also solved the broken hash on the 970. A sync issue.., Will submit soon.

[djm34]

your release 78 is so crappy that I lose even 50kh/s...  most of that is on you (as usual)
but I was asked to use it so...

I have submitted some improvements @ github.

gtx 960: +20khash
gtx:970: +40khash
gtx 750ti: +5khash

I removed this call:

cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync);

I have also solved the broken hash on the 970. A sync issue.., Will submit soon.

actually cudadevicescheduleblockingsync makes no difference to me (actually this is the cuda 7.5 best parameter which slows down the card when run with cuda 6.5

[bensam1231]

so basically you expect other dev to work on the opensource so to make your private miner faster
pallas says thank you to you  

I will study your new optimized Neoscrypt code and learn.
You should study my code as well. We use different ways to reach our goal. I do small compiler/assembly optimalizations while you are reinventing the algorithm. You should buy my private miner and analyze it. I will give you a discount of 0.05 BTC.  

that's your problem, you need to study my code to get something out, while if you knew what you were doing it is the original code would have to read to make it easier for the gpu.
Basically you don't get any real upgrade because you don't look at what the code does, you are just trying random stuff expecting it will work based on what and other devs wrote... well good luck with that

That is basically how science works in a nutshell. You build off other peoples research. If they get past hurdles you couldn't it gives you ideas and different ways of looking at things to help surpass their work. Remaking the wheel is stupid and it's actually pretty asinine that developers assume everyone should learn from the ground up without assistance besides wizard style 'hints'. This is also the fun that's entailed with 'nix when I was using it. People would hint at what I needed to do instead of tell me what I needed to do and why to make it seem like their knowledge has more meaning.

no it isn't how science works.
Yes you build on other people works, but you do it with an understanding of what has been done before and the inner concept behind... and
once you have reached a dead end (optimization wise...) you get back to the basis.

And in the case of neoscrypt if you don't see that fastkdf is just a massive rotation, there isn't much you can do on top of what has been done...

Have you ever read a research paper? They give you everything. We're operating under the assumption that everyone involved knows the basics and they need to go from there, which includes crossing common hurdles with new techniques.

I can only imagine how you guys would work in groups. Instead of trying to build concepts you're like 'I know this piece of information and we can spend the next 20 minutes while you try to guess how I got there and rub my penis in the process!'.


Does anyone have a batch file to restart a miner every so often?

[sp_]

your release 78 is so crappy that I lose even 50kh/s...  most of that is on you (as usual)
but I was asked to use it so...

I have submitted some improvements @ github.
gtx 960: +20khash
gtx:970: +40khash
gtx 750ti: +5khash
I removed this call:
cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync);
I have also solved the broken hash on the 970. A sync issue.., Will submit soon.

actually cudadevicescheduleblockingsync makes no difference to me (actually this is the cuda 7.5 best parameter which slows down the card when run with cuda 6.5

I have sendt version 5 of the SP-MOD private to the donators. It contains a bugfixed and improved  NEOSCRYPT kernal. Works on all models including the gtx 970.

[sp_]

You can become a subscriber for only 0.1BTC. (0.05 discount for DJM34)

My privater miner contains all the algorithms on nicehash optimized. 0-30%

[sp_]

My 980ti is doing 835khash when compiled with cuda 6.5.
Good job.

Sp-Mod private 5 is doing 900Khash on the 980ti standard clocks. 7.7% faster than djm34 's opensource version.

[djm34]

My 980ti is doing 835khash when compiled with cuda 6.5.
Good job.

Sp-Mod private 5 is doing 900Khash on the 980ti standard clocks. 7.7% faster than djm34 's opensource version.

well considering you reported completely false number in the first place

[djm34]

your release 78 is so crappy that I lose even 50kh/s...  most of that is on you (as usual)
but I was asked to use it so...

I have submitted some improvements @ github.
gtx 960: +20khash
gtx:970: +40khash
gtx 750ti: +5khash
I removed this call:
cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync);
I have also solved the broken hash on the 970. A sync issue.., Will submit soon.

actually cudadevicescheduleblockingsync makes no difference to me (actually this is the cuda 7.5 best parameter which slows down the card when run with cuda 6.5

I have sendt version 5 of the SP-MOD private to the donators. It contains a bugfixed and improved  NEOSCRYPT kernal. Works on all models including the gtx 970.

considering the perf you show on the 750ti, I suspect you just make out number...
(anyhow, I get paid to watch your shit show... I shouldn't care too much   but yeah that just confirms that you are just an asshole)

right you are not an asshole, just some jealous ridiculous prick... yeah I understand it is difficult to realize that without me (or other) releasing new kernel your whole business is falling apart. 

[sp_]

Your miner doesn't work. It gives "Does not validate on the cpu error messages" and lower rates on the pool. (gtx 970)

I had to spend half a day to fix it.
I also found some optimalizations.

[djm34]

Your miner doesn't work. It gives "Does not validate on the cpu error messages" and lower rates on the pool. (gtx 970)

I had to spend half a day to fix it.
I also found some optimalizations.

lol one day to uncomment the __syncthread(), which gives problem on a card I don't have, you're slow   (hope it wasn't too difficult)
(considering I haven't even finished to release it  )

[djm34]

so basically you expect other dev to work on the opensource so to make your private miner faster
pallas says thank you to you 

I will study your new optimized Neoscrypt code and learn.
You should study my code as well. We use different ways to reach our goal. I do small compiler/assembly optimalizations while you are reinventing the algorithm. You should buy my private miner and analyze it. I will give you a discount of 0.05 BTC. 

that's your problem, you need to study my code to get something out, while if you knew what you were doing it is the original code would have to read to make it easier for the gpu.
Basically you don't get any real upgrade because you don't look at what the code does, you are just trying random stuff expecting it will work based on what and other devs wrote... well good luck with that

That is basically how science works in a nutshell. You build off other peoples research. If they get past hurdles you couldn't it gives you ideas and different ways of looking at things to help surpass their work. Remaking the wheel is stupid and it's actually pretty asinine that developers assume everyone should learn from the ground up without assistance besides wizard style 'hints'. This is also the fun that's entailed with 'nix when I was using it. People would hint at what I needed to do instead of tell me what I needed to do and why to make it seem like their knowledge has more meaning.

no it isn't how science works.
Yes you build on other people works, but you do it with an understanding of what has been done before and the inner concept behind... and
once you have reached a dead end (optimization wise...) you get back to the basis.

And in the case of neoscrypt if you don't see that fastkdf is just a massive rotation, there isn't much you can do on top of what has been done...

Have you ever read a research paper? They give you everything. We're operating under the assumption that everyone involved knows the basics and they need to go from there, which includes crossing common hurdles with new techniques.

I can only imagine how you guys would work in groups. Instead of trying to build concepts you're like 'I know this piece of information and we can spend the next 20 minutes while you try to guess how I got there and rub my penis in the process!'.


Does anyone have a batch file to restart a miner every so often?

I read quite a few (and wrote some as well) so ?

[Eliovp]

My 980ti is doing 835khash when compiled with cuda 6.5.
Good job.

Sp-Mod private 5 is doing 900Khash on the 980ti standard clocks. 7.7% faster than djm34 's opensource version.

Sp, put that on my mining hardware comparison website if you Will.

You can Also include the image and write a comment after it's activated if needed :-)

Check my sig for link 😁

[djm34]

My 980ti is doing 835khash when compiled with cuda 6.5.
Good job.

Sp-Mod private 5 is doing 900Khash on the 980ti standard clocks. 7.7% faster than djm34 's opensource version.

Sp, put that on my mining hardware comparison website if you Will.

You can Also include the image and write a comment after it's activated if needed :-)

Check my sig for link 😁

actually the perf reported by sp are pretty shitty... which is pitty considering he uses my kernel.
my 980 gets into the 800kh/s (stock clock 830 with the __syncthreads 850 without)
not really good at optimizing he is

[scryptr]

I APOLOGIZE FOR FALLING ASLEEP--

But I did wake up and found 2 and 1/2 pages of posts that threaten to put me back to sleep.

My miners toiled away like zombies while I was in bed  In a few days, if the beer holds out, I'll be able to donate again... or maybe buy beer instead.       --scryptr

[djm34]

I APOLOGIZE FOR FALLING ASLEEP--

But I did wake up and found 2 and 1/2 pages of posts that threaten to put me back to sleep.

My miners toiled away like zombies while I was in bed  In a few days, if the beer holds out, I'll be able to donate again... or maybe buy beer instead.       --scryptr

you just need more popcorn (I am grateful to myself to have used the moderation feature on my thread )

[pallas]

I APOLOGIZE FOR FALLING ASLEEP--

But I did wake up and found 2 and 1/2 pages of posts that threaten to put me back to sleep.

My miners toiled away like zombies while I was in bed  In a few days, if the beer holds out, I'll be able to donate again... or maybe buy beer instead.       --scryptr

you just need more popcorn (I am grateful to myself to have used the moderation feature on my thread )

you deleted one of my posts, now I'm gonna troll you forever!!!!

just joking ;-D

[sp_]

Just downloaded and compiled your latest version DJM34. The does not validate messages are gone on my gtx 970.
.
Compiled with Visual C++ 18 using Nvidia CUDA Toolkit 7.5

  Based on pooler cpuminer 2.3.2 and the tpruvot@github fork
  CUDA support by Christian Buchner, Christian H. and DJM34
  Includes optimizations implemented by sp, klaust, tpruvot, tsiv and pallas.

[2016-01-24 17:10:23] NVAPI GPU monitoring enabled.
[2016-01-24 17:10:23] 1 miner thread started, using 'neoscrypt' algorithm.
[2016-01-24 17:10:23] Binding thread 0 to cpu 0 (mask 1)
[2016-01-24 17:10:25] GPU #0 Found nounce 000bbb86
[2016-01-24 17:10:39] GPU #0 Found nounce 009955c5
[2016-01-24 17:10:39] GPU #0: GeForce GTX 970, 639
[2016-01-24 17:10:41] GPU #0 Found nounce 00aab9f6
[2016-01-24 17:10:41] GPU #0: GeForce GTX 970, 635
[2016-01-24 17:10:41] Total: 597.10 kH/s
[2016-01-24 17:10:43] GPU #0 Found nounce 00bb9b78
[2016-01-24 17:10:43] GPU #0: GeForce GTX 970, 617
[2016-01-24 17:10:43] Total: 601.99 kH/s
[2016-01-24 17:11:27] GPU #0 Found nounce 02667a39
[2016-01-24 17:11:27] GPU #0: GeForce GTX 970, 634
[2016-01-24 17:11:27] Total: 608.49 kH/s
[2016-01-24 17:11:43] CTRL_C_EVENT received, exiting once miner jobs complete.  Ctrl+C again to abort miner jobs
[2016-01-24 17:11:43] GPU #0 Found nounce 030286b5
[2016-01-24 17:11:43] GPU #0: GeForce GTX 970, 633
[2016-01-24 17:11:43] Total: 612.61 kH/s
[2016-01-24 17:11:43] CTRL_C_EVENT received, aborting miner jobs

[sp_]

Here is my buyable private miner:

                       SP-Mod Private #5
Compiled with Visual C++ 18 using Nvidia CUDA Toolkit 7.5

 Based on pooler cpuminer 2.3.2 and the tpruvot@github fork
 CUDA support by Christian Buchner, Christian H. and DJM34
 Includes optimizations implemented by sp, klaust, tpruvot, tsiv and pallas.

2016-01-24 17:15:25] NVAPI GPU monitoring enabled.
2016-01-24 17:15:25] 1 miner thread started, using 'neoscrypt' algorithm.
2016-01-24 17:15:25] Binding thread 0 to cpu 0 (mask 1)
2016-01-24 17:15:27] GPU #0 Found nounce 000bbb86
2016-01-24 17:15:31] GPU #0 Found nounce 00354a70
2016-01-24 17:15:31] GPU #0: GeForce GTX 970, 669
2016-01-24 17:15:31] Total: 587.70 kH/s
2016-01-24 17:15:40] GPU #0 Found nounce 009955c5
2016-01-24 17:15:40] GPU #0: GeForce GTX 970, 672
2016-01-24 17:15:40] Total: 608.69 kH/s
2016-01-24 17:15:42] GPU #0 Found nounce 00aab9f6
2016-01-24 17:15:42] GPU #0: GeForce GTX 970, 668
2016-01-24 17:15:42] Total: 620.64 kH/s
2016-01-24 17:15:44] GPU #0 Found nounce 00b9a60e
2016-01-24 17:15:44] GPU #0: GeForce GTX 970, 669
2016-01-24 17:15:44] Total: 628.68 kH/s
2016-01-24 17:15:50] GPU #0 Found nounce 00fd8a32
2016-01-24 17:15:50] GPU #0: GeForce GTX 970, 664

[sp_]

If let it run for a while yours is making 630khash, and my version is doing 670khash. (+6.34%)