CCminer(SP-MOD) Modded GPU kernels.

[sp_]

I was thinking of a hybrid miner when I noticed how well cpuminer-multi hashed lyra2. If the best performing
cpu sub-algo could be offloaded from the gpu, the gpu could work more on the other sub-algos. Even if the offloaded
sub-algo is slower on a cpu wouldn't the parallelization result in a higher overall hash rate?

The problem is the pci-e bandwidth. Moving buffers through 1x pcie will slow.. I was thinking let the cpu work in parallell to find hashes togeter with the gpu.
The cpu code is already written... But perhaps bether to just run 2 miners. one for the cpu and one for the gpu..

[bensam1231]

Something to consider about Axiom, you don't get the amount of coins the pools show per day. Since it's hybrid PoW/PoS, you get about half the quoted coins by the pool, which makes it not every profitable, CPU or GPU mining.

Hello,
I plan to retire my 6x750ti and since everybody is talking about gigabyte 970 what model to choose? There are two Windforce models: GV-N970WF3-4GD and GV-N970WF3OC-4GD...

retire? ...
the best system ( outside of density ) that you can use currently to date ...
another 18 will be added to the farm very shortly to thefarm - once all the 'other' stuff is got out the way ( including all the personal 'stuff' ) ...
are you in australia? ... ill happily have a home for them ...
#crysx

How many cards do you have now? Where I live a used 750ti goes for around $100 on the second hand market. I am expanding slowly, but ROI is still 6-12Months. The reward/risc is high.

What are you mining to get a 6-12 month ROI?!?! Oo

Mining Quark for instance right now, at $.1144 KwH for a 970 nets $.32 per day after power. That ROI in about 1000 days, or three~ years. Back before SAK crashed, it was about 8-12 months for ROI.



Edit: Release .57 is broken for Neoscrypt. One of my machines (W8) it doesn't do anything and hangs when it connects to the pool, my other machine (W10) only one of the 970s works even though it tries to start all three of them.

[joblo]

I was thinking of a hybrid miner when I noticed how well cpuminer-multi hashed lyra2. If the best performing
cpu sub-algo could be offloaded from the gpu, the gpu could work more on the other sub-algos. Even if the offloaded
sub-algo is slower on a cpu wouldn't the parallelization result in a higher overall hash rate?

The problem is the pci-e bandwidth. Moving buffers through 1x pcie will slow.. I was thinking let the cpu work in parallell to find hashes togeter with the gpu.
The cpu code is already written... But perhaps bether to just run 2 miners. one for the cpu and one for the gpu..

I was wondering about bandwidth.  I don't know the details about data movement between cpu and gpu
but that never stpos me from speculating.
What if the cpu handles either the first algo in the chain or the last algo? Would that reduce the data flow
or is the handoff between algos more data intensive than either the start or end of the chain?

I do a quasi hybrid whenever lyra is paying well by running ccminer & cpuminer-multi simultaneously.
I haven't noticed a drop in ccminer performance even when cpuminer-multi is running n threads.
The cpu miners of other algos don't hash nearly as well so it's not worth it.

[djm34]

Axiom with a 80-120% faster cpu miner.
https://github.com/nicehash/cpuminer-multi/releases
By using cpu-simdinstructions.
I think I want to finish the gpu miner now  

Downloaded the win-64bit and can't get it to work error: "unable to start correctly (0xc000007b) click to close."

there are 3 versions, you need to run the correct one for your cpu. (amd or intel?)

cpuminer64_SSE.exe miner: ~220 H/s
cpuminer64_AVX.exe miner: ~240 H/s
cpuminer64_AVX2.exe miner: ~260 H/s

I think the 750ti is only doing around 150 H/s

Perhaps it would be nice to make a hybrid miner. (cpu+gpu)

that would be good ...

a hybrid with the option to use cpu OR gpu - not just cpu AND gpu ...

btw - ive seen somewhere that the 750ti does around the 350H mark ...

#crysx

actually, it isn't possible (whatever hybrid means because in that case, I don't really understand):
The fast part (shabal) is slowed down by mem access.
So moving shabal to cpu would slow down, and saving to ram instead of vram would slow down again.

The only way to increase the speed here is by playing with warp shuffle which won't be practical since the hash is updated at each iteration...
and probably not faster

I was thinking of a hybrid miner when I noticed how well cpuminer-multi hashed lyra2. If the best performing
cpu sub-algo could be offloaded from the gpu, the gpu could work more on the other sub-algos. Even if the offloaded
sub-algo is slower on a cpu wouldn't the parallelization result in a higher overall hash rate?

may-be it is worth trying (well until the 10th... for lyra), don't know.
Problem being to transfer the hashes from the gpu to the cpu do a loop over those hashes, then send them back to the gpu...
not sure if there would be a lot to gain...

[djm34]

I was thinking of a hybrid miner when I noticed how well cpuminer-multi hashed lyra2. If the best performing
cpu sub-algo could be offloaded from the gpu, the gpu could work more on the other sub-algos. Even if the offloaded
sub-algo is slower on a cpu wouldn't the parallelization result in a higher overall hash rate?

The problem is the pci-e bandwidth. Moving buffers through 1x pcie will slow.. I was thinking let the cpu work in parallell to find hashes togeter with the gpu.
The cpu code is already written... But perhaps bether to just run 2 miners. one for the cpu and one for the gpu..

I was wondering about bandwidth.  I don't know the details about data movement between cpu and gpu
but that never stpos me from speculating.
What if the cpu handles either the first algo in the chain or the last algo? Would that reduce the data flow
or is the handoff between algos more data intensive than either the start or end of the chain?

the only algo you would want to put on the cpu is in the middle... for the rest the gpu is much faster than the cpu

[zTheWolfz]

Axiom with a 80-120% faster cpu miner.
https://github.com/nicehash/cpuminer-multi/releases
By using cpu-simdinstructions.
I think I want to finish the gpu miner now  

Downloaded the win-64bit and can't get it to work error: "unable to start correctly (0xc000007b) click to close."

there are 3 versions, you need to run the correct one for your cpu. (amd or intel?)

cpuminer64_SSE.exe miner: ~220 H/s
cpuminer64_AVX.exe miner: ~240 H/s
cpuminer64_AVX2.exe miner: ~260 H/s

I think the 750ti is only doing around 150 H/s

Perhaps it would be nice to make a hybrid miner. (cpu+gpu)

Found the problem and fixed on the SR2, needed to install vcredist_64.
24 cores @ 3.33Ghz getting up to 290.xx h/s at max. about 250h/s with 20 cores, leaving cores free for GPU's to use.

Edit: NiceHash-Axiom-Launcher-v1.0.0.0 works really nice, quick easy setup.
Will rename and test with other cpuminer_x64.exe, SSE and so on from the other zip later to find the best core +hash for my rigs..

[sp_]

Submitted some more speedups in groestl on the 970.

Groestl is now peaking at 23.8 MHASH(benchmark mode)
Quark 16550(benchmark mode)

gigabyte windforce oc


I think the 750ti is running slower with the latest changes. I need to make two seperate kernals.

[sp_]

What are you mining to get a 6-12 month ROI?!?! Oo
Mining Quark for instance right now, at $.1144 KwH for a 970 nets $.32 per day after power. That ROI in about 1000 days, or three~ years. Back before SAK crashed, it was about 8-12 months for ROI.
Edit: Release .57 is broken for Neoscrypt. One of my machines (W8) it doesn't do anything and hangs when it connects to the pool, my other machine (W10) only one of the 970s works even though it tries to start all three of them.

Cost of power: $0,06KWH
Cost of a used board (750ti) $100
Average outside temp in the summer. 10C (I live close to the north pole)
Private kernals +30%

[zTheWolfz]

What are you mining to get a 6-12 month ROI?!?! Oo
Mining Quark for instance right now, at $.1144 KwH for a 970 nets $.32 per day after power. That ROI in about 1000 days, or three~ years. Back before SAK crashed, it was about 8-12 months for ROI.
Edit: Release .57 is broken for Neoscrypt. One of my machines (W8) it doesn't do anything and hangs when it connects to the pool, my other machine (W10) only one of the 970s works even though it tries to start all three of them.

Cost of power: $0,06KWH
Cost of a used board (750ti) $100
Average outside temp in the summer. 10C (I live close to the north pole)
Private kernals +30%

Get LightShots http://app.prntscr.com/en/index.html
easy to use no need to take pictures then upload to the PC.
Click lightshots, drag over what you want to post, it will upload to web, view image, then copy link and post.

[innerchaos]

What are you mining to get a 6-12 month ROI?!?! Oo
Mining Quark for instance right now, at $.1144 KwH for a 970 nets $.32 per day after power. That ROI in about 1000 days, or three~ years. Back before SAK crashed, it was about 8-12 months for ROI.
Edit: Release .57 is broken for Neoscrypt. One of my machines (W8) it doesn't do anything and hangs when it connects to the pool, my other machine (W10) only one of the 970s works even though it tries to start all three of them.

Cost of power: $0,06KWH
Cost of a used board (750ti) $100
Average outside temp in the summer. 10C (I live close to the north pole)
Private kernals +30%

Get LightShots http://app.prntscr.com/en/index.html
easy to use no need to take pictures then upload to the PC.
Click lightshots, drag over what you want to post, it will upload to web, view image, then copy link and post.

I like the Illegal copy of windows message in lower right corner

[GKar]

With the latest commit, I get on Quark 38142 Kh/s on my 6 card EVGA 750 Ti SC with no OC, No Bios change Windows 8.1 Pro (6357 per card).   If I put the fans to 100% I can get about 15 Kh/s more per card.  

X11 19070 Kh/s Rig / 3178.33 Kh/s per card  (Note this is with any -g or -i parameter)

thats the great thing about this fork - sp puts the intensity levels ( -i ) as high as he can as default in the code so that the miner works at peak speed ...

this can backfire with some cards - as the intensity levels are a little too high at default - but usually works quite well overall ...

for those who want to mine as a donation for sp's work - i will be working on getting the donation links running and active today ...

a few more issues to be rid of and the servers ( with donation links ) will be ready to mine with ...

will notify here - https://bitcointalk.org/index.php?topic=1089744 ...

#crysx

Yeah... there is very little configuration for this miner compared to SG. You just plug in pool and algo information and that's it.

That also means people aren't rewarded for tweaking the miner in addition to setting it up, but you know. That's just the elitist miner in me talking.

sp is working on oc setting tuning within the miner though ...

its just limited by the nvidia toolkit and from nvidia themselves ...

eg - memclock and gpuclock is possible under linux using nvidia-smi - BUT - nvidia themselves are limiting availability to the cards that can be oc'd to those that are the 'elite' upper range cards ... my gigabyte 750ti oc lp card is not on that list ...

so when there is more flexibility to do that - i am sure there will be the switches / parameters that will be built into ccminer ( if it is still called that of course ) that will allow easy oc - like that in sgminer ...

i really do like the sgminer 'stats-candy' ...

one can see all that one needs - in one screen ... except for the current block number ( not just the hash ) ... unlike the rolling list that ccminer has ...

for the layman and lazy miner - this is a good thing ... for tweakers and power users / miners - this is a little lame ... but works ...

#crysx

Tpruv's windows version does allow setting gpu clock on the command line.   This is a feature I'd like to see in SP's mods.
 
fyi: I was Angora but lost that account as pw got changed and could not get the reset email.

[GKar]

work finished since a good week... (actually I helped vtc team in tuning their new algo, so the code even predate the algo  )

Interesting. The  vertcoin botnet has over 50% of the hashrate, will they be able to hardfork?

- Botnets have free electricity
- Botnet owners will dump the coins for bitcoins and crash the market price.

The same botnet also is on feathercoin and is on ipominer.   I've pulled my miners off that pool due to that.   Good to hear that vert is addressing the botnet.
 
If the bot owner dumps and the price falls, it will rebound.

[blue-yu]

What are you mining to get a 6-12 month ROI?!?! Oo
Mining Quark for instance right now, at $.1144 KwH for a 970 nets $.32 per day after power. That ROI in about 1000 days, or three~ years. Back before SAK crashed, it was about 8-12 months for ROI.
Edit: Release .57 is broken for Neoscrypt. One of my machines (W8) it doesn't do anything and hangs when it connects to the pool, my other machine (W10) only one of the 970s works even though it tries to start all three of them.

Cost of power: $0,06KWH
Cost of a used board (750ti) $100
Average outside temp in the summer. 10C (I live close to the north pole)
Private kernals +30%

Hey, what are Your OC values, and what cards are You using?

[sp_]

Tpruv's windows version does allow setting gpu clock on the command line.   This is a feature I'd like to see in SP's mods.
 fyi: I was Angora but lost that account as pw got changed and could not get the reset email.

You can also set the gpu clock in my fork. Windows only.(use the latest drivers) Setting the clock will only work on the highend cards. gtx 970/980 etc.

the settings are called:

--gpu-engine and gpu-memspeed

[sp_]

I like the Illegal copy of windows message in lower right corner

It's not illegal It's a trial copy. To remove the error message I type  slmgr -rearm in the administrator window and I get 3 more months of trial.

[sp_]

work finished since a good week... (actually I helped vtc team in tuning their new algo, so the code even predate the algo  )

Interesting. The  vertcoin botnet has over 50% of the hashrate, will they be able to hardfork?
- Botnets have free electricity
- Botnet owners will dump the coins for bitcoins and crash the market price.

The same botnet also is on feathercoin and is on ipominer.   I've pulled my miners off that pool due to that.   Good to hear that vert is addressing the botnet.
 If the bot owner dumps and the price falls, it will rebound.

The new optimized Axiom cpu miner is perfekt for the botnets. The devs battle to avoid gpu-mining might be the final nail in the coffin.

[sp_]

Hey, what are Your OC values, and what cards are You using?

Factory clocks, different cards. Mostly without the 6pin power connector.

[blue-yu]

Hey, what are Your OC values, and what cards are You using?

Factory clocks, different cards. Mostly without the 6pin power connector.

These are great results for quark - i think. I am getting 6600-6800 with my gigabyte 750ti with OC +150cpu +300mem. I see one of Your gpu is doing 8600! Other are 7900, 7700,...

What am I doing wrong?

[sp_]

Hey, what are Your OC values, and what cards are You using?

Factory clocks, different cards. Mostly without the 6pin power connector.

These are great results for quark - i think. I am getting 6600-6800 with my gigabyte 750ti with OC +150cpu +300mem. I see one of Your gpu is doing 8600! Other are 7900, 7700,...
What am I doing wrong?

Private kernal

Average for 6 cards is around 7MHASH on the 750ti. The opensource version does around 5.9MHASH. (factory clocks)

[blue-yu]

Hey, what are Your OC values, and what cards are You using?

Factory clocks, different cards. Mostly without the 6pin power connector.

These are great results for quark - i think. I am getting 6600-6800 with my gigabyte 750ti with OC +150cpu +300mem. I see one of Your gpu is doing 8600! Other are 7900, 7700,...
What am I doing wrong?

Private kernal

Average for 6 cards is around 7MHASH on the 750ti. The opensource version does around 5.9MHASH. (factory clocks)

Aaaa, that is the trick. How much btc is that top secret project