CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 707.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: crysx on April 10, 2016, 08:18:47 PM

sp ...
im having issues with my email from the isp side - and am not receiving ONE email from ANYONE ...
could you please send to another email address if possible - just as a temporary solution until i can get my emails fixed again? ... ill pm you the new email if you can ...
i havent received not one of the updates ...
tanx ...
#crysx

Why can't you get your account back here on bitcointalk? I don't trust newbies.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: alexis78 on April 10, 2016, 04:07:15 PM

I didnt get this about GSPREC, you claim credits for replacing 8/16 bit rotations with __byte_perm? or for using __funnelshift?

The point is that I have already optimized blake-256 to run 100% faster than the tvpruvot version. (based on sph code) This is the code you forked and improved in your vanilla mod. I removed the usage of the c_sigma[16][16] table by moving it to the instruction cache. Small simple modifications, that give a big boost.
Byteperm instead of funnels shift gave another 10% boost.

https://github.com/sp-hash/ccminer/commit/9c8976ec8ec4f880a9420176c287008f0d91fe1c

My decred is 300% faster than the tvpruvot 14 rounds blake-256 implementation.

crysx

sr. member

Activity: 364

Merit: 260

--- ChainWorks Industries ---

Quote from: sp_ on April 09, 2016, 02:35:41 PM

Vanilla (V-cash) sp-mod #6 sendt to the donators. Please test and report your numbers compared to the alexis Kernal.

Compute 5.0 have a default intensity of 29.
Compute 5.2 have a default intensity of 30.

I get faster hashrates on all my cards. gtx 750,750ti,950,960,970,980,980ti

Compiled with the 364.51 driver. Cuda 7.5 32 bit.

sp ...

im having issues with my email from the isp side - and am not receiving ONE email from ANYONE ...

could you please send to another email address if possible - just as a temporary solution until i can get my emails fixed again? ... ill pm you the new email if you can ...

i havent received not one of the updates ...

tanx ...

#crysx

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: pallas on April 10, 2016, 04:12:35 PM

I didn't sell it :-D
BTW the problem with windows wasn't in my kernel.

What was the problem? Did you test #4 on Linux yet? What hashrates are you getting?

pallas

legendary

Activity: 2716

Merit: 1094

Black Belt Developer

Quote from: sp_ on April 10, 2016, 04:10:42 PM

Quote from: alexis78 on April 10, 2016, 04:07:15 PM

I got angry because 2 people reported back.
And they said that it's not faster.
Yet you claimed 30->15->8% faster

Pallas claimed +10% and he was wrong on the 750ti. But he was also right on the gtx 970 and Linux.

I didn't sell it :-D
BTW the problem with windows wasn't in my kernel.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: alexis78 on April 10, 2016, 04:07:15 PM

I got angry because 2 people reported back.
And they said that it's not faster.
Yet you claimed 30->15->8% faster

Pallas claimed +10% and he was wrong on the 750ti and windows (less than 1%). But he was also right on the gtx 970 and Linux.

alexis78

member

Activity: 61

Merit: 10

Quote from: sp_ on April 10, 2016, 03:38:09 PM

Here is some blake-256 history before my modded kernals. 225MHASH

https://github.com/sp-hash/ccminer/commits/c0f1a819dae6c2ed9882db961239e32b8683c00b/blake32.cu

here is some of my modded history: (blake-256 is implemented in more than one file)

https://github.com/sp-hash/ccminer/commits/windows/Algo256/cuda_blake256.cu

If you wonder who made the GSPREC macro first it was me (used in the Alexis vcash kernal):

Lyra2v2 +200Khash on gtx970 (+2%)
https://github.com/sp-hash/ccminer/commit/edc63ab0d0c529b307ed8cf8fb12644fa1811ead

You fork my work, and then you get angy when I fork and optimize yours.. Why do you opensource?

I didnt get this about GSPREC, you claim credits for replacing 8/16 bit rotations with __byte_perm? or for using __funnelshift?

I got angry because 2 people reported back.
And they said that it's not faster.
Yet you claimed 30->15->8% faster

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: myagui on April 10, 2016, 03:47:07 PM

@sp_
Open sourcing something that you sold as private to other folks does not seem like a fair deal, just my 2 cents.
Claiming performance boosts on account of a default intensity bump is pretty f#$k'ed up, again just my 2 cents.
(That's 4 cents I've donated already, you're welcome! Grin

)

All my decred donators have a faster version. Decred #4 is slow and contain bugs. I think this is fair. I will opensource some of the old Spreadcoin private miners as well. F.ex spreadcoinminer #5. (the latest is version #9). Pallas just did a 10% increase opensource so decred #4 is not 15% faster anymore. (but doesn't work on windows/750ti)

myagui

legendary

Activity: 1154

Merit: 1001

@sp_
Open sourcing something that you sold as private to other folks does not seem like a fair deal, just my 2 cents.
Claiming performance boosts on account of a default intensity bump is pretty f#$k'ed up, again just my 2 cents.
(That's 4 cents I've donated already, you're welcome! Grin

)

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Here is some blake-256 history before my modded kernals. 14 rounds 225MHASH

https://github.com/sp-hash/ccminer/commits/c0f1a819dae6c2ed9882db961239e32b8683c00b/blake32.cu

here is some of my modded history: (blake-256 is implemented in more than one file)

https://github.com/sp-hash/ccminer/commits/windows/Algo256/cuda_blake256.cu

If you wonder who made the GSPREC macro first it was me (used in the Alexis vcash kernal):

faster lyra2 (gsprec macro) @sp-hash sp-hash committed on May 5 2015
https://github.com/sp-hash/ccminer/commit/9c8976ec8ec4f880a9420176c287008f0d91fe1c

Blakecoin 10% faster, small speedup lyra (byteperm instead of rotations)
https://github.com/sp-hash/ccminer/commit/60426259b061d5428609e622ca63d225ab70bdaa

Lyra2v2 +200Khash on gtx970 (+2%)
https://github.com/sp-hash/ccminer/commit/edc63ab0d0c529b307ed8cf8fb12644fa1811ead

etc..

You fork my work, and then you get angy when I fork and optimize yours.. Why do you opensource?

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: joblo on April 10, 2016, 11:21:54 AM

This is getting out of hand. But let's remember how this all started, when SP went closed after
importing lots of open source improvements that were contributed in good faith by various developpers.

I have opensourced my decred private #4 Original work by me..

Here is The First blake-256 kernal: (14 rounds 220MHASH on the 750ti) Mostly copied from the sph c version of blake-256

Tanguy Pruvot - Nov. 2014

https://github.com/sp-hash/ccminer/blob/c0f1a819dae6c2ed9882db961239e32b8683c00b/Algo256/blake256.cu

In the sp-mod project I have made it 300% faster. (decred 14 rounds)

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: alexis78 on April 10, 2016, 03:05:32 PM

So, what are you trying to say? That you are essentially selling my source with a greater hardcoded intensity?

I am not selling anything. I optimize for for my cards. 750ti,950,960. Why don't you try my decred #4 and check the speed.
The point is that I am not lying when I write that my kernal was more than 30% faster. I compiled yours, tested it, and was surprised that it was so slow.
Then I found the "problem", and then I showed you where to fix the "bug" that was causing the loss of speed(only on windows).

If operatingsystem=windows then reduce the speed 30%

alexis78

member

Activity: 61

Merit: 10

Quote from: sp_ on April 10, 2016, 03:02:36 PM

Quote from: alexis78 on April 10, 2016, 02:32:30 PM

I came here because you reported 30%,

Because your kernal used a default intensity of 24 on windows. That's pretty stupid isn't it. You loose 30%, it doesn't matter how good the code is, if you launch your kernal with retarded default parameters.. In your code, you case on windows, and reduce the speed with 1 line of code..
Cryptominers only care about two things. Speed and Stability.

So, what are you trying to say? That you are essentially selling my source with a greater hardcoded default intensity?

Intensity on ccminer is under the user defined parameters. The hardcoded ones are for the newcomers in order not to freeze their systems on the first place while digging into ccminer.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: alexis78 on April 10, 2016, 02:32:30 PM

I came here because you reported 30%,

Because your kernal used a default intensity of 24 on windows. That's pretty stupid isn't it. You loose 30%, it doesn't matter how good the code is, if you launch your kernal with retarded default parameters.. In your code, you case on windows, and reduce the speed with 1 line of code..
Cryptominers only care about two things. Speed and Stability.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

In my fork I have made a new parameter called -X (xintensity) just like in sgminer. Most of the kernals have -X as default and not -i

airwalker2662

member

Activity: 116

Merit: 10

Quote from: sp_ on April 10, 2016, 02:42:23 PM

Quote from: airwalker2662 on April 10, 2016, 02:37:34 PM

I am curious of seeing what the default intensity is per miner. Is their a command argument that will show intensity in ccminer? I know debug does not.

There is no command, but for some algos you can check the gpu memory being used with gpu-z, and then try to match. Or just look in my opens sourcecode.

Wouldn't this be a simple addition to include with debug? To just pass the number in the algo used for the card being used? I think it would be very useful..

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: djm34 on April 10, 2016, 02:36:00 PM

he isn't a newbie, he wrote the fast whirlpoolx algo you did paste in your release.
I think we are several to call your copy and paste a scam... selling to someone a program where the only thing you change is the intensity, is a scam. There is no hard work. You are not fooling any one at that level.

I base my work on opensource just like the rest of you. I made whirlpoolx 10-15% faster after Alexis's first commit.. Some of the speedup was increasing intensities and a rewrite to more than one nonce per thread. All my work is documented on github and the credits are preserved. (first commit 8 march 2015)
the whirlpoolx kernal have 18 commits:

https://github.com/sp-hash/ccminer/commits/windows/x15/whirlpoolx.cu

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: airwalker2662 on April 10, 2016, 02:37:34 PM

I am curious of seeing what the default intensity is per miner. Is their a command argument that will show intensity in ccminer? I know debug does not.

There is no command, but for some algos you can check the gpu memory being used with gpu-z, and then try to match. Or just look in my opens sourcecode.

sp_

legendary

Activity: 2954

Merit: 1087

Team Black developer

Quote from: alexis78 on April 10, 2016, 02:32:30 PM

I came here because you reported 30%, 15%, 8% increase in the open sourced vcash kernel.
Since nobody validated this increase and people were reporting that your hashrate is rather similar or slightly lower

The different NVIDIA graphic cards behave different. If you test your kernal on gtx 950, gtx 960 and the gtx960M. I did the first vanilla mod on the gtx 960m. and my results are real and can be verified. Why don't you start with decred #4 opensource. What are you getting on Linux?

airwalker2662

member

Activity: 116

Merit: 10

Quote from: alexis78 on April 10, 2016, 02:32:30 PM

I came here because you reported 30%, 15%, 8% increase in the open sourced vcash kernel.
Since nobody validated this increase and people were reporting that your hashrate is rather similar or slightly lower
i admit i got out of temper (my bad).
Especially when your argument about the boost was that you hardcoded a bigger intensity.

As i said above, I dont care where people invest their btcs.
I was here to find out if you were speaking the truth about the increase.

I am curious of seeing what the default intensity is per miner. Is their a command argument that will show intensity in ccminer? I know debug does not.

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 707. (Read 2347641 times)