Topic: Wolf's XMR/BCN/DSH CPUMiner - 2x speed compared to LucasJones' - NEW 06/20/2014 - page 17. (Read 547140 times)

Curl 7.31.1 compiled with :
./configure --disable-shared --enable-static --prefix=/usr/local --disable-ldap --disable-sspi
make -j 4;make install;

miner still errors out:
./autogen.sh
./configure CFLAGS="-static'
checking for the version of libcurl... 7.37.1
checking for libcurl >= version 7.15.2... yes
checking whether libcurl is usable... no
configure: error: Missing required libcurl >= 7.15.2

Any ideas ?

the static binary can include everything.
did you try:

./configure --enable-static

or

CFLAGS='-static'

?

I know it is, is there any other way to get it to run on centos ? If i compile on ubuntu and run on centos it complains about glibc version (2.14 required, centos runs 2.12). Any other way ? Is it possible to include glibc in the static binary ?

EDIT : Removing MADV_HUGEPAGE from cpu-miner.c fixes the problem but with a significant loss in speed. I need the extra speed!!

You can't, because CentOS is garbage. I had to add in a flag to disable some optimizations for that shit OS.

really, because 10,000/528 = ~19.  19 is a "few machines?"

edit: and that diff is pathetically low.  444 is for a pentium 2 or someshit.

To answer my own question, it appears as though the improvement is noticed only when the server has a large number of cpus / cache etc.
Older machines seem to perform better with a larger amount of threads, even when cpu count is lower.

I'm breaking > 10 000 H/s without even breaking a sweat across a few machines.


Thank you

Why does it automatically set thread count to ~cpu count?

Using your formula:

[2014-08-03 13:19:33] accepted: 12/12 (100.00%), 514.93 H/s at diff 444 (yay!!!)
[2014-08-03 13:19:34] accepted: 13/13 (100.00%), 513.02 H/s at diff 444 (yay!!!)
[2014-08-03 13:19:35] accepted: 14/14 (100.00%), 528.15 H/s at diff 444 (yay!!!)
[2014-08-03 13:19:35] accepted: 15/15 (100.00%), 525.17 H/s at diff 444 (yay!!!)

Perhaps this should be added to the original post? Its a BIG improvement.

Any updated miners ? The GPU now seems to have a massive advantage over CPUs now on monero

Don't set threads equal to number of CPU threads, that is wrong.
The general rule is threads = floor(L3 cache / 1024)
Use this: num_threads=$(($(cat /proc/cpuinfo | grep "cache size" | uniq | cut -d":" -f2 | cut -d" " -f2)/1024))
Should be 12 for X5650

How though? I've tried on a variety of *very expensive* hardware from Dell.
Each has 100gigs of ram, 24 cpus, etc..

Ubuntu 14.04:

# sysctl -w vm.nr_hugepages=72  (confirmed that it is set)
# git clone https://github.com/wolf9466/cpuminer-multi.git
# cd cpuminer-multi
# ./autogen.sh
# ./configure CFLAGS="-march=native"
# make
# make install
# screen -d -m -S minerd /usr/local/bin/minerd -o stratum+tcp://mro.pool.minergate.com:45560 -p x -u x
# screen -r (after a few hours of running)
[2014-08-03 06:05:25] accepted: 3952/7916 (49.92%), 288.94 H/s at diff 444 (yay!!!)

example of hardware: 24 CPUs - Intel(R) Xeon(R) CPU E5-2420 0 @ 1.90GHz
another example:       24 CPUs - Intel(R) Xeon(R) CPU X5650  @ 2.67GHz

You should be getting atleast 400 H/s out of those CPUs (mine are hashing at 451 H/s)

Hello all,

Has anyone successfully compiled Wolfs cpuminer under centOS 6.4?

I tried for a couple hours yesterday to make it work, and even removed some of the values under the makefile but can't seem to get it to compile.

I'm stuck on this error..

./cpu-miner.c:1065: error: ‘MADV_HUGEPAGE’ undeclared (first use in this function)
./cpu-miner.c:1065: error: (Each undeclared identifier is reported only once
./cpu-miner.c:1065: error: for each function it appears in.)

Any help would be greatly appreciated!

Going to check this out as well, running it without the optimizations.

Btw.. Just thought i'd mention this:

Latest CPUMiner compiled from git: (24 x Intel(R) Xeon(R) CPU X5650 @ 2.67GHz)

Over a 20 minute duration:

[2014-08-02 20:25:25] accepted: 191/234 (81.62%), 230.74 H/s at diff 980 (yay!!!)
[2014-08-02 20:25:29] accepted: 192/235 (81.70%), 230.63 H/s at diff 980 (yay!!!)
[2014-08-02 20:25:29] accepted: 193/236 (81.78%), 230.61 H/s at diff 980 (yay!!!)
[2014-08-02 20:25:31] accepted: 194/237 (81.86%), 230.16 H/s at diff 980 (yay!!!)
[2014-08-02 20:25:40] accepted: 195/238 (81.93%), 231.01 H/s at diff 980 (yay!!!)

Then i removed all of the "optimizations" from the Makefile:

From:

am__append_3 = -Ofast -flto -fuse-linker-plugin -funroll-loops -fvariable-expansion-in-unroller -ftree-loop-if-convert-stores -fmerg
e-all-constants -fbranch-target-load-optimize2 -fsched2-use-superblocks -maes

to :

am__append_3 = -maes

The result?

[2014-08-02 20:26:44] accepted: 189/231 (81.82%), 289.83 H/s at diff 1384 (yay!!!)
[2014-08-02 20:26:45] accepted: 190/232 (81.90%), 290.09 H/s at diff 1384 (yay!!!)
[2014-08-02 20:27:03] accepted: 191/233 (81.97%), 291.41 H/s at diff 1384 (yay!!!)
[2014-08-02 20:27:07] accepted: 192/234 (82.05%), 290.47 H/s at diff 1384 (yay!!!)

There are numerous posts/warnings on-line about over-aggressive compile-time "optimizations".
Just thought I'd post my results/findings which concur.

Regards,
Cami

Have someone compiled it on OSX?

It's quite a difference. I have 8 of those monster servers that are freshly installed (+updated) with Ubuntu 14.04.
They are all identical and unused (no running services, crons etc..) but the results are rather inconsistent across each so perhaps something else could be influencing things.
I'll dig a bit deeper and see if anything comes of it.