Wolf's XMR/BCN/DSH CPUMiner - 2x speed compared to LucasJones' - NEW 06/20/2014 - page 20.

Wolf0

member

Activity: 81

Merit: 1002

It was only the wind.

Quote from: AceCobra1 on June 18, 2014, 04:43:43 PM

Quote from: Wolf0 on June 18, 2014, 03:41:16 PM

Quote from: AceCobra1 on June 18, 2014, 01:11:06 PM

Just a quick check and I am getting roughly 220 KH/s with my QA82 ES Xeon processor with 12 GB of RAM... would more RAM make any difference?

No, but try a different number of threads.

Quote from: Alpinist on June 18, 2014, 02:37:46 PM

Can I run this miner to merge mine on Minergate?

Yes.

Quote from: bitspender on June 18, 2014, 01:40:26 PM

Another question, i'm running multiple VPS.

When i use on Ubuntu, i get around 100 H/s
On the same hardware, but running windows image, i get 280 ??

You're probably forgetting hugepages.

Mining with 9 threads for some reason produces 284 hash compared to say 16.... why would that be the case? Take it hyperthreading makes no difference? Also, in your front page, you said the Amazon EC2 manages 1.2KH/s.... which CPUs would those be ?!?!

Hyperthreading is pretty much useless here. Yep, and those are dual Xeons that cost $6500+ each.

billotronic

legendary

Activity: 1610

Merit: 1000

Crackpot Idealist

This might vary a little depending on what ubuntu version you are using and ffs I have not run this in a while so it also might have a few un-needed steps but it should get you started. This is tested to work for ubuntu 12.04 x64 with AES capable CPU. This is all done in your terminal. The files are downloaded to your home folder (ie /home/cpuminer-multi)

First your dependencies:

Code:

sudo add-apt-repository ppa:costamagnagianfranco/autoconf -y
sudo apt-get update
sudo apt-get install build-essential screen automake m4 openssl libssl-dev git libjson0 libjson0-dev libcurl4-openssl-dev autoconf python-software-properties -y

Next get the miner source and compile:

Code:

git clone https://github.com/wolf9466/cpuminer-multi.git
cd cpuminer-multi
./autogen.sh
./configure CFLAGS="-march=native"
make

Optimize system for mining

Code:

threads=`nproc`
sub="1"
mul="3"
thr=$(($threads - $sub))
thruse=$(($thr * 3))
sudo sysctl -w vm.nr_hugepages=$thruse

Start your miner with screen

Code:

sudo screen ./minerd -a cryptonight -o stratum+tcp://mine.moneropool.com:3333 -u 485VWHzZMe8hGJCXDArFVf1KUqjbwUnKqCEfPTzcU8zt6geeCMaaXECEMYsxFwQdgKi9LVX3HNMb1FQoMLKpbFRE5rUGCNe -p x

sleepdog

full member

Activity: 183

Merit: 100

Can anyone link me to a guide suitable for a Linux beginner who doesn't know his github from his twatspoke on how to obtain and compile this miner, starting from a fresh ubuntu install?

I'd like to run some side by side comparisons.

Wolf0

member

Activity: 81

Merit: 1002

It was only the wind.

Quote from: Alpinist on June 18, 2014, 04:23:06 PM

What .bat should I write to merge mine on MinerGate?
Did somebody do that?

The Fantomcoin website has instructions, I think.

cayars

full member

Activity: 168

Merit: 100

Quote from: Wolf0 on July 13, 2014, 10:16:39 AM

Quote from: cayars on July 13, 2014, 09:11:27 AM

Quote from: Wolf0 on July 13, 2014, 08:15:28 AM

CN is not only CPU mining by a long shot. GPUs will be able to do better than CPUs - in fact, they already do. BBR, on the other hand, might be actually GPU-resistant. That totally fucked blockchain-in-mah-scratchpad thing causes an ouch slow copy from host memory to GPU memory very often. On top of that, I don't think you can interrupt kernels sometimes - so you'd have to wait for the kernel to finish wasting its time hashing old data.

Nope BBR is not GPU resistant and is being mined with GPUs.

Yes, it's being mined with GPUs. But the GPU implementation is slower than a good CPU one.

Let's just say I'll respectfully disagree.

cayars

full member

Activity: 168

Merit: 100

Quote from: Wolf0 on July 13, 2014, 08:15:28 AM

CN is not only CPU mining by a long shot. GPUs will be able to do better than CPUs - in fact, they already do. BBR, on the other hand, might be actually GPU-resistant. That totally fucked blockchain-in-mah-scratchpad thing causes an ouch slow copy from host memory to GPU memory very often. On top of that, I don't think you can interrupt kernels sometimes - so you'd have to wait for the kernel to finish wasting its time hashing old data.

Nope BBR is not GPU resistant and is being mined with GPUs.

Wolf0

member

Activity: 81

Merit: 1002

It was only the wind.

Quote from: AceCobra1 on June 18, 2014, 01:11:06 PM

Just a quick check and I am getting roughly 220 KH/s with my QA82 ES Xeon processor with 12 GB of RAM... would more RAM make any difference?

No, but try a different number of threads.

Quote from: Alpinist on June 18, 2014, 02:37:46 PM

Can I run this miner to merge mine on Minergate?

Yes.

Quote from: bitspender on June 18, 2014, 01:40:26 PM

Another question, i'm running multiple VPS.

When i use on Ubuntu, i get around 100 H/s
On the same hardware, but running windows image, i get 280 ??

You're probably forgetting hugepages.

Rytir_fik

full member

Activity: 139

Merit: 100

Quote from: Wolf0 on July 13, 2014, 04:33:17 AM

Quote from: Rytir_fik on July 13, 2014, 03:31:17 AM

Quote from: ?? on ??

Quote from: primer- on July 12, 2014, 04:28:21 PM

Quote from: Wolf0 on July 12, 2014, 04:25:15 PM

Oh, I see. Odd... maybe the older CPUs did hyperthreading better?

Maybe. Newer CPUs definitely dont like the default -t option.

I'm also getting a %10-%20 difference on a i7 4770 (210-230 when using -t 7 compared to 265-280 when using -t 4).

in one i5 laptop I get more hashes with -t 1 than with -t 3, for aes-ni cpus less threads means more performance.

Of course it could be correct. In case you have only 3MB cache memory on your processor then the -t 1 will be faster (see my previous post with explanation). How much memory your processor has you can check eg. here: http://ark.intel.com/products/family/75024/4th-Generation-Intel-Core-i5-Processors#@All

Makes a lot of sense. There's no reason why you couldn't run more threads, but it would run a lot slower since it has to traipse out to main memory for the scratchpad all the damned time... but I wonder if the hardware prefetcher is smart enough to get that it should probably stuff the whole scratchpad in cache?

I do not know how exactly it works but this is one of the reasons why CryptoNight is declared as Only CPU-mining & ASIC-resistant. I am noob, I posted here information which I got from devs of BCN as you can see my Question on Github (because I thought that this is bug) https://github.com/amjuarez/bytecoin/issues/23.
.... and Wolfi thx for the miner in my case of i5-2500K @ 4.3 is about 15% faster (v. 05-30-2014) than Minergare client (v3.0) (on Minergate pool) Wink

Rytir_fik

full member

Activity: 139

Merit: 100

Quote from: ?? on ??

Quote from: primer- on July 12, 2014, 04:28:21 PM

Quote from: Wolf0 on July 12, 2014, 04:25:15 PM

Oh, I see. Odd... maybe the older CPUs did hyperthreading better?

Maybe. Newer CPUs definitely dont like the default -t option.

I'm also getting a %10-%20 difference on a i7 4770 (210-230 when using -t 7 compared to 265-280 when using -t 4).

in one i5 laptop I get more hashes with -t 1 than with -t 3, for aes-ni cpus less threads means more performance.

Of course it could be correct. In case you have only 3MB cache memory on your processor then the -t 1 will be faster (see my previous post with explanation). How much memory your processor has you can check eg. here: http://ark.intel.com/products/family/75024/4th-Generation-Intel-Core-i5-Processors#@All

Rytir_fik

full member

Activity: 139

Merit: 100

Quote from: primer- on July 12, 2014, 04:28:21 PM

Quote from: Wolf0 on July 12, 2014, 04:25:15 PM

Oh, I see. Odd... maybe the older CPUs did hyperthreading better?

Maybe. Newer CPUs definitely dont like the default -t option.

I'm also getting a %10-%20 difference on a i7 4770 (210-230 when using -t 7 compared to 265-280 when using -t 4).

In case you use it for CryptoNight hash algorithm then the answer could be this. CryptoNight hash algorithm requires 2 MB of cache for each of the mining threads. In my case of i5-2500K with 4 cores and 6MB I am able to use only -t 3 properly, so any additional thread (over 3) will only slow the process down.

primer-

legendary

Activity: 1092

Merit: 1000

Quote from: Wolf0 on July 12, 2014, 04:25:15 PM

Oh, I see. Odd... maybe the older CPUs did hyperthreading better?

Maybe. Newer CPUs definitely dont like the default -t option.

I'm also getting a %10-%20 difference on a i7 4770 (210-230 when using -t 7 compared to 265-280 when using -t 4).

primer-

legendary

Activity: 1092

Merit: 1000

Quote from: Wolf0 on July 12, 2014, 04:04:51 PM

Quote from: primer- on July 12, 2014, 04:50:34 AM

I'm getting different results when running the miner on all threads vs only physical cores.

On newer CPUs i'm getting up to 50% more power when running only on physical cpu cores. On older CPUs (also with AES) i get 50% less when using only physical cores.
Why is this ?

That isn't obvious? Newer CPUs tend to be faster, and I don't mean in terms of clock speeds. They improve shit like out of order execution, speculative execution, branch prediction...

I think you misunderstood my post, example below :

dual E5-2620 (12 cores/24 threads)

-t 23 (default when run without -t tag) does around 310H/s
-t 12 (physical cores) does around 390H/s

Dual E5620 (8 cores/16 threads)
-t 15 (default when run without -t tag) does around 150H/s
-t 8 (physical cores) does around 100H/s

EDIT:Dual E5620 figures might not be correct, i performed tests earlier today (couple hours) and no longer have access. Difference was 50%

primer-

legendary

Activity: 1092

Merit: 1000

I'm getting different results when running the miner on all threads vs only physical cores.

On newer CPUs i'm getting up to 50% more power when running only on physical cpu cores. On older CPUs (also with AES) i get 50% less when using only physical cores.
Why is this ?

ajiekceu4

newbie

Activity: 38

Merit: 0

Download links broken. (ottrbutt.com and 46.105.182.112 didn't response).

Wolf0

member

Activity: 81

Merit: 1002

It was only the wind.

Quote from: HardwarePal on June 18, 2014, 04:25:43 AM

Em ... Im a newb on that Sad

No, I mean, I think there are Windows binaries on the github.

sleepdog

full member

Activity: 183

Merit: 100

I meant that if yours is quicker than his on AES-NI systems it must be down to that part of the code that uses AES-NI.

I'm running 2008 R2 Enterprise.

Wolf0

member

Activity: 81

Merit: 1002

It was only the wind.

Quote from: HardwarePal on June 18, 2014, 04:21:01 AM

Any links to Lucas software for NON AES-NI for windows other then the github?

Why? What's wrong with the github?

sleepdog

full member

Activity: 183

Merit: 100

Similar results here. I ran some tests with a dual E5520 server.

cpu-multi / LucasJones (these are the same thing, right?) v2.3.3 gets max 148 H/s and is pretty much always over 140.

Wolf's newly provided non-AES version gets 118 H/s max, with a range of 100 - 118.

Looks like any advantages of Wolf's miner come from the AES-NI code.

NeuroticFish

legendary

Activity: 3668

Merit: 6382

Looking for campaign manager? Contact icopress!

In case that this is of any interest:

I did some tests with W0lf's latest non-AES miner (yeah, unfortunately I cannot "install" AES for now).

On my I7-920, -t 8, it shows an average of 60 H/s.
Cpuminer-multi (1.0.3) shows about 70 H/s.

The numbers on the pool have a big variation area because I get - quite often - connection errors, so I don't have a reliable way to compare.

Is there any chance that the miner hash reporting is wrong?
Is there anything I can do to avoid the connection problems?
Are there some details I am missing?

----------

I've read your explanation about 32bit vs 64bit on XMR thread... hats off to you, sir.

dragonmike

hero member

Activity: 1274

Merit: 556

Probably a compatibility issue.
Testing on crypto-pool.fr gives more consistent results. Using -t 2 and -t 4 give similar result of between 40 and 50 H/s on the miner and the website... which is also consistent with the figure I would get from the Minergate client... except there I'd get 20% more from merged mining.

Topic: Wolf's XMR/BCN/DSH CPUMiner - 2x speed compared to LucasJones' - NEW 06/20/2014 - page 20. (Read 547105 times)