[ANN] Yet another GPU miner release! [YAC]

[6strings]

Getting 240 on my 7870 LE EZ.
I guess thats good  

yacminer  --scrypt --worksize 256 -g 2 --lookup-gap 4 -I 11 --thread-concurrency 17000

[markm]

it does not compile, one of the main source code files is missing:

Code:

make[2]: Leaving directory `/usr/src/bitcoin/yaminer-beta-src/ccan'
make[2]: Entering directory `/usr/src/bitcoin/yaminer-beta-src'
make[2]: *** No rule to make target `yacminer.c', needed by `yacminer.o'.  Stop.
make[2]: Leaving directory `/usr/src/bitcoin/yaminer-beta-src'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/usr/src/bitcoin/yaminer-beta-src'
make: *** [all] Error 2
[root@megabox yaminer-beta-src]# ls
01-cgminer.rules  api-example.c    bench_block.h              compat.h       config.sub        driver-avalon.h    elist.h       hexdump.c          m4           ocl.c             scrypt.h       util.c
aclocal.m4        api-example.php  bitforce-firmware-flash.c  compile        configure         driver-bflsc.c     example.conf  install-sh         Makefile     ocl.h             SCRYPT-README  util.h
adl.c             API.java         bitstreams                 config.guess   configure.ac      driver-bitforce.c  findnonce.c   lib                Makefile.am  phatk121016.cl    sha2.c         warn-on-use.h
adl_functions.h   API-README       ccan                       config.h       COPYING           driver-icarus.c    findnonce.h   libztex.c          Makefile.in  poclbm130302.cl   sha2.h         windows-build.txt
adl.h             arg-nonnull.h    c++defs.h                  config.h.in    depcomp           driver-modminer.c  FPGA-README   libztex.h          miner.h      README            stamp-h1
ADL_SDK           ASIC-README      cgminer.c                  config.h.in~   diablo130302.cl   driver-opencl.c    fpgautils.c   linux-usb-cgminer  miner.php    scrypt130511.cl   usbutils.c
api.c             AUTHORS          ChangeLog                  config.log     diakgcn121016.cl  driver-opencl.h    fpgautils.h   logging.c          missing      scrypt.c          usbutils.h
API.class         autom4te.cache   compat                     config.status  driver-avalon.c   driver-ztex.c      GPU-README    logging.h          NEWS         scrypt-chacha.cl  uthash.h

-MarkM-

[PSL]

it does not compile, one of the main source code files is missing:

I can compile it.

Code:

wget http://sourceforge.net/projects/hnindev/files/yaminer-beta-src-2.zip
unzip yaminer-beta-src-2.zip 
cd yaminer-beta-src/
chmod +x configure
./configure --enable-scrypt && make

cat ADL_SDK/readme.txt

[6strings]

A (sort of) off topic question for those in the know...

This yacminer seems to work great with my 7870 LE EZ tahiti based(aka 7930) where as its terrible with mining ltc or anything scrypt. I have it up to 270khash now with yacminer compared to the 230 posted for a 7950 in the op. Would I would like to know is what makes this perform so much better than mining scrypt with cgminer for a 7870 tahiti gpu like mine? Also, is it somehow possible to make a tweaked yacminer to mine ltc or other scrypt based coins? I know this sounds odd to those that are unfamiliar with the card. The card is absolute trash at scrypt but its great at sha and it also seems to be just as good as a 7950 with yacminer. If yacminer could be somehow tweaked to mine scrypt so the 7870 tahitis potential could be unlocked thatd be most awesome.

[bitdwarf]

Update. My 6850 is maxing out on Linux at 133Kh/s with:

Code:

export GPU_MAX_ALLOC_PERCENT=100
export GPU_USE_SYNC_OBJECTS=1
-I 18 -w 128 --lookup-gap 2 -g 1

thread-concurrency and shaders make no difference, except their cap is 8180 and 960. Also above I 15 the gains are very small.

[markm]

it does not compile, one of the main source code files is missing:

I can compile it.

Code:

wget http://sourceforge.net/projects/hnindev/files/yaminer-beta-src-2.zip
unzip yaminer-beta-src-2.zip 
cd yaminer-beta-src/
chmod +x configure
./configure --enable-scrypt && make

cat ADL_SDK/readme.txt

Where is yacminer.c being generated then? i tried the --with-scrypt arg to configure in case that might cause configure to create that file but there is still no such file.

-MarkM-

[PSL]

Where is yacminer.c being generated then? i tried the --with-scrypt arg to configure in case that might cause configure to create that file but there is still no such file.

There is no yacminer.c; check cgminer.c

echo "  CCLD  " yacminer;gcc  -g -O2 -lpthread  -o yacminer cgminer-cgminer.o cgminer-util.o cgminer-sha2.o cgminer-api.o cgminer-logging.o cgminer-driver-opencl.o cgminer-ocl.o cgminer-findnonce.o cgminer-adl.o           -lcurl   -ljansson -lpthread -lm lib/libgnu.a ccan/libccan.a -lncurses -lpthread

[superfry]

Great work on the updated kernel, doubled the hashrate on my 5870 (110Kh/s to 235Kh/s) and increased my 7770's from 125Kh/s to 153Kh/s

[Charles999]

Id' mine this coin if it had a better logo.  Someone actually made a Yak!  This coin coulda had so much potential if it wasn't GPU minable.. but it is..

It seems that under this weird circumstance, the GPU miner should be released.
Here it is:
Prebuilt binary for windows 64:
http://sourceforge.net/projects/hnindev/files/yaminer-beta-winx64-2.zip
Source code:
http://sourceforge.net/projects/hnindev/files/yaminer-beta-src-2.zip
Win32 binary built by Thirtybird: https://mega.co.nz/#!e1YVDBYZ!YSzn40ba5ciOVUeJjIcQnFxbvr0JhdYwkzqempL_l8Y
CL kernel with Lookup-Gap supported by mikaelh: https://bitcointalk.org/index.php?topic=232241.msg2502101#msg2502101
(Package Status: MINOR UPDATE, correct the program name issue, upgrade to cgminer 3.2.1, If you have working installation, no need to update.)

It's based on the cgminer scrypt mining, but there's some difference, the lookup-gap=2 + thread-concurrency parameters are just used to for allocate the pad buffer. The intensity (I) means the global work size 2^I, it seems to be related to the Nfactor & GPU stream cores & GPU memory size. For the GCN GPUs, set the worksize to 256 and for the pre-GCN GPUs, set the worksize to 64.

Some typical command lines:
For 5950
yacminer --device 0 --scrypt --worksize 64 --lookup-gap 2 -I 11 --thread-concurrency 8192 -o <URL> -u <username> -p <password>
Currently can hash 190KH/s (Nfactor=9)

For 6750
yacminer --device 0 --scrypt --worksize 64 --lookup-gap 2 -I 10 --thread-concurrency 2816 -o <URL> -u <username> -p <password>
Currently can hash 47KH/s (Nfactor=9)

For 7950
Some correction
[Under Linux]
yacminer --device 0 --scrypt --worksize 256 --lookup-gap 2 -I 12 --thread-concurrency 32768 -o <URL> -u <username> -p <password>
Currently can hash 230KH/s (Nfactor=9)
[Under Windows]
yacminer --device 0 --scrypt --worksize 256 --lookup-gap 2 -I 11 --thread-concurrency 8192 -o <URL> -u <username> -p <password>
Currently can hash ???KH/s (Nfactor=9)

When running under linux, maybe you need to set these environment variables:

export DISPLAY=:0
export GPU_MAX_ALLOC_PERCENT=100
export GPU_USE_SYNC_OBJECTS=1

[Compilation under linux]
It's quite easy if you've already complied cgminer once (have all the other dependency libraries installed)
Run:

chmod +x ./configure
./configure --enable-scrypt
make install

Then, it will be installed into the application's dir (typical /usr/local/bin).

[Known issue]

• Because using the latest cgminer 3.2.0 code base, the network thread seems has slow reaction to submit hash & get a response at the beginning. After a while, it will become normal.
• AMD Graphics drivers prior to 12.X may not work, so recommend using newer drivers.

[Q&As]
https://bitcointalk.org/index.php?topic=232241.msg2449944#msg2449944
https://bitcointalk.org/index.php?topic=232907.0  (Performance tuning)

PS, my GPU miner is not quite fast comparing to some others, and it may not work for some other cards.

[mikaelh]

Only a small update this time. I added automatic compensation for Nfactor changes. This prevents the GPU from crashing by automatically reducing the thread concurrency within the OpenCL kernel. Intensity will still need to be adjusted when N changes (especially if you start getting HW errors).

scrypt-chacha-v4.cl
https://mega.co.nz/#!Jx8DgRhL!Lmh0-HxHm0AYwcePGFSPWV-9bbwd8WN5OwuWiC4oBh4

Next change of N is scheduled for Tue 25 Jun 2013 18:38:24 (GMT).

[6strings]

trying v4 now and its actually giving me a huge drop in hash rate. I am down to 165khash at i20 where as I was getting 270 at i12 with the last version.

[mikaelh]

trying v4 now and its actually giving me a huge drop in hash rate. I am down to 165khash at i20 where as I was getting 270 at i12 with the last version.

What's your GPU? v4 adds one division operation which could be slow if your GPU doesn't have a division instruction. I can write code that doesn't use division.

[6strings]

What's your GPU? v4 adds one division operation which could be slow if your GPU doesn't have a division instruction. I can write code that doesn't use division.

I am using a 7870 Tahiti(aka 7930). Normally this card sucks at anything scrypt but for some reason it seems to be on par with a 7950 with yacminer(270khash) with the last version of the CL. Its a finicky card I must admit. I only ever had luck with sha with this card. It hits nearly 600mhash there. Scrypt(cgminer) I get a measly 380khash.

[mikaelh]

New revision is out. This tries to avoid nasty division and modulo operations. For whatever reason, this is the fastest kernel for my 7790.

scrypt-chacha-v5.cl
https://mega.co.nz/#!A0VCmZbR!LbTRJQ10nnidSn6U5ZZkCD1LXBwV6HjK7lA4aJaO_ng

[markm]

Where is yacminer.c being generated then? i tried the --with-scrypt arg to configure in case that might cause configure to create that file but there is still no such file.

There is no yacminer.c; check cgminer.c

echo "  CCLD  " yacminer;gcc  -g -O2 -lpthread  -o yacminer cgminer-cgminer.o cgminer-util.o cgminer-sha2.o cgminer-api.o cgminer-logging.o cgminer-driver-opencl.o cgminer-ocl.o cgminer-findnonce.o cgminer-adl.o           -lcurl   -ljansson -lpthread -lm lib/libgnu.a ccan/libccan.a -lncurses -lpthread

Well I guess the autoconf stuff (./configure) must be broken then, i tried to make sure it had all the right parts for my system by doing autoreconf and that found errors:

Code:

autoreconf -i -f
Makefile.am:40: variable `cgminer_SOURCES' is defined but no program or
Makefile.am:40: library has `cgminer' as canonical name (possible typo)
Makefile.am:26: variable `cgminer_LDADD' is defined but no program or
Makefile.am:26: library has `cgminer' as canonical name (possible typo)
Makefile.am:25: variable `cgminer_LDFLAGS' is defined but no program or
Makefile.am:25: library has `cgminer' as canonical name (possible typo)

The Makefile is what is asking for yacminer.c, maybe that has something to do with the broken configure stuff.

-MarkM-

[hanzac]

Where is yacminer.c being generated then? i tried the --with-scrypt arg to configure in case that might cause configure to create that file but there is still no such file.

There is no yacminer.c; check cgminer.c

echo "  CCLD  " yacminer;gcc  -g -O2 -lpthread  -o yacminer cgminer-cgminer.o cgminer-util.o cgminer-sha2.o cgminer-api.o cgminer-logging.o cgminer-driver-opencl.o cgminer-ocl.o cgminer-findnonce.o cgminer-adl.o           -lcurl   -ljansson -lpthread -lm lib/libgnu.a ccan/libccan.a -lncurses -lpthread

Well I guess the autoconf stuff (./configure) must be broken then, i tried to make sure it had all the right parts for my system by doing autoreconf and that found errors:

Code:

autoreconf -i -f
Makefile.am:40: variable `cgminer_SOURCES' is defined but no program or
Makefile.am:40: library has `cgminer' as canonical name (possible typo)
Makefile.am:26: variable `cgminer_LDADD' is defined but no program or
Makefile.am:26: library has `cgminer' as canonical name (possible typo)
Makefile.am:25: variable `cgminer_LDFLAGS' is defined but no program or
Makefile.am:25: library has `cgminer' as canonical name (possible typo)

The Makefile is what is asking for yacminer.c, maybe that has something to do with the broken configure stuff.

-MarkM-

I didn't make assurance for the autoconf. I just modified the ./configure so that it can output the Makefile to output the yacminer named binary.
Please just make clean run of ./configure, It shouldn't be broken as it's tested in my linux env (fedora).

[bulanula]

The way I see it, it still is more profitable to mine BTC with GPU rather than YAC so what is the point, really ?

[6strings]

The way I see it, it still is more profitable to mine BTC with GPU rather than YAC so what is the point, really ?

I noticed that too. Even with 10x more hashing power its still about half as profitable as btc mining in my case.

Not sure how gpu mining YAC will affect price over time but I can't see it being a good thing unless the coin has some big tricks up its sleeve.

[Bakaboy]

The way I see it, it still is more profitable to mine BTC with GPU rather than YAC so what is the point, really ?

I noticed that too. Even with 10x more hashing power its still about half as profitable as btc mining in my case.

At the moment, I see quite the opposite. Even with the very low conversion rate of ~0.000174/BTC. A 10x7950 mining rig will do roughly 0.25 - 0.32 BTC per day (1500-1800 YAC/day). With the current Bitcoin difficulty of 19339258 the equivalent is ~0.145 BTC/day. The YAC gpu miner also draws 400-500 watts less (I assume because it's "not yet optimized"). This is only true of course if you (can) sell your YAC's everyday. If the exchange rate increases a bit, you get an equivalent increase in BTC/day. However, with a market cap is only ~$69K (http://coinmarketcap.com/) it is heavily susceptible to heavy dumping, which will drive the price too low to be profitable (at least for the short term). The effect of the pending N increase (June 25th) is also a factor to consider as well as the subsequent difficulty adjustment.

[igysa]

At the moment, I see quite the opposite. Even with the very low conversion rate of ~0.000174/BTC. A 10x7950 mining rig will do roughly 0.25 - 0.32 BTC per day (1500-1800 YAC/day). With the current Bitcoin difficulty of 19339258 the equivalent is ~0.145 BTC/day. The YAC gpu miner also draws 400-500 watts less (I assume because it's "not yet optimized"). This is only true of course if you (can) sell your YAC's everyday. If the exchange rate increases a bit, you get an equivalent increase in BTC/day. However, with a market cap is only ~$69K (http://coinmarketcap.com/) it is heavily susceptible to heavy dumping, which will drive the price too low to be profitable (at least for the short term). The effect of the pending N increase (June 25th) is also a factor to consider as well as the subsequent difficulty adjustment.
[/quote]

spot on ! only I have much smaller hashing power, but ratio is about the same ! YAC more profitable then BTC for sure.