Open Source ZEC (ZCash) GPU Miner AMD & NVidia (up to 45 sol/s on RX480)

[dianasta]

3 x Sapphire 290X  ~ 51 sols/s


[GPU 0] T=-1C A=-1% sols=16.829
[GPU 1] T=-1C A=-1% sols=17.005
[GPU 2] T=-1C A=-1% sols=17.040

hello you can give us your setup memory clock please because i have 4 R9 290 running at 10.7 sol his would be cool thank you

 

Everything is at stock

Core Clock 1000 MHz
MEM Clock 1250 MHz

[Marvell1]

tried it on my 470 six gpu rig got this

GPU 0 found share
Share accepted.
[GPU 0] T=-1C A=-1% sols=17.193
[GPU 1] T=-1C A=-1% sols=17.689
[GPU 2] T=-1C A=-1% sols=17.235
[GPU 3] T=-1C A=-1% sols=17.008
[GPU 4] T=-1C A=-1% sols=16.353
[GPU 5] T=-1C A=-1% sols=17.227
(ST/INV/DUP): 31x 0ch(0/0/0)
GPU 4 found share
Share accepted.
GPU 3 found share
Share accepted.

sometimes as high as 18.5 no higher than that though

[nerdralph]

nice! clocking ~18Sols/s on the same R9 390X that does ~22 using my own brew. your miner has one major advantage though...it actually comes up with solutions that are accepted by the pool  

An optimal solution should be able to do near 100 H/s (i.e. 200 sol/s) on a R9 380.  With 2 million records and 32-64 bytes of memory IO per record * 10 rounds = 1.28GB of memory bandwidth per hash.  The R9 380 with a memory clock of 1.5Ghz has 192GB/s of bandwidth.  The hard part is making full use of that bandwidth when you have small random IO.  Ethash is just at the fringe of being able to maximize memory bandwidth with 128-byte random reads.  Since an equihash solver requires a lot of small reads and writes, the algorithm and data structures need to be optimized for the GPU memory architecture.  i.e. 8 x 32-bit channels working with 2K pages and a 256-byte stride per wavefront.

I expect by the 28th we'll see GPU miners doing 50-60 sol/s, and close to 100 sol/s a month from now.  Hopefully one of the faster ones will be mine. :-)

[snakey]

470 stock bios 14-15sols
470 modded bios 17-18sols
gpu-z 55-60watts
temps 55C
460 stock bios 8sols
gpu-z 25-30watts
temps 41C

4Gb or 8GB cards?

[IOTUSA]

nice! clocking ~18Sols/s on the same R9 390X that does ~22 using my own brew. your miner has one major advantage though...it actually comes up with solutions that are accepted by the pool  

An optimal solution should be able to do near 100 H/s (i.e. 200 sol/s) on a R9 380.  With 2 million records and 32-64 bytes of memory IO per record * 10 rounds = 1.28GB of memory bandwidth per hash.  The R9 380 with a memory clock of 1.5Ghz has 192GB/s of bandwidth.  The hard part is making full use of that bandwidth when you have small random IO.  Ethash is just at the fringe of being able to maximize memory bandwidth with 128-byte random reads.  Since an equihash solver requires a lot of small reads and writes, the algorithm and data structures need to be optimized for the GPU memory architecture.  i.e. 8 x 32-bit channels working with 2K pages and a 256-byte stride per wavefront.

I expect by the 28th we'll see GPU miners doing 50-60 sol/s, and close to 100 sol/s a month from now.  Hopefully one of the faster ones will be mine. :-)

Jtoomin (fastest known GPU miner)  claim to have done just that. Optimized the memory handling.

[snakey]

nice! clocking ~18Sols/s on the same R9 390X that does ~22 using my own brew. your miner has one major advantage though...it actually comes up with solutions that are accepted by the pool  

An optimal solution should be able to do near 100 H/s (i.e. 200 sol/s) on a R9 380.  With 2 million records and 32-64 bytes of memory IO per record * 10 rounds = 1.28GB of memory bandwidth per hash.  The R9 380 with a memory clock of 1.5Ghz has 192GB/s of bandwidth.  The hard part is making full use of that bandwidth when you have small random IO.  Ethash is just at the fringe of being able to maximize memory bandwidth with 128-byte random reads.  Since an equihash solver requires a lot of small reads and writes, the algorithm and data structures need to be optimized for the GPU memory architecture.  i.e. 8 x 32-bit channels working with 2K pages and a 256-byte stride per wavefront.

I expect by the 28th we'll see GPU miners doing 50-60 sol/s, and close to 100 sol/s a month from now.  Hopefully one of the faster ones will be mine. :-)

So its deff H/s not Sol/S to go off?

Everyone was telling me its the same thing or to go off Sol/s?

Is it better to wait off on these cards to see if some updates work better on other cards first?

[snakey]

IOTUSA - Yes they are claiming to get 40Sol/s

[xfunk]

Sorry for noob question, but could you give a little bit more detailed instructions how to build it from source? i'm on ubuntu 14.04, want to try miner on my current setup. Thanks in advance.

[nerdralph]

nice! clocking ~18Sols/s on the same R9 390X that does ~22 using my own brew. your miner has one major advantage though...it actually comes up with solutions that are accepted by the pool  

An optimal solution should be able to do near 100 H/s (i.e. 200 sol/s) on a R9 380.  With 2 million records and 32-64 bytes of memory IO per record * 10 rounds = 1.28GB of memory bandwidth per hash.  The R9 380 with a memory clock of 1.5Ghz has 192GB/s of bandwidth.  The hard part is making full use of that bandwidth when you have small random IO.  Ethash is just at the fringe of being able to maximize memory bandwidth with 128-byte random reads.  Since an equihash solver requires a lot of small reads and writes, the algorithm and data structures need to be optimized for the GPU memory architecture.  i.e. 8 x 32-bit channels working with 2K pages and a 256-byte stride per wavefront.

I expect by the 28th we'll see GPU miners doing 50-60 sol/s, and close to 100 sol/s a month from now.  Hopefully one of the faster ones will be mine. :-)

Jtoomin (fastest known GPU miner)  claim to have done just that. Optimized the memory handling.

40sol/s for a R9 390 is still on the slow side...

[nerdralph]

nice! clocking ~18Sols/s on the same R9 390X that does ~22 using my own brew. your miner has one major advantage though...it actually comes up with solutions that are accepted by the pool  

An optimal solution should be able to do near 100 H/s (i.e. 200 sol/s) on a R9 380.  With 2 million records and 32-64 bytes of memory IO per record * 10 rounds = 1.28GB of memory bandwidth per hash.  The R9 380 with a memory clock of 1.5Ghz has 192GB/s of bandwidth.  The hard part is making full use of that bandwidth when you have small random IO.  Ethash is just at the fringe of being able to maximize memory bandwidth with 128-byte random reads.  Since an equihash solver requires a lot of small reads and writes, the algorithm and data structures need to be optimized for the GPU memory architecture.  i.e. 8 x 32-bit channels working with 2K pages and a 256-byte stride per wavefront.

I expect by the 28th we'll see GPU miners doing 50-60 sol/s, and close to 100 sol/s a month from now.  Hopefully one of the faster ones will be mine. :-)

So its deff H/s not Sol/S to go off?

Everyone was telling me its the same thing or to go off Sol/s?

Is it better to wait off on these cards to see if some updates work better on other cards first?

Wagner's algorithm produces 2 solutions to the generalized birthday problem.  Each equihash run produces 2 solutions, so 1 equihash/s = 2 solutions/s.

If you want to know the details, read the paper.  It's a heavy read though.  It wasn't until the 2nd or 3rd time I read it that I started to really understand the algorithm enough to start thinking about coding.
https://www.internetsociety.org/sites/default/files/blogs-media/equihash-asymmetric-proof-of-work-based-generalized-birthday-problem.pdf

[xleejohnx]

470 stock bios 14-15sols
470 modded bios 17-18sols
gpu-z 55-60watts
temps 55C
460 stock bios 8sols
gpu-z 25-30watts
temps 41C

4Gb or 8GB cards?

all 4GB cards

[snakey]

nice! clocking ~18Sols/s on the same R9 390X that does ~22 using my own brew. your miner has one major advantage though...it actually comes up with solutions that are accepted by the pool  

An optimal solution should be able to do near 100 H/s (i.e. 200 sol/s) on a R9 380.  With 2 million records and 32-64 bytes of memory IO per record * 10 rounds = 1.28GB of memory bandwidth per hash.  The R9 380 with a memory clock of 1.5Ghz has 192GB/s of bandwidth.  The hard part is making full use of that bandwidth when you have small random IO.  Ethash is just at the fringe of being able to maximize memory bandwidth with 128-byte random reads.  Since an equihash solver requires a lot of small reads and writes, the algorithm and data structures need to be optimized for the GPU memory architecture.  i.e. 8 x 32-bit channels working with 2K pages and a 256-byte stride per wavefront.

I expect by the 28th we'll see GPU miners doing 50-60 sol/s, and close to 100 sol/s a month from now.  Hopefully one of the faster ones will be mine. :-)

So its deff H/s not Sol/S to go off?

Everyone was telling me its the same thing or to go off Sol/s?

Is it better to wait off on these cards to see if some updates work better on other cards first?

Wagner's algorithm produces 2 solutions to the generalized birthday problem.  Each equihash run produces 2 solutions, so 1 equihash/s = 2 solutions/s.

If you want to know the details, read the paper.  It's a heavy read though.  It wasn't until the 2nd or 3rd time I read it that I started to really understand the algorithm enough to start thinking about coding.
https://www.internetsociety.org/sites/default/files/blogs-media/equihash-asymmetric-proof-of-work-based-generalized-birthday-problem.pdf

Thank you 

[snakey]

470 stock bios 14-15sols
470 modded bios 17-18sols
gpu-z 55-60watts
temps 55C
460 stock bios 8sols
gpu-z 25-30watts
temps 41C

4Gb or 8GB cards?

all 4GB cards

Thanks

[Tmdz]

nice! clocking ~18Sols/s on the same R9 390X that does ~22 using my own brew. your miner has one major advantage though...it actually comes up with solutions that are accepted by the pool  

An optimal solution should be able to do near 100 H/s (i.e. 200 sol/s) on a R9 380.  With 2 million records and 32-64 bytes of memory IO per record * 10 rounds = 1.28GB of memory bandwidth per hash.  The R9 380 with a memory clock of 1.5Ghz has 192GB/s of bandwidth.  The hard part is making full use of that bandwidth when you have small random IO.  Ethash is just at the fringe of being able to maximize memory bandwidth with 128-byte random reads.  Since an equihash solver requires a lot of small reads and writes, the algorithm and data structures need to be optimized for the GPU memory architecture.  i.e. 8 x 32-bit channels working with 2K pages and a 256-byte stride per wavefront.

I expect by the 28th we'll see GPU miners doing 50-60 sol/s, and close to 100 sol/s a month from now.  Hopefully one of the faster ones will be mine. :-)

So its deff H/s not Sol/S to go off?

Everyone was telling me its the same thing or to go off Sol/s?

Is it better to wait off on these cards to see if some updates work better on other cards first?

Wagner's algorithm produces 2 solutions to the generalized birthday problem.  Each equihash run produces 2 solutions, so 1 equihash/s = 2 solutions/s.

If you want to know the details, read the paper.  It's a heavy read though.  It wasn't until the 2nd or 3rd time I read it that I started to really understand the algorithm enough to start thinking about coding.
https://www.internetsociety.org/sites/default/files/blogs-media/equihash-asymmetric-proof-of-work-based-generalized-birthday-problem.pdf

Thanks for the info.

 I see people swapping the terms all the time which makes it confusing to figure out which one they actually mean.

[Arceny]

Problems with building on 14.04:

Code:

[ 80%] Building CXX object CMakeFiles/zcashcpuclient.dir/zcashcpuclient.cpp.o
/root/src/xpmclient/zcashcpuclient.cpp:8:38: fatal error: /data/build/inc/printhex.h: No such file or directory
 #include "/data/build/inc/printhex.h"
                                      ^
compilation terminated.
make[2]: *** [CMakeFiles/zcashcpuclient.dir/zcashcpuclient.cpp.o] Error 1
make[1]: *** [CMakeFiles/zcashcpuclient.dir/all] Error 2
make: *** [all] Error 2

Can't locate printhex.h anywhere

[adaseb]

R9 7950 = 9H/S
R9 7970 = 10H/S
R9 280X= 11H/S
R9 290 = 17H/S

[snakey]

R9 7950 = 9H/S
R9 7970 = 10H/S
R9 280X= 11H/S
R9 290 = 17H/S

Hi, is that deff H/S or Sol/S?

Thanks

[adaseb]

Whichever the program displays, I forgot. Maybe its Sol/s

[IOTUSA]

Whichever the program displays, I forgot. Maybe its Sol/s

They are the same. It's preferred to refer to it as Sols or Sol/s to avoid confusion.

[adaseb]

So does it work on any other pools like Nicehash? Suprnova? Or is this Getwork/Getblocktemplate and not Stratum yet