Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 839. (Read 2347659 times)

No. I have done more. My kernel compiles down to 110 registers and no spillbytes. Yours is 213 regs. Do a filecompare and check for yourself.

you just tweaked kernel parameters (which were adjustable by the user in my version btw), so no there is no real difference...
(except the huge overclock on your cards)

Your Lyra2v2 kernal is slow on the gtx 950 and the gtx 960. I have improved it. It is faster on all the maxwell models. My modded kernals are 100-200KHASH bether on the 750tis'

Around 5%.

It's enough to prove that my work is faster.

considering, I never worked on quark, your comparison is irrelevant as usual

I have modded and optimized all the kernals in ccminer, including your neoscrypt and lyra2v2 kernals..
My kernals are the fastest opensource there is for the maxwell cards. If you want to prove me wrong, please release something new. It's been a while now.

no they're not faster...  
And your work is mostly to wait for other people to release them for you... (sorry for not playing by your rules... )

now those are nice figures ...

could you test it in x11 when you have time please? ...

im very curious ...

#crysx

27.5 MHASH in quark stock clocks (release 76)  I have the G1 OC card.
It did 13MHASH in x11 in an earlier release,  but haven't tested the latest build.

i know ...

my cards have ALWAYS shown lower numbers than that of the other miners ...

i have never been able to find out why this is so ...

BUT ... the upside is that in comparison to what these cards were getting - these figures are better hashrates ...

these cards have always done below the 2800kh mark on x11 ... then when i upgraded to f22x54 and c75 - the numbers jumped slightly to 2800kh ...

now they are averaging more - compiled in the exact same way that they were before - except with ccminer-spmod76 ...

what are your numbers on the stock 980ti g1 for x11 and quark? ...

#crysx

But your cards are showing lower numbers than others. I get 2.9-3.1 on most of my cards on the factory clocks. With Overclocking you can reach 3.4-3.5 in the x11 algo. on the 750ti.

edit: I see that your cards are running with a corespeed of 1176 Mhz. this is very low. x11 performs best between 1300  and 1400

sp ...

ccminer-spmod76 has succeeded in hashing more compiled in f22x64c75 than the previous versions on f22x64c65 ...

the average hashrate i used to get per card ( gigabyte 750ti oc lp ) was 2800kh ... now the average is 2847kh ...

this is proof enough for me that you have improved it on x11 more than the previous versions ...

below is a copy of the most recent output of ccminer-spmod76 using the setting - ./ccminer -o stratum+tcp://donate-sp.granitecoin.com:7003/ -O chrysophylax.ace:x -a x11 -X 29 ...

-------

[2015-12-14 23:42:29] GPU #1: GeForce GTX 750 Ti, 2830 (T= 74C F= 60% C=1176/2700)
[2015-12-14 23:42:29] GPU #2: GeForce GTX 750 Ti, 2859 (T= 74C F= 60% C=1176/2700)
[2015-12-14 23:42:30] accepted: 126/133 (94.74%), 14237 kH/s yes!
[2015-12-14 23:42:30] accepted: 127/134 (94.78%), 14237 kH/s yes!
[2015-12-14 23:42:41] GPU #4: GeForce GTX 750 Ti, 2847 (T= 65C F= 54% C=1176/2700)
[2015-12-14 23:42:42] accepted: 128/135 (94.81%), 14237 kH/s yes!
[2015-12-14 23:42:44] GPU #2: GeForce GTX 750 Ti, 2857 (T= 74C F= 60% C=1176/2700)
[2015-12-14 23:42:45] accepted: 129/136 (94.85%), 14237 kH/s yes!
[2015-12-14 23:42:47] donate-sp.granitecoin.com:7003/ x11 block 116788
[2015-12-14 23:42:47] GPU #2: GeForce GTX 750 Ti, 2855 (T= 74C F= 60% C=1176/2700)
[2015-12-14 23:42:47] GPU #4: GeForce GTX 750 Ti, 2850 (T= 65C F= 54% C=1176/2700)
[2015-12-14 23:42:47] GPU #3: GeForce GTX 750 Ti, 2845 (T= 71C F= 57% C=1176/2700)
[2015-12-14 23:42:47] GPU #1: GeForce GTX 750 Ti, 2829 (T= 74C F= 60% C=1176/2700)
[2015-12-14 23:42:47] GPU #0: GeForce GTX 750 Ti, 2861 (T= 76C F= 62% C=1189/2700)
[2015-12-14 23:42:53] GPU #1: GeForce GTX 750 Ti, 2830 (T= 74C F= 60% C=1176/2700)
[2015-12-14 23:42:54] accepted: 130/137 (94.89%), 14237 kH/s yes!

-------

this is currently running on the donation link http://donate-sp.granitecoin.com:7003/ on nicehash - https://www.nicehash.com/?p=miners&addr=1CTiNJyoUmbdMRACtteRWXhGqtSETYd6Vd&a=3&l=0 ...

tanx ...

#crysx

Just different ways to work. I don't wait 6 months to publish my kernals. They are getting faster every week, and most of my work is opensource.

Not my problem, this algo was developed for cuda 6.5 (the newest algo, still unpublished, which is a lot faster would be the only one I would consider for upgrade...).
This is funny how when it comes to announce small and meaningless speed increase, you are there, and how it becomes everyone else fault when an algo gets slower 

19-september 2014 (ccminer DJM34 version)
http://cryptomining-blog.com/3503-crypto-mining-performance-of-the-new-nvidia-geforce-gtx-980/




Quark was hashing at 12322 on the reference gtx 980 card (2048 shaders)
Quark is now  hashing at 12300 on a overclocked gtx 960 oc. (1024 shaders) and around 20MHASH on the reference 980 cards.(ccminer sp-mod release 76)'

A total of 63% gain

Thanks for your support

oh yes, I forgot the 5% stuff, we see all the time and nobody can notice 

I just gave you 5% more quark on compute 5.2 devices(release 76). Everybody Know I have a faster private kernal.

But for only a few beers in donations,  I will not publish it.


For the donators: if you donate a total amount of 0.1 btc you can get a copy of my private kernals.

1. Spreadcoin 10-20% faster (0.1BTC) (full sourcecode and linux compatible)
2. Cryptonight 10% faster (0.1BTC)
3. pentablake 100-120% faster (0.3BTC)

Please restore the performance of your neoscrypt kernal on cuda 7.5. My 750ti is only hashing @ 60KHASH.  66% slower

except we know that nothing noticeable comes out at the end...
so as far as I am concerned, these are just baseless promises...

sp ...

how do you find the test machine going? ...

i think its is powering along - even at the minimal rate that the 750ti runs at ...

in all honesty - i knew this would be a good thing when the code was tuned for c7.5 and wrote that earlier ...

the tests have proven that is the case - and not only the case - but shows there is still room for improvement ...

i am currently building a machine that will mine on all 5 cards for the donation-sp link - and will take the test machine off ... ill set it to x11 and see how that factors in for the speed and longer term performance of the cards with the latest git build ... if the average rate is more than 2800kh in x11 on these cards - then you have succeeded in doing better than c6.5 builds ...

this machine will be ready soon ... in fact - within the next half hour ...

its then sleep for me - then all day work on the other system i have been wanting to build for so long ... some of the components are already here ... the next few weeks will be the rest of the components for the granite 'grunt' system ... ill be building that system in fedora 23 x64 c7.5 ... cant wait ...

keep an eye on the x11 eu stratum at nicehash in the next 30mins ... https://www.nicehash.com/?p=miners&addr=1CTiNJyoUmbdMRACtteRWXhGqtSETYd6Vd&a=3&l=0 ... ill leave that running for a couple of days to donate a further bit of btc for you ...

#crysx

X11 and neoscrypt is performing terrible in release 74 compiled for cuda 7.5.
I have almost reached the performance of the cuda 6.5 build now with only 4 kernals modified. (x11, x13 (750ti))

release 76-git is 20% faster than a vanilla build of release 74 using cuda 7.5 build (x86)