search

fireice-uk / xmr-stak

Free Monero RandomX Miner and unified CryptoNight miner
GNU General Public License v3.0
4.06k stars 1.79k forks source link

I'd like the version for the xeon phi also. #1242

Closed ardatious closed 6 years ago

ardatious commented 6 years ago

Please provide as much as possible information to reproduce the issue.

Basic information

Compile issues

Issue with the execution

AMD OpenCl issue

  run `clinfo` and add the output here

Stability issue

komawoyo commented 6 years ago

with a bit of googling you can find a software for xeon phi specific, I would rather have this project focus on the current backend

psychocrypt commented 6 years ago

Which xeon phi do you mean. KNL or older?

ardatious commented 6 years ago

The KNL 7210 in my case. I am using the SuperMicro 2u4n server with four phi nodes. I am using it with lukminer which is a great piece of software. However someone on another thread here (closed now) said he used the stak software on it with 7250 and got 22kh/s with 7250s. This is fast. I'd like to find out how he got it working, but I don't have his email and the thread is closed. I haven't run stak on it, but cpuminer-multi only sees 16 processors and runs at 373h/s.

I'm wondering if he used opencl or did something else. The 7210s are slower than the 7250s, but not by 2 to 1.

In response to the other comment, I think the idea of open source development is that we can use all the past contributions to it and contribute our own, as long as we are willing to use future users to do the same. I would not mind starting a new fork if I can get this running with the phi.

After I looked at the cpuminer-multi code, I saw several things that would benefit from optimization. Then I found stak, which has many optimizations in place.

I have 26 more 4n2u servers on order and would like to get the best profitability out of each one. If the stak software will help me do that, I would like to use it.

I am also mining on old Dell servers and I will be testing xmr-stak there very soon.

The newest intel servers we sell in our other business have up to 18 cores per CPU, but they are too expensive for mining, I think.

on the SuperMicro server, we are booting ubuntu 16.04.4 on the xeon phi cpu. it is not a coprocessor. There is no other cpu chip in the system.

with lukminer mining to xmr on nicehash, we generated $14.50 gross revenue in one 24 hour period, at a cost of about $3.50 in electricity at 11 cents a kwh. We are planning to put in our own natural-gas generator which should get the electric cost below 5 cent per kwh.

Since the electric cost is basically fixed, even a small increase in revenue makes a big difference in the profit.

On Thu, Mar 29, 2018 at 12:40 AM, psychocrypt notifications@github.com wrote:

Which xeon phi do you mean. KNL or older?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fireice-uk/xmr-stak/issues/1242#issuecomment-377127780, or mute the thread https://github.com/notifications/unsubscribe-auth/AjhwB_B5tnATb1fSPj53sJ-3KsFZtE9fks5tjHPmgaJpZM4S92Lg .

psychocrypt commented 6 years ago

Do you tried xmr-stak. KNL is also x86 therefore the miner supports the system without changes.

ardatious commented 6 years ago

I haven't tried it yet. I'm looking for the portable download for linux bu the latest release only has a zip file for windows.

so next i will try the linux install and compile route.

the one page that i have found for doing that skips the step of gitting the project, but I think I know how to do that.

i am not sure how it is going to behave when I do what you suggest, but I will try it and report back.

The person who said he got it working at high speed is J. Daniel Medeiros (I hope I spelled that right) but I don't have his email and the thread with his post is closed. I don't mind floundering around to do the same thing, that is what we all did before we had the web. (I have been programming since punched card days.) I approached him on FB but no response yet.

On Thu, Mar 29, 2018 at 4:09 PM, psychocrypt notifications@github.com wrote:

Do you tried xmr-stak. KNL is also x86 therefore the miner supports the system without changes.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fireice-uk/xmr-stak/issues/1242#issuecomment-377373375, or mute the thread https://github.com/notifications/unsubscribe-auth/AjhwB6je3LYv2uM2cuGpnWP1taHMCIFjks5tjU2fgaJpZM4S92Lg .

ardatious commented 6 years ago

i had problems in make. im running on a fresh install of linux 16.04.4

i used the instructions from this page:

https://github.com/fireice-uk/xmr-stak/blob/master/doc/compile_Linux.md

specifically

Ubuntu / Debian

sudo apt install libmicrohttpd-dev libssl-dev cmake

build-essential libhwloc-dev git clone https://github.com/fireice-uk/xmr-stak.git mkdir xmr-stak/build cd xmr-stak/build cmake .. make install

with these modificatons:

sudo apt install git

cmake .. -DCUDA_ENABLE=OFF -DOpenCL_ENABLE=OFF

my system is a dell r410 poweredge server with two intel xeon cpus, no graphics cards.

cmake worked fine

however make generated 2.3 gazillion errors like:

/home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:788:26: note: ‘_mm_cvtsi128_si64’ declared here, later in the translation unit idx0 = _mm_cvtsi128_si64(ax0); ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:718:25: error: ‘_mm_cvtsi128_si64’ was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive] idx = _mm_cvtsi128_si64(c); \ ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:807:3: note: in expansion of macro ‘CN_STEP3’ CN_STEP3(ax1, bx1, cx1, l1, ptr1, idx1); ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:788:26: note: ‘_mm_cvtsi128_si64’ declared here, later in the translation unit idx0 = _mm_cvtsi128_si64(ax0); ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:718:25: error: ‘_mm_cvtsi128_si64’ was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive] idx = _mm_cvtsi128_si64(c); \ ^ im going to try it again and capture the output

I understand that many or most of your users are running windows, since most of the action is on the gpu(s).

besides the phi servers, I hope to run stak on 90 dell r410s I have as leftovers. lukminer runs on it, but is not really optimized for cpus.

I also have 64 d610 and d620 blade servers I want to run it on.

having fun....

On Thu, Mar 29, 2018 at 4:24 PM, Ardatious llc ardatious@gmail.com wrote:

I haven't tried it yet. I'm looking for the portable download for linux bu the latest release only has a zip file for windows.

so next i will try the linux install and compile route.

the one page that i have found for doing that skips the step of gitting the project, but I think I know how to do that.

i am not sure how it is going to behave when I do what you suggest, but I will try it and report back.

The person who said he got it working at high speed is J. Daniel Medeiros (I hope I spelled that right) but I don't have his email and the thread with his post is closed. I don't mind floundering around to do the same thing, that is what we all did before we had the web. (I have been programming since punched card days.) I approached him on FB but no response yet.

On Thu, Mar 29, 2018 at 4:09 PM, psychocrypt notifications@github.com wrote:

Do you tried xmr-stak. KNL is also x86 therefore the miner supports the system without changes.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fireice-uk/xmr-stak/issues/1242#issuecomment-377373375, or mute the thread https://github.com/notifications/unsubscribe-auth/AjhwB6je3LYv2uM2cuGpnWP1taHMCIFjks5tjU2fgaJpZM4S92Lg .

ardatious commented 6 years ago

ha ha first error message is 'you are trying to do a 32 bit build, i know this will all end in tears'

and it did

here is my cpu info for intel es5620

lscpu Architecture: i686 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 8 On-line CPU(s) list: 0-7 Thread(s) per core: 1 Core(s) per socket: 4 Socket(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 44 Model name: Intel(R) Xeon(R) CPU E5620 @ 2.40GHz Stepping: 2 CPU MHz: 2393.797 BogoMIPS: 4787.59 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 256K L3 cache: 12288K

On Thu, Mar 29, 2018 at 4:48 PM, Ardatious llc ardatious@gmail.com wrote:

i had problems in make. im running on a fresh install of linux 16.04.4

i used the instructions from this page:

https://github.com/fireice-uk/xmr-stak/blob/master/doc/compile_Linux.md

specifically

Ubuntu / Debian

sudo apt install libmicrohttpd-dev libssl-dev cmake build-essential libhwloc-dev
git clone https://github.com/fireice-uk/xmr-stak.git
mkdir xmr-stak/build
cd xmr-stak/build
cmake ..
make install

with these modificatons:

sudo apt install git

cmake .. -DCUDA_ENABLE=OFF -DOpenCL_ENABLE=OFF

my system is a dell r410 poweredge server with two intel xeon cpus, no graphics cards.

cmake worked fine

however make generated 2.3 gazillion errors like:

/home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:788:26: note: ‘_mm_cvtsi128_si64’ declared here, later in the translation unit idx0 = _mm_cvtsi128_si64(ax0); ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:718:25: error: ‘_mm_cvtsi128_si64’ was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive] idx = _mm_cvtsi128_si64(c); \ ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:807:3: note: in expansion of macro ‘CN_STEP3’ CN_STEP3(ax1, bx1, cx1, l1, ptr1, idx1); ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:788:26: note: ‘_mm_cvtsi128_si64’ declared here, later in the translation unit idx0 = _mm_cvtsi128_si64(ax0); ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:718:25: error: ‘_mm_cvtsi128_si64’ was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive] idx = _mm_cvtsi128_si64(c); \ ^ im going to try it again and capture the output

I understand that many or most of your users are running windows, since most of the action is on the gpu(s).

besides the phi servers, I hope to run stak on 90 dell r410s I have as leftovers. lukminer runs on it, but is not really optimized for cpus.

I also have 64 d610 and d620 blade servers I want to run it on.

having fun....

On Thu, Mar 29, 2018 at 4:24 PM, Ardatious llc ardatious@gmail.com wrote:

I haven't tried it yet. I'm looking for the portable download for linux bu the latest release only has a zip file for windows.

so next i will try the linux install and compile route.

the one page that i have found for doing that skips the step of gitting the project, but I think I know how to do that.

i am not sure how it is going to behave when I do what you suggest, but I will try it and report back.

The person who said he got it working at high speed is J. Daniel Medeiros (I hope I spelled that right) but I don't have his email and the thread with his post is closed. I don't mind floundering around to do the same thing, that is what we all did before we had the web. (I have been programming since punched card days.) I approached him on FB but no response yet.

On Thu, Mar 29, 2018 at 4:09 PM, psychocrypt notifications@github.com wrote:

Do you tried xmr-stak. KNL is also x86 therefore the miner supports the system without changes.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fireice-uk/xmr-stak/issues/1242#issuecomment-377373375, or mute the thread https://github.com/notifications/unsubscribe-auth/AjhwB6je3LYv2uM2cuGpnWP1taHMCIFjks5tjU2fgaJpZM4S92Lg .

ardatious commented 6 years ago

architecture i686 means I have 32 bit ubuntu installed. i don't know why it installed that instead of 64 bit. ill go try to figure that out. i did the install on this same machine

On Thu, Mar 29, 2018 at 5:05 PM, Ardatious llc ardatious@gmail.com wrote:

ha ha first error message is 'you are trying to do a 32 bit build, i know this will all end in tears'

and it did

here is my cpu info for intel es5620

lscpu Architecture: i686 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 8 On-line CPU(s) list: 0-7 Thread(s) per core: 1 Core(s) per socket: 4 Socket(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 44 Model name: Intel(R) Xeon(R) CPU E5620 @ 2.40GHz Stepping: 2 CPU MHz: 2393.797 BogoMIPS: 4787.59 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 256K L3 cache: 12288K

On Thu, Mar 29, 2018 at 4:48 PM, Ardatious llc ardatious@gmail.com wrote:

i had problems in make. im running on a fresh install of linux 16.04.4

i used the instructions from this page:

https://github.com/fireice-uk/xmr-stak/blob/master/doc/compile_Linux.md

specifically

Ubuntu / Debian

sudo apt install libmicrohttpd-dev libssl-dev cmake build-essential libhwloc-dev
git clone https://github.com/fireice-uk/xmr-stak.git
mkdir xmr-stak/build
cd xmr-stak/build
cmake ..
make install

with these modificatons:

sudo apt install git

cmake .. -DCUDA_ENABLE=OFF -DOpenCL_ENABLE=OFF

my system is a dell r410 poweredge server with two intel xeon cpus, no graphics cards.

cmake worked fine

however make generated 2.3 gazillion errors like:

/home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:788:26: note: ‘_mm_cvtsi128_si64’ declared here, later in the translation unit idx0 = _mm_cvtsi128_si64(ax0); ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:718:25: error: ‘_mm_cvtsi128_si64’ was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive] idx = _mm_cvtsi128_si64(c); \ ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:807:3: note: in expansion of macro ‘CN_STEP3’ CN_STEP3(ax1, bx1, cx1, l1, ptr1, idx1); ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:788:26: note: ‘_mm_cvtsi128_si64’ declared here, later in the translation unit idx0 = _mm_cvtsi128_si64(ax0); ^ /home/ardatious/xmr-stak/xmrstak/backend/cpu/crypto/cryptonight_aesni.h:718:25: error: ‘_mm_cvtsi128_si64’ was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive] idx = _mm_cvtsi128_si64(c); \ ^ im going to try it again and capture the output

I understand that many or most of your users are running windows, since most of the action is on the gpu(s).

besides the phi servers, I hope to run stak on 90 dell r410s I have as leftovers. lukminer runs on it, but is not really optimized for cpus.

I also have 64 d610 and d620 blade servers I want to run it on.

having fun....

On Thu, Mar 29, 2018 at 4:24 PM, Ardatious llc ardatious@gmail.com wrote:

I haven't tried it yet. I'm looking for the portable download for linux bu the latest release only has a zip file for windows.

so next i will try the linux install and compile route.

the one page that i have found for doing that skips the step of gitting the project, but I think I know how to do that.

i am not sure how it is going to behave when I do what you suggest, but I will try it and report back.

The person who said he got it working at high speed is J. Daniel Medeiros (I hope I spelled that right) but I don't have his email and the thread with his post is closed. I don't mind floundering around to do the same thing, that is what we all did before we had the web. (I have been programming since punched card days.) I approached him on FB but no response yet.

On Thu, Mar 29, 2018 at 4:09 PM, psychocrypt notifications@github.com wrote:

Do you tried xmr-stak. KNL is also x86 therefore the miner supports the system without changes.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fireice-uk/xmr-stak/issues/1242#issuecomment-377373375, or mute the thread https://github.com/notifications/unsubscribe-auth/AjhwB6je3LYv2uM2cuGpnWP1taHMCIFjks5tjU2fgaJpZM4S92Lg .

ardatious commented 6 years ago

i installed ubuntu 16.04.4 64 bit

i had to use the -f on apt commands to install cmake and git

did cmake and make and everything seemed to have worked.

where do I go to run the binary? i browsed thru several directories and cant find it

hieixu commented 6 years ago

binary is under your build/bin directory. you can copy it to somewhere like /usr/bin/

ardatious commented 6 years ago

I forgot you cannot execute xmr-stak, you have to execute ./xmr-stak

im an old unix guy. i don't know if all of linux is this stupid (like not allowing root login, but you have to type sudo a billion times to do the same thing)

meanwhile back at the ranch, i compiled everything up and ran xmr-stak. 272 hashes per second on a dell r410, versus lukminer's 372. have'nt actually tested xmr-stak on nicehash for 24 hour timed trial yet.

did the same setup again and ran it on one node of the supermicro phi 7210 system.

for some reason it only ran on the even numbered cores, 32 out of 64, at 8 hashes per second on a core.

i manually edited cpu.txt and added the odd numbered cores, that gave me 64 cores running at 6.7 hashes per second.

total was about 480 hashes per second. for reference, lukminer on a dual 5620 8-core dell r410 runs at 372 hashes per second.

lukminer on the 7210 phi machine node runs on the 4 virtual threads per core, using 254 threads and a total hash rate of 2570 hashes per second (cryptonight).

multiply that by 4 for all nodes.

however he is writing the code to take full advantage of the phi. his numbers check out with the predictions from whattomine. a 4 7210 node system generates a smidgen more than 24 nvidia gtf 1060s. the 4 node system pulls 1350 watts while crunching.

the person who reported getting 22khs on a 7210 node or coprocessor, may have been using opencl.

On Fri, Mar 30, 2018 at 6:41 PM, hieixu notifications@github.com wrote:

binary is under your build/bin directory. you can copy it to somewhere like /usr/bin/

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fireice-uk/xmr-stak/issues/1242#issuecomment-377646848, or mute the thread https://github.com/notifications/unsubscribe-auth/AjhwB42O0FO_nFUc6Z3aKxtOrZyf0B-Cks5tjsK7gaJpZM4S92Lg .

psychocrypt commented 6 years ago

If you like you can also use 4 threads per core in xmr-stak. Set low power mode to 4.

ardatious commented 6 years ago

could you reopen the thread where j Daniel says he got 22k on the phi, so I can ask him how he did it?

On Sat, Mar 31, 2018 at 2:05 AM, psychocrypt notifications@github.com wrote:

If you like you can also use 4 threads per core in xmr-stak. Set low power mode to 4.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/fireice-uk/xmr-stak/issues/1242#issuecomment-377672005, or mute the thread https://github.com/notifications/unsubscribe-auth/AjhwB_CRF5jPH0Me28ergCA10KI8SqOCks5tjyqhgaJpZM4S92Lg .