CCminer(SP-MOD) Modded GPU kernels.

[jjjordan]

sp_, may I get the latest spreadminer you have released?
By the way, I have more than ten EVGA GTX 970s for sale. I'm in US, if anybody's interested - please PM.
This one: EVGA GeForce GTX 970 SC GAMING ACX 2.0 Part Number: 04G-P4-2974-KR
http://www.evga.com/Products/Specs/GPU.aspx?pn=8296b494-5d1d-400d-b2e8-d00ccd71e088

[alexis78]

Since your revision of my source has nothing to offer on the majority of the gpus, <edit>even 6 months after the release</edit> (except the 1.7% fee you take in your pocket).
Can we see the source code please?

Hi! Can I have your btc address for donation? Cannot find it on readme at github.
Better to put it here.
Your miner is absolutely the best in all algos I tested!

please do forward any donations to Epsylon.
He is taking great care of the community by keeping ccminer free - open sourced - and up to date from any aspect.
Consider my contributions as a donation to epsylon's work too

[sp_]

Since your revision of my source has nothing to offer on the majority of the gpus, <edit>even 6 months after the release</edit> (except the 1.7% fee you take in your pocket).
Can we see the source code please?

Why don't you extract the ptx, you will see that it is different.
I left the exe wide open.

I get +2-6% faster on all the cards I have tested. Gtx 1060 3gb, gtx 1070, gtx 970.

100% tdp, small oc. -500 on the memory.


My kernel vs your kernel (gtx 1060 3gb) same launch config, compiled with the same compiler (cuda 7.5), same driver:



You can profile it yourself.

[alexis78]

Since your revision of my source has nothing to offer on the majority of the gpus, <edit>even 6 months after the release</edit> (except the 1.7% fee you take in your pocket).
Can we see the source code please?

Why don't you extract the ptx, you will see that it is different.
I left the exe wide open.

I get +2-6% faster on all the cards I have tested. Gtx 1060 3gb, gtx 1070, gtx 970.

100% tdp, small oc. -500 on the memory.


My kernel vs your kernel (gtx 1060 3gb):



You can profile it yourself.

How come that only you are getting +2-6% out of these and everyone else gets a maximum of 0.5% increase? (scrolling back in your thread you can observe these benchmarks)

...talk is cheap show us the code

[sp_]

How come that only you are getting +2-6% out of these and everyone else gets a maximum of 0.5% increase? (scrolling back in your thread you can observe these benchmarks)
...talk is cheap show us the code

Because the first version I released yesterday had a bug with 100% cpu usage. I have updated the .7z file this morning with a working version.

[alexis78]

How come that only you are getting +2-6% out of these and everyone else gets a maximum of 0.5% increase? (scrolling back in your thread you can observe these benchmarks)
...talk is cheap show us the code

Because the first version I released yesterday had a bug with 100% cpu usage. I have updated the .7z file this morning with a working version.

I'm talking for tests taken from 28th of October and after. Not just yesterday....

Well, regarding the 100% cpu usage "bug".
I assume you tried to mess with the cudaScheduleYield and blockingSync. You should not. there is a reason why I (and everyone else) choose the blockingSync approach over ScheduleYield.

I'm still asking for the source

[newmz]

I respect both of you guys: @sp_  and @alexis78 for what you can do. I just use your code to mine, I can't write it.

I have to say it's very interesting to see you argue about who can produce more hash-rate from their code, but it would be better for all of us out here leeching off your work if you guys would collaborate rather than fight.

I do see  the point made by @alexis78 though - surely open source is better for everyone?

[alexis78]

I respect both of you guys: @sp_  and @alexis78 for what you can do. I just use your code to mine, I can't write it.

I have to say it's very interesting to see you argue about who can produce more hash-rate from their code, but it would be better for all of us out here leeching off your work if you guys would collaborate rather than fight.

I do see  the point made by @alexis78 though - surely open source is better for everyone?

@newmz there is no fight here

Open source is better for everyone but there can be no ccminer community with someone parasitically asking for donations, applying fees, to code that he never wrote or non-existent increases.

All i do here one more time (after the blake-8round sp's default intensity fiasco), is to make sure that everyone will mine efficiently without paying fees to any ccminer pimp.

But in the end, it will be the community's decision which of the both ways to walk, with all pros and cons that imply

[newmz]

I respect both of you guys: @sp_  and @alexis78 for what you can do. I just use your code to mine, I can't write it.

I have to say it's very interesting to see you argue about who can produce more hash-rate from their code, but it would be better for all of us out here leeching off your work if you guys would collaborate rather than fight.

I do see  the point made by @alexis78 though - surely open source is better for everyone?

@newmz there is no fight here

Open source is better for everyone but there can be no ccminer community with someone parasitically asking for donations, applying fees, to code that he never wrote or non-existent increases.

All i do here one more time (after the blake-8round sp's default intensity fiasco), is to make sure that everyone will mine efficiently without paying fees to any ccminer pimp.

But in the end, it will be the community's decision which of the both ways to walk, with all pros and cons that imply

@alexis78 OK I understand and I guess I agree with everything you said. I have often wondered what was going on here with someone offering private versions for a "donation" that will hash better, etc when I see all the stuff at ccminer.org that is open source. I didn't understand but now maybe I understand a bit better....

I must say that your ccminer gets me the best hash-rate on X11, Lyra2REv2 and cryptonight when I mine those algos - so thank you very much for that! Right now I'm mining equihash but I don't think ccminer does that yet right?

[sp_]

Open source is better for everyone but there can be no ccminer community with someone parasitically asking for donations, applying fees, to code that he never wrote or non-existent increases.

The first opensource LBRY GPU kernel and ccminer integration  was done by me 15-16 Jul 2016 (github). 2 days later tvpruvot released his version of the kernel.
After that I have been doing private modded kernels with more speed. It's normal to get payed when you do a job. You got payed by the LBRY team in october (5 months later).
(because at the time the LBRY sp-mod #2 was  around 15% faster than the opensource)

You have never released an official binary of your work. (ccminer 2.0 is using your kernel, but with slow results) That's why I use the profiler to measure the real speed.

6months have passed and LBRY is profitable to mine again...

Now it's time to add another 10% ..

[alexis78]

The first opensource LBRY GPU kernel and ccminer integration  was done by me 15-16 Jul 2016 (github). 2 days later tvpruvot released his version of the kernel.

Please not that mambo jumbo... not to me... i know how to read code

I'm still waiting for the changes you've done to my source code and added a fee on top, not only the latest one, but all of your public revisions. Then we can talk about contributions/commits/integrations etc. Thank you

You keep chewing the "i got paid" by the lbry team. I didnt ask for it and i was not the only one. It was more a team effort.

ccminer2.0 hosts a simple copy paste of our previous work, just to compete with your "complete rewrite claiming"

[NiceHashSupport]

@alexis78, please, check your PM.

[sp_]

...and i was not the only one. It was more a team effort.

But you got payed. It was a job. You based your work on other opensource coders and got payed for it.  Why don't you compile a binary and we can compare the assembly code. I want you to compile it with the compiler and cuda version of your choice.
Please add it to your github. exefile.

Because when I compile your work, I get a 6% slower LBRY Kernel. I might have missed something?? The speed is not there. Are you compiling with compute 6.1?

[alexis78]

Because when I compile your work, I get a 6% slower LBRY Kernel. I might have missed something?? The speed is not there. Are you compiling with compute 6.1?

I'm not compiling for compute 6.1 the configuration of the compilation is already set there in order to make sure that non-tech people can compile their own binaries, and they managed to

Why can't you?

And why can't you just release the source code of all your revisions to my code so that everyone can read/check/test/comment?

speaking of... Nicehash... i remember people arguing here regarding the publication of the "source code" of nanashi's improvements and how the binary was violating licenses and stuff. (He did not just mess with the default intensity, Nanashi's lyra2v2 was a marvelous complete rewrite of the kernel).

[pallas]

nanashi version added warp shuffle (which is not trivial and a very good speed improvement, even though probably not his own idea).
his sources has always been open, but nicehash chose to pay him anyway.
nodoby compiles for compute 6.1 because it's slower: pascal cards run just fine with compute 5.2.

[sp_]

speaking of... Nicehash... i remember people arguing here regarding the publication of the "source code" of nanashi's improvements and how the binary was violating licenses and stuff. (He did not just mess with the default intensity, Nanashi's lyra2v2 was a marvelous complete rewrite of the kernel).

TSIV had a shuffle kernel right after the launch of lyra2v2. I think Djm34 had one as well.  Marvellous code doesn't give you the profit, speed and stabillity does.

[alexis78]

@pallas

My point is not whether the source was public or not.
My point is that people got nervous when they could not find the source code.

@SP_

speed and stabillity does for sure.
Can we see the changes to my source code throughout all your public revisions of the binary based on my fork which gave a 0% increase and 1.7% fee? Thank you

[sp_]

nodoby compiles for compute 6.1 because it's slower: pascal cards run just fine with compute 5.2.

My latest Nexus mod is compiled with cuda 6.1.

Why do you think NVIDIA release a new compute version?  You just need to rewrite abit for more speed.

[sp_]

Can we see the changes to my source code throughout all your public revisions of the binary based on my fork which gave a 0% increase and 1.7% fee? Thank you

Yes. I will release the donation code later.

[alexis78]

Can we see the changes to my source code throughout all your public revisions of the binary based on my fork which gave a 0% increase and 1.7% fee? Thank you

Yes. I will release the donation code later.

Would be really helpful for the flow if you released all revisions of the public one with the fee