CCminer(SP-MOD) Modded GPU kernels.

[mintminty59]

Will you be releasing a already compiled version for windows?

1.5.52(sp-mod)
@sp-hash sp-hash released this 11 minutes ago

-Faster myriad-groestlcoin (myr-gr)
-Faster quark
-compiled with the latest nvidia driver(31-mai-2015) (980ti support)


1.5.52(sp-MOD) is available here: (06-6-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.52

The sourcecode is available here:

https://github.com/sp-hash/ccminer

[1714913407]

1714913407

[1714913407]

1714913407

[sp_]

released. I had to rollback a change in aes that was not faster.

[Grout]

I'm having trouble compiling on linux (ubuntu 15.04):

Code:

nvcc -gencode=arch=compute_52,code=\"sm_52,compute_52\" -gencode=arch=compute_50,code=\"sm_50,compute_50\" -I/usr/local/cuda/include -I.  --ptxas-options="-v" --maxrregcount=128 -o heavy/heavy.o -c heavy/heavy.cu
nvcc fatal   : Unsupported gpu architecture 'compute_52'
Makefile:2048: recipe for target 'heavy/heavy.o' failed
make[2]: *** [heavy/heavy.o] Error 1

[joblo]

I'm having trouble compiling on linux (ubuntu 15.04):

Code:

nvcc -gencode=arch=compute_52,code=\"sm_52,compute_52\" -gencode=arch=compute_50,code=\"sm_50,compute_50\" -I/usr/local/cuda/include -I.  --ptxas-options="-v" --maxrregcount=128 -o heavy/heavy.o -c heavy/heavy.cu
nvcc fatal   : Unsupported gpu architecture 'compute_52'
Makefile:2048: recipe for target 'heavy/heavy.o' failed
make[2]: *** [heavy/heavy.o] Error 1

You need the version of cuda 6.5 that supports 9xx cards.

https://developer.nvidia.com/cuda-downloads-geforce-gtx9xx

If you don't have any 9xx cards you can remove compute_52 from Makefile.am.
Edit: assuming Linux, edit project properties on windows.
Note the "+" was removed from the compute_50 line.

Code:

#nvcc_ARCH = -gencode=arch=compute_52,code=\"sm_52,compute_52\"
nvcc_ARCH  = -gencode=arch=compute_50,code=\"sm_50,compute_50\"
#nvcc_ARCH += -gencode=arch=compute_35,code=\"sm_35,compute_35\"
#nvcc_ARCH += -gencode=arch=compute_30,code=\"sm_30,compute_30\"

You can also run 9xx cards with only compute_50 but presumably at lower performance.

[mintminty59]

released. I had to rollback a change in aes that was not faster.

doesnt work for me under windows

Cuda error in func 'x11_simd512_cpu_init' at line 614 : invalid texture reference.

[Grout]

I'm having trouble compiling on linux (ubuntu 15.04):

Code:

nvcc -gencode=arch=compute_52,code=\"sm_52,compute_52\" -gencode=arch=compute_50,code=\"sm_50,compute_50\" -I/usr/local/cuda/include -I.  --ptxas-options="-v" --maxrregcount=128 -o heavy/heavy.o -c heavy/heavy.cu
nvcc fatal   : Unsupported gpu architecture 'compute_52'
Makefile:2048: recipe for target 'heavy/heavy.o' failed
make[2]: *** [heavy/heavy.o] Error 1

You need the version of cuda 6.5 that supports 9xx cards.

https://developer.nvidia.com/cuda-downloads-geforce-gtx9xx

If you don't have any 9xx cards you can remove compute_52 from Makefile.am.
Edit: assuming Linux, edit project properties on windows.
Note the "+" was removed from the compute_50 line.

Code:

#nvcc_ARCH = -gencode=arch=compute_52,code=\"sm_52,compute_52\"
nvcc_ARCH  = -gencode=arch=compute_50,code=\"sm_50,compute_50\"
#nvcc_ARCH += -gencode=arch=compute_35,code=\"sm_35,compute_35\"
#nvcc_ARCH += -gencode=arch=compute_30,code=\"sm_30,compute_30\"

You can also run 9xx cards with only compute_50 but presumably at lower performance.

Is it a new thing? I compiled 1.5.51 without any problem.
And since my rig only has 750 Ti, I shouldn't even need compute 5.2 for now. Thanks for giving me the line to change. It works.

[joblo]

I'm having trouble compiling on linux (ubuntu 15.04):

Code:

nvcc -gencode=arch=compute_52,code=\"sm_52,compute_52\" -gencode=arch=compute_50,code=\"sm_50,compute_50\" -I/usr/local/cuda/include -I.  --ptxas-options="-v" --maxrregcount=128 -o heavy/heavy.o -c heavy/heavy.cu
nvcc fatal   : Unsupported gpu architecture 'compute_52'
Makefile:2048: recipe for target 'heavy/heavy.o' failed
make[2]: *** [heavy/heavy.o] Error 1

You need the version of cuda 6.5 that supports 9xx cards.

https://developer.nvidia.com/cuda-downloads-geforce-gtx9xx

If you don't have any 9xx cards you can remove compute_52 from Makefile.am.
Edit: assuming Linux, edit project properties on windows.
Note the "+" was removed from the compute_50 line.

Code:

#nvcc_ARCH = -gencode=arch=compute_52,code=\"sm_52,compute_52\"
nvcc_ARCH  = -gencode=arch=compute_50,code=\"sm_50,compute_50\"
#nvcc_ARCH += -gencode=arch=compute_35,code=\"sm_35,compute_35\"
#nvcc_ARCH += -gencode=arch=compute_30,code=\"sm_30,compute_30\"

You can also run 9xx cards with only compute_50 but presumably at lower performance.

Is it a new thing? I compiled 1.5.51 without any problem.
And since my rig only has 750 Ti, I shouldn't even need compute 5.2 for now. Thanks for giving me the line to change. It works.

Yes, fairly new. Since most people have 9xx cards now it only makes sense. The Windows binaries have
had compute 5.2 for a while, you would only run into problems if you tried to compile.

[bathrobehero]

I've seen reviews where they've shown the cards are cooler without backplates. They don't do anything except trap heat. Sometimes they don't even make contact with the memory, which is the only reason to have a backplate in the first place. If it doesn't make contact with anything except the screw mounts it's essentially a insulator.

A lot of backplates and heatsink designs have no contact with the memory or with the VRM modules which means they are pretty much useless and they really are just trapping heat. Unless you manually put thermal pads in between to fill the gap and let the heat actually trasfer because air is terrible at transfering heat.

[bensam1231]

Anyone have problems undervolting. Im trying to undervolt my cards right now and it wont stick. It just rebounds back to 0 in afterburner when I click apply.

Also memory OC doesn't seem to do anything. My memory always seems to operate at the same speed, regardless of what it's set for. It even 'underclocks' under normal memory parameters for the card. All of my cards sit at 3005 for memory speed while mining. When I OC them, they spike when mining starts and when it stops, but they go down to 3005 normally.

1.5.52(sp-mod)
@sp-hash sp-hash released this 11 minutes ago

-Faster myriad-groestlcoin (myr-gr)
-Faster quark
-compiled with the latest nvidia driver(31-mai-2015) (980ti support)

1.5.52 seems just as fast as the last three versions for quark unless I'm missing something.

Setting intensity higher then the default of 24 always causes the miner to error out as well.

In my example the 970 only has one 8pin connector instead of a 6+8. Same watts, less areas for delivery.

My 970 has 2x6 pin PCIe connections which is the same as 1x8 which is enough for the rated power
with headroom for OCing. Some 980s (165W) have only 2x6. That shouldn't cause problems for the
connectors unless they are defective. The only real benefit I see with connector overkill is offloading
the PCIe bus, a good idea in multi GPU rigs.

As far as areas of delivery there are only 2, the bus on one corner of the card, and the connector cluster
at the opposite corner. Adding pins to the connector cluster doesn't create more areas of delivery, it
only increases the capacity of the cluster area. Adding more areas requires power connectors at different
locations on the card, but that would complicate cable management and airflow.

Anyway it looks like your problem is fixed and I don't expect your connectors to melt, but don't sue me.

It's not about what the connectors are rated at, rather the amount of pins available for delivery. Rated delivery for a socket is just a specification. Just the same as you should be able to use 750tis without PCI-E connectors and have no problem, it doesn't always work that way in practice and you end up with boards with burned out traces.

Electrical transfer doesn't work like a thermal transfer.

I wasn't talking about problems for the connectors. Just talking about system stability and ability to OC. I mentioned the problem I had in passing, which has absolutely nothing to do with my GPUs or what I was talking about. I never expected/expect my connector to melt.

I've seen reviews where they've shown the cards are cooler without backplates. They don't do anything except trap heat. Sometimes they don't even make contact with the memory, which is the only reason to have a backplate in the first place. If it doesn't make contact with anything except the screw mounts it's essentially a insulator.

A lot of backplates and heatsink designs have no contact with the memory or with the VRM modules which means they are pretty much useless and they really are just trapping heat. Unless you manually put thermal pads in between to fill the gap and let the heat actually trasfer because air is terrible at transfering heat.

Yup... They also take up more space between cards reducing airflow and act like a hot plate, keeping cards around them hotter.

[aionsz]

have rig with 4 970 and 2 980.
mining speed always jump from 65000 khs to 101 000. with avarage speed 85000. why this is happen? on others rig speed doesnt jump so huge
and sometimes it closes with 28 error.

[go6ooo1212]

release 50 is a little bit faster than 51 and 52 on qubit , to my mixed 750ti/970 rig...

[Handofmidas]

@bensam: When OCing memory, make sure you do so for P2 mode. I think afterburner will only do P0.

[bensam1231]

@bensam: When OCing memory, make sure you do so for P2 mode. I think afterburner will only do P0.

How do you do this or what do you use?

[hashbrown9000]

quark profitability dropping. major hit last week. anyone have suggestions on what else to mine?

[Handofmidas]

@bensam: When OCing memory, make sure you do so for P2 mode. I think afterburner will only do P0.

How do you do this or what do you use?

Nvidia inspector. Also see here:

https://bitcointalk.org/index.php?topic=824764.0

[sambiohazard]

This is crazy if true. Everyone should wait for Pascal to come out in Q1 2016. Start saving for it now or once other s upgrade your Maxwells can't compete.

http://www.guru3d.com/news-story/nvidia-big-pascal-taped-out-gp100-silicon.html

[bensam1231]

@bensam: When OCing memory, make sure you do so for P2 mode. I think afterburner will only do P0.

How do you do this or what do you use?

Nvidia inspector. Also see here:

https://bitcointalk.org/index.php?topic=824764.0

Ah nice... It takes a bit of finagaling, but seems to work. It appears that you have to set your p3 clock higher to set p2 higher. Using 'unlock maximum' on p2 causes the program to error out. It also seems like you can't set p3 higher then +710mhz or it errors out as well.

[chrysophylax]

This is crazy if true. Everyone should wait for Pascal to come out in Q1 2016. Start saving for it now or once other s upgrade your Maxwells can't compete.

http://www.guru3d.com/news-story/nvidia-big-pascal-taped-out-gp100-silicon.html

so there really isnt anything different about what has been happening in the industry ... just following the same steps ... which is great ...

technological advancements at this rate is a very good thing ... but wait? ... why? ...

we have never done that - and never will ...

waiting means that that you will miss out on what is NOW ...

the crypto sphere is going to be a very different animal in 6months time - as will the commercially available products to mine with ...

6months is a massive amount of time to mine and trade and save - all at the same time ...

who knows - btc may just jump back to the $1000USD mark again ...

why would you want to miss out on gaining all those crypto coins so that you CAN purchase these cards when they come out next year ... its what we do ...

i hope they do bring this technology all out sooner than later - as the next generation of chips will be a massive jump in hashrate if history is anything to go by ...

tanx for the read ...

#crysx

[Taek]

I am not sure if there is a process, but could we possibly get Sia into sgminer?

A buggy implementation of the Sia miner here: https://github.com/NebulousLabs/Sia-GPU-Miner

About how long would it take? The algorithm above is not very optimized, we're mostly (for now) concerned about having something that works for most people and isn't super buggy.

[chrysophylax]

I am not sure if there is a process, but could we possibly get Sia into sgminer?

A buggy implementation of the Sia miner here: https://github.com/NebulousLabs/Sia-GPU-Miner

About how long would it take? The algorithm above is not very optimized, we're mostly (for now) concerned about having something that works for most people and isn't super buggy.

pm'd you ...

#crysx