Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

CryptonightR testing (second round) #5

Open
SChernykh opened this Issue Jan 11, 2019 · 56 comments

Comments

Projects
None yet
@SChernykh
Copy link
Owner

SChernykh commented Jan 11, 2019

Test results will be collected here. Please post your hashrate and power consumption (if you can measure it) compared to latest xmrig/xmrig-amd release on CryptonightV2.

1-2% hashrate drop is normal because CPU/GPU miner code is not fully optimized yet. In case there is a significant hashrate drop on some hardware, we have an option to reduce random math size appropriately.

Basic algorithm description:

Optimized CPU miner:

Optimized GPU miner:

Test pools:

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 11, 2019

GPU tests

AMD Radeon RX 560 on Windows 10 1809: core @ 1196 MHz, memory @ 2200 MHz, 1 Click PBE Timing Straps, 18.6.1 drivers, monitor plugged in, 2 threads (intensity 512, worksize 32, strided_index 2, mem_chunk 2, unroll 8):

  • CN/2 515.5 H/s
  • CN/R 513.8-515.1 H/s, average 514.3 H/s

AMD Vega 64 (stock, 2 threads, intensity 1920, worksize 16, strided_index 2, mem_chunk 2, unroll_factor 8):

  • CN/2 1765.5 H/s
  • CN/R 1763.8-1765.6 H/s, average 1764.3 H/s
@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 12, 2019

CPU tests

AMD Ryzen 5 2600 @ 4 GHz, single thread:

  • CN/2 93.8 H/s
  • CN/R 91.6-93.3 H/s, average 92.0 H/s

I've been experimenting with Ryzen a lot, and this variance doesn't come from random math directly, it's somehow related to subtle instruction decoding stalls that happen for some code sequences. But I know how to fix it, it'll just take some more time.

Intel Pentium G5400 (Coffee Lake 3.7 GHz, 2 threads):

  • CN/2 143.2 H/s
  • CN/R 138.4-141.3 H/s, average 140 H/s

Intel Core i7-2600 (Sandy Bridge 3.4 GHz, 4 threads):

  • CN/2 267.9 H/s
  • CN/R 259.8-265.2 H/s, average 263.4 H/s

Dual AMD Opteron 6276 (16 physical cores, 24 threads)

  • CN/2 720 H/s
  • CN/R 752.9-786.8 H/s, average 771 H/s

Fluctuations are much smaller now compared to the first version.

@Gingeropolous

This comment has been minimized.

Copy link

Gingeropolous commented Jan 12, 2019

CN/2
[2019-01-12 07:48:44] speed 10s/60s/15m 298.4 267.5 n/a H/s max 312.7 H/s
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
| 0 | -1 | 75.0 | 66.7 | n/a |
| 1 | -1 | 75.7 | 67.3 | n/a |
| 2 | -1 | 74.6 | 66.6 | n/a |
| 3 | -1 | 75.6 | 67.3 | n/a |

CN/R

[2019-01-12 07:42:03] speed 10s/60s/15m 293.8 222.2 n/a H/s max 294.9 H/s
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
| 0 | -1 | 74.3 | 60.9 | n/a |
| 1 | -1 | 74.2 | 60.8 | n/a |
| 2 | -1 | 71.8 | 59.6 | n/a |
| 3 | -1 | 73.3 | 69.3 | n/a |

model name : Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz (4 cores, 8 threads)

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 12, 2019

Did you test it for only 1 minute? You probably hit one of the slowest code sequences. Core 3 is also 9 H/s faster than other 3 cores which shouldn't happen - something was probably running in the background and spoiled test result.

@wowario

This comment has been minimized.

Copy link

wowario commented Jan 12, 2019

Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz (1) x64 AES

CN/2

[2019-01-12 18:46:13] new job from pool.wowne.ro:3333 diff 1080 algo cn/2
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
|      0 |        0 |    16.9 |    16.3 |    16.7 |
|      1 |        0 |    17.0 |    16.4 |    16.8 |
|      2 |        0 |    16.9 |    16.4 |    16.7 |
[2019-01-12 18:46:14] speed 10s/60s/15m 50.7 49.0 50.2 H/s max 50.9 H/s

CN/R

[2019-01-12 19:06:49] new job from testnet.wowne.ro:3333 diff 1950 algo cn/r height 350
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
|      0 |        0 |    17.9 |    16.5 |    17.2 |
|      1 |        0 |    18.0 |    16.5 |    17.2 |
|      2 |        0 |    17.9 |    16.5 |    17.2 |
[2019-01-12 19:06:50] speed 10s/60s/15m 53.8 49.4 51.7 H/s max 54.0 H/s
@jorgealonso108

This comment has been minimized.

Copy link

jorgealonso108 commented Jan 12, 2019

Intel® Core™ i5-5250U CPU @ 1.60GHz × 4

  • ABOUT XMRig/2.8.5-dev gcc/7.3.0
  • LIBS libuv/1.18.0 OpenSSL/1.1.0g microhttpd/0.9.59
  • CPU Intel(R) Core(TM) i5-5250U CPU @ 1.60GHz (1) x64 AES

Hello, I am trying to provide some feedback for testing but always get this error...

[2019-01-12 12:01:29] new job from testnet.wowne.ro:3333 diff 5001 algo cn/r height 354
[2019-01-12 12:01:29] rejected (0/1) diff 5000 "Low difficulty share" (569 ms)

[2019-01-12 12:03:29] speed 10s/60s/15m 60.3 56.7 n/a H/s max 62.2 H/s
[2019-01-12 12:03:30] new job from testnet.wowne.ro:3333 diff 1600 algo cn/r height 354
[2019-01-12 12:03:30] rejected (0/3) diff 1599 "Low difficulty share" (507 ms)
[2019-01-12 12:03:30] speed 10s/60s/15m 60.1 56.8 n/a H/s max 62.2 H/s

Does testnet always reject shares? Just let me know... I have a few cpu's i can test and provide feedback

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 12, 2019

@jorgealonso108 It shouldn't reject shares. Did you compile the latest version?

Edit: I've compiled it using GCC 7.3.0 on Ubuntu 18.04 and GCC 8.2.1 on Windows 10 (MSYS2 build) and both builds worked fine, all shares were accepted.

@jorgealonso108

This comment has been minimized.

Copy link

jorgealonso108 commented Jan 12, 2019

I'm on Ubuntu 18.04 clean/fresh install....I have used the same version miner "XMRig/2.8.5-dev gcc/7.3.0" on other pools and have had no problems, its actually the first time I have seen that error...I don't really know what it means, but i would like to help fix the problem.
Maybe latency to the pool? i.e. (569 ms) (507 ms)
I can try it on a few cloud vM's closer to the pool?
Maybe pool software?
I do mine wownero and monero, cpu's only...let me know what i can do to help?

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 12, 2019

@jorgealonso108 How exactly did you compile it? Did you do

git clone https://github.com/SChernykh/xmrig.git
cd xmrig
git checkout CryptonightR
mkdir build
cd build
cmake ..
make

???

@jorgealonso108

This comment has been minimized.

Copy link

jorgealonso108 commented Jan 12, 2019

No...I compiled it from download zip folder...not with git clone
I get this error with the above instructions...git checkout error?

ubuntu@ubuntu:$ git clone https://github.com/SChernykh/xmrig.git
Cloning into 'xmrig'...
remote: Enumerating objects: 59, done.
remote: Counting objects: 100% (59/59), done.
remote: Compressing objects: 100% (29/29), done.
remote: Total 5795 (delta 35), reused 41 (delta 30), pack-reused 5736
Receiving objects: 100% (5795/5795), 1.81 MiB | 1.24 MiB/s, done.
Resolving deltas: 100% (4256/4256), done.
ubuntu@ubuntu:
$ git checkout CryptonightR
fatal: not a git repository (or any of the parent directories): .git
ubuntu@ubuntu:~$

@jorgealonso108

This comment has been minimized.

Copy link

jorgealonso108 commented Jan 12, 2019

Please, if you can send me git clone exact instructions...I will follow them exactly.
I was use xmrig, love it, never have a problem. I must be doing something wrong?

[ 86%] Building CXX object CMakeFiles/xmrig.dir/src/crypto/CryptonightR_gen.cpp.o
In file included from /home/ubuntu/Downloads/xmrig-CryptonightR/src/crypto/CryptonightR_gen.cpp:2:0:
/home/ubuntu/Downloads/xmrig-CryptonightR/src/crypto/variant4_random_math.h:182:12: warning: ‘int v4_random_math_init(V4_Instruction*, uint64_t)’ defined but not used [-Wunused-function]
static int v4_random_math_init(struct V4_Instruction* code, const uint64_t height)
^~~~~~~~~~~~~~~~~~~
[ 88%] Building CXX object CMakeFiles/xmrig.dir/src/common/log/SysLog.cpp.o
[ 90%] Building CXX object CMakeFiles/xmrig.dir/src/api/Api.cpp.o
cc1plus: warning: unrecognized command line option ‘-Wno-class-memaccess’
[ 91%] Building CXX object CMakeFiles/xmrig.dir/src/api/ApiRouter.cpp.o

Not trying to waste your time, just trying to help.

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 12, 2019

@jorgealonso108 I edited my post above: cd xmrig shoud be done before git checkout - you can try again.

@jorgealonso108

This comment has been minimized.

Copy link

jorgealonso108 commented Jan 12, 2019

Same problem as before...following these instructions exactly:
git clone https://github.com/SChernykh/xmrig.git
cd xmrig
git checkout CryptonightR
mkdir build
cd build
cmake ..
make

ubuntu@ubuntu:$ git clone https://github.com/SChernykh/xmrig.git
Cloning into 'xmrig'...
remote: Enumerating objects: 59, done.
remote: Counting objects: 100% (59/59), done.
remote: Compressing objects: 100% (29/29), done.
remote: Total 5795 (delta 35), reused 41 (delta 30), pack-reused 5736
Receiving objects: 100% (5795/5795), 1.81 MiB | 2.76 MiB/s, done.
Resolving deltas: 100% (4256/4256), done.
ubuntu@ubuntu:
$ cd xmrig
ubuntu@ubuntu:/xmrig$ git checkout CryptonightR
Branch 'CryptonightR' set up to track remote branch 'CryptonightR' from 'origin'.
Switched to a new branch 'CryptonightR'
ubuntu@ubuntu:
/xmrig$ mkdir build
ubuntu@ubuntu:/xmrig$ cd build
ubuntu@ubuntu:
/xmrig/build$ cmake ..
-- The C compiler identification is GNU 7.3.0
-- The CXX compiler identification is GNU 7.3.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found UV: /usr/lib/x86_64-linux-gnu/libuv.a
-- Found OpenSSL: /usr/lib/x86_64-linux-gnu/libcrypto.so (found version "1.1.0g")
-- The ASM compiler identification is GNU
-- Found assembler: /usr/bin/cc
-- Looking for syslog.h
-- Looking for syslog.h - found
-- Found MHD: /usr/lib/x86_64-linux-gnu/libmicrohttpd.so
-- Configuring done
-- Generating done
-- Build files have been written to: /home/ubuntu/xmrig/build
ubuntu@ubuntu:~/xmrig/build$ make
Scanning dependencies of target xmrig-asm
Scanning dependencies of target xmrig-asm
[ 1%] Building ASM object CMakeFiles/xmrig-asm.dir/src/crypto/asm/cnv2_main_loop.S.o
[ 3%] Building ASM object CMakeFiles/xmrig-asm.dir/src/crypto/asm/CryptonightR_template.S.o
[ 4%] Building ASM object CMakeFiles/xmrig-asm.dir/src/crypto/asm/CryptonightR_64_template.S.o

All good till it gets here...

[ 85%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_skein.c.o
[ 86%] Building CXX object CMakeFiles/xmrig.dir/src/crypto/CryptonightR_gen.cpp.o
In file included from /home/ubuntu/xmrig/src/crypto/CryptonightR_gen.cpp:2:0:
/home/ubuntu/xmrig/src/crypto/variant4_random_math.h:182:12: warning: ‘int v4_random_math_init(V4_Instruction*, uint64_t)’ defined but not used [-Wunused-function]
static int v4_random_math_init(struct V4_Instruction* code, const uint64_t height)
^~~~~~~~~~~~~~~~~~~
cc1plus: warning: unrecognized command line option ‘-Wno-class-memaccess’
[ 88%] Building CXX object CMakeFiles/xmrig.dir/src/common/log/SysLog.cpp.o
[ 90%] Building CXX object CMakeFiles/xmrig.dir/src/api/Api.cpp.o
[ 91%] Building CXX object CMakeFiles/xmrig.dir/src/api/ApiRouter.cpp.o
[ 93%] Building CXX object CMakeFiles/xmrig.dir/src/common/api/Httpd.cpp.o
[ 95%] Building CXX object CMakeFiles/xmrig.dir/src/common/api/HttpRequest.cpp.o
[ 96%] Building CXX object CMakeFiles/xmrig.dir/src/common/net/Tls.cpp.o
[ 98%] Building CXX object CMakeFiles/xmrig.dir/src/crypto/Asm.cpp.o
[100%] Linking CXX executable xmrig
[100%] Built target xmrig
ubuntu@ubuntu:~/xmrig/build$

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 12, 2019

It's just a warning, you have xmrig binary at this point.

@Gingeropolous

This comment has been minimized.

Copy link

Gingeropolous commented Jan 12, 2019

cnv2

[2019-01-12 16:11:47] speed 10s/60s/15m 316.8 315.3 314.9 H/s max 325.5 H/s
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
| 0 | -1 | 78.7 | 78.4 | 78.3 |
| 1 | -1 | 79.6 | 79.3 | 79.2 |
| 2 | -1 | 78.8 | 78.5 | 78.4 |
| 3 | -1 | 79.5 | 79.2 | 79.1 |

cnr

[2019-01-12 15:50:17] speed 10s/60s/15m 342.3 342.4 338.7 H/s max 342.7 H/s
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
| 0 | -1 | 86.3 | 86.3 | 85.3 |
| 1 | -1 | 84.9 | 85.0 | 84.0 |
| 2 | -1 | 86.3 | 86.3 | 85.3 |
| 3 | -1 | 84.9 | 84.9 | 84.0 |

model name : Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz

yeah, turns out i was running xmr-stak for some GPUs. must have been using enough of something on the CPU to run the GPUs.

@jorgealonso108

This comment has been minimized.

Copy link

jorgealonso108 commented Jan 12, 2019

Ok here are my results

Intel® Core™ i5-5250U CPU @ 1.60GHz × 4
sudo sysctl -w vm.nr_hugepages=128

figures shown are max h/s...CN/r seems to be more stable and slightly faster, see details below...
Computer was completely left alone during testing...not even mouse movements...trying to get an exact apples to apples comparison.

CN/r 64.3
CN/v2 63.3

sudo ./xmrig --print-time 1 --max-cpu-usage 100 --cpu-priority 4 -o testnet.wowne.ro:3333 -u 9zuKdp6MrQaBSw85QP2fFh8zznN8tPbpJQZhX6wyVGHb5TdJu4QYcPK7EtfhUtnYbVUCFkNM9sPfViN5HvmXxV9iEVJP2o8 -p x -k
commit 5367300

  • ABOUT XMRig/2.8.5-dev gcc/7.3.0
  • LIBS libuv/1.18.0 OpenSSL/1.1.0g microhttpd/0.9.59
  • CPU Intel(R) Core(TM) i5-5250U CPU @ 1.60GHz (1) x64 AES
  • CPU L2/L3 0.5 MB/3.0 MB
  • THREADS 2, cryptonight, av=0, donate=5%
  • ASSEMBLY auto:intel
  • POOL #1 testnet.wowne.ro:3333 variant auto
  • COMMANDS hashrate, pause, resume
    [2019-01-12 16:21:17] use pool testnet.wowne.ro:3333 159.69.41.143
    [2019-01-12 16:21:17] new job from testnet.wowne.ro:3333 diff 5000 algo cn/r height 385
    [2019-01-12 16:21:17] READY (CPU) threads 2(2) huge pages 2/2 100% memory 4.0 MB

[2019-01-12 16:51:11] speed 10s/60s/15m 64.1 64.1 63.6 H/s max 64.3 H/s
[2019-01-12 16:51:12] accepted (50/0) diff 2160 (324 ms)
[2019-01-12 16:51:12] speed 10s/60s/15m 64.1 63.9 63.6 H/s max 64.3 H/s
[2019-01-12 16:51:13] speed 10s/60s/15m 64.0 63.8 63.6 H/s max 64.3 H/s
[2019-01-12 16:51:14] speed 10s/60s/15m 64.1 64.1 63.6 H/s max 64.3 H/s
[2019-01-12 16:51:14] new job from testnet.wowne.ro:3333 diff 2190 algo cn/r height 386
[2019-01-12 16:51:15] speed 10s/60s/15m 64.0 63.9 63.6 H/s max 64.3 H/s

sudo ./xmrig --print-time 1 --max-cpu-usage 100 --cpu-priority 4 -o pool.wowne.ro:3333 -u Wo4hunKKcZ9JhCrqpFMotQQAiJAiATuJLNAK2HrPLoNzK8hkqNbf9t8gmx6bzAQrXRMnWnoELoiD6GTv8guPBRwH1FpjujDKH -p x -k
commit 2b0b71b

  • ABOUT XMRig/2.8.3 gcc/7.3.0
  • LIBS libuv/1.18.0 OpenSSL/1.1.0g microhttpd/0.9.59
  • CPU Intel(R) Core(TM) i5-5250U CPU @ 1.60GHz (1) x64 AES
  • CPU L2/L3 0.5 MB/3.0 MB
  • THREADS 2, cryptonight, av=0, donate=5%
  • ASSEMBLY auto:intel
  • POOL #1 pool.wowne.ro:3333 variant auto
  • COMMANDS hashrate, pause, resume
    [2019-01-12 17:33:56] READY (CPU) threads 2(2) huge pages 2/2 100% memory 4.0 MB
    [2019-01-12 17:33:56] use pool pool.wowne.ro:3333 195.201.115.12
    [2019-01-12 17:33:56] new job from pool.wowne.ro:3333 diff 5000 algo cn/2

[2019-01-12 17:59:37] speed 10s/60s/15m 61.7 60.2 60.1 H/s max 63.3 H/s
[2019-01-12 17:59:37] new job from pool.wowne.ro:3333 diff 1740 algo cn/2
[2019-01-12 17:59:38] speed 10s/60s/15m 61.8 60.1 60.1 H/s max 63.3 H/s
[2019-01-12 17:59:39] speed 10s/60s/15m 61.7 60.1 60.1 H/s max 63.3 H/s
[2019-01-12 17:59:40] speed 10s/60s/15m 61.5 60.1 60.1 H/s max 63.3 H/s
[2019-01-12 17:59:40] accepted (52/0) diff 1740 (326 ms)
[2019-01-12 17:59:41] speed 10s/60s/15m 61.5 60.2 60.1 H/s max 63.3 H/s
[2019-01-12 17:59:42] speed 10s/60s/15m 61.3 60.2 60.1 H/s max 63.3 H/s

@fuwa0529

This comment has been minimized.

Copy link

fuwa0529 commented Jan 13, 2019

AMD Ryzen 3 2200G, Linux 4.20.0, 2 threads:

  • CN/2: speed 10s/60s/15m 127.3 116.3 116.2 H/s max 134.0 H/s
  • CN/R: speed 10s/60s/15m 126.7 127.2 122.2 H/s max 130.5 H/s
@nssy

This comment has been minimized.

Copy link

nssy commented Jan 15, 2019

Dual Xeon: Linux
Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz (2) x64 AES

CN/2

[2019-01-15 11:32:02] new job from xmrpool.eu:7777 diff 10000 algo cn/2 height 0
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
|      0 |        0 |    50.1 |    50.1 |     n/a |
|      1 |        1 |    50.9 |    51.0 |     n/a |
|      2 |        2 |    43.0 |    43.1 |     n/a |
|      3 |        3 |    51.9 |    51.9 |     n/a |
|      4 |        4 |    51.3 |    51.4 |     n/a |
|      5 |        5 |    48.0 |    48.0 |     n/a |
|      6 |        6 |    47.2 |    47.2 |     n/a |
|      7 |        7 |    46.0 |    46.0 |     n/a |
|      8 |        8 |    46.3 |    46.3 |     n/a |
|      9 |        9 |    51.3 |    51.3 |     n/a |
|     10 |       10 |    51.4 |    51.4 |     n/a |
|     11 |       11 |    42.1 |    42.2 |     n/a |
|     12 |       12 |    50.1 |    50.1 |     n/a |
|     13 |       13 |    49.2 |    49.2 |     n/a |
|     14 |       14 |    50.4 |    50.4 |     n/a |
|     15 |       15 |    51.2 |    51.2 |     n/a |
|     16 |       16 |    43.2 |    43.2 |     n/a |
|     17 |       17 |    52.0 |    52.0 |     n/a |
|     18 |       18 |    51.5 |    51.5 |     n/a |
|     19 |       19 |    48.1 |    48.1 |     n/a |
|     20 |       20 |    47.4 |    47.4 |     n/a |
|     21 |       21 |    46.0 |    46.0 |     n/a |
|     22 |       22 |    46.4 |    46.4 |     n/a |
|     23 |       23 |    51.3 |    51.3 |     n/a |
|     24 |       24 |    51.4 |    51.4 |     n/a |
|     25 |       25 |    42.1 |    42.1 |     n/a |
|     26 |       26 |    50.1 |    50.1 |     n/a |
|     27 |       27 |    49.2 |    49.2 |     n/a |
|     28 |       30 |    43.3 |    43.4 |     n/a |
|     29 |       39 |    42.4 |    42.4 |     n/a |
|     30 |       44 |    43.5 |    43.5 |     n/a |
|     31 |       53 |    42.4 |    42.4 |     n/a |
[2019-01-15 11:32:03] speed 10s/60s/15m 1530.6 1531.2 n/a H/s max 1531.6 H/s

CN/R

[2019-01-15 11:29:30] new job from testnet.wowne.ro:3333 diff 47610 algo cn/r height 641
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
|      0 |        0 |    49.6 |    49.5 |     n/a |
|      1 |        1 |    50.5 |    50.4 |     n/a |
|      2 |        2 |    35.2 |    35.2 |     n/a |
|      3 |        3 |    51.4 |    51.3 |     n/a |
|      4 |        4 |    50.9 |    50.7 |     n/a |
|      5 |        5 |    47.5 |    47.4 |     n/a |
|      6 |        6 |    46.7 |    46.6 |     n/a |
|      7 |        7 |    45.5 |    45.4 |     n/a |
|      8 |        8 |    45.8 |    45.7 |     n/a |
|      9 |        9 |    50.8 |    50.7 |     n/a |
|     10 |       10 |    50.9 |    50.8 |     n/a |
|     11 |       11 |    34.6 |    34.7 |     n/a |
|     12 |       12 |    49.6 |    49.5 |     n/a |
|     13 |       13 |    48.8 |    48.7 |     n/a |
|     14 |       14 |    49.9 |    49.8 |     n/a |
|     15 |       15 |    50.6 |    50.5 |     n/a |
|     16 |       16 |    35.3 |    35.4 |     n/a |
|     17 |       17 |    51.4 |    51.4 |     n/a |
|     18 |       18 |    50.9 |    50.9 |     n/a |
|     19 |       19 |    47.5 |    47.4 |     n/a |
|     20 |       20 |    46.8 |    46.7 |     n/a |
|     21 |       21 |    45.5 |    45.4 |     n/a |
|     22 |       22 |    45.8 |    45.8 |     n/a |
|     23 |       23 |    50.7 |    50.6 |     n/a |
|     24 |       24 |    50.8 |    50.8 |     n/a |
|     25 |       25 |    34.6 |    34.6 |     n/a |
|     26 |       26 |    49.5 |    49.5 |     n/a |
|     27 |       27 |    48.7 |    48.7 |     n/a |
|     28 |       30 |    35.2 |    35.3 |     n/a |
|     29 |       39 |    34.6 |    34.7 |     n/a |
|     30 |       44 |    35.3 |    35.4 |     n/a |
|     31 |       53 |    34.6 |    34.6 |     n/a |
[2019-01-15 11:29:44] speed 10s/60s/15m 1455.6 1454.3 n/a H/s max 1469.1 H/s

So approx 5% speed difference
CN/2: [2019-01-15 11:32:03] speed 10s/60s/15m 1530.6 1531.2 n/a H/s max 1531.6 H/s
CN/R: [2019-01-15 11:29:44] speed 10s/60s/15m 1455.6 1454.3 n/a H/s max 1469.1 H/s

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 15, 2019

@nssy Do you have any double threads in your config? Double hash code for cn/r is far from optimized yet.
Can you test both algorithms only with single threads?

@nssy

This comment has been minimized.

Copy link

nssy commented Jan 15, 2019

No double hash threads used.

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 15, 2019

@nssy Still, you have 32 threads on only 28 physical cores, so 4 cores run 2 single threads each. This configuration is much slower because single thread already uses 80% of a single CPU core compute capacity. I'll need to implement double thread using SSE instructions to fix this slowdown.

@nssy

This comment has been minimized.

Copy link

nssy commented Jan 15, 2019

Ok I have run both with 28 threads instead.

Dual Xeon: Linux, 28 threads

Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz (2) x64 AES

CN/2

[2019-01-15 12:25:09] READY (CPU) threads 28(28) huge pages 28/28 100% memory 56.0 MB
....
[2019-01-15 12:28:46] new job from xmrpool.eu:7777 diff 240009 algo cn/2 height 0
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
|      0 |        0 |    50.7 |    50.6 |     n/a |
|      1 |        1 |    51.4 |    51.4 |     n/a |
|      2 |        2 |    52.1 |    52.1 |     n/a |
|      3 |        3 |    52.3 |    52.3 |     n/a |
|      4 |        4 |    51.8 |    51.8 |     n/a |
|      5 |        5 |    48.4 |    48.4 |     n/a |
|      6 |        6 |    47.7 |    47.7 |     n/a |
|      7 |        7 |    46.4 |    46.4 |     n/a |
|      8 |        8 |    46.8 |    46.8 |     n/a |
|      9 |        9 |    51.7 |    51.7 |     n/a |
|     10 |       10 |    51.8 |    51.8 |     n/a |
|     11 |       11 |    50.7 |    50.7 |     n/a |
|     12 |       12 |    50.5 |    50.5 |     n/a |
|     13 |       13 |    49.7 |    49.7 |     n/a |
|     14 |       14 |    50.8 |    50.8 |     n/a |
|     15 |       15 |    51.5 |    51.5 |     n/a |
|     16 |       16 |    52.1 |    52.1 |     n/a |
|     17 |       17 |    52.4 |    52.4 |     n/a |
|     18 |       18 |    51.9 |    51.9 |     n/a |
|     19 |       19 |    48.4 |    48.4 |     n/a |
|     20 |       20 |    47.7 |    47.7 |     n/a |
|     21 |       21 |    46.4 |    46.4 |     n/a |
|     22 |       22 |    46.8 |    46.8 |     n/a |
|     23 |       23 |    51.7 |    51.7 |     n/a |
|     24 |       24 |    51.8 |    51.8 |     n/a |
|     25 |       25 |    50.7 |    50.7 |     n/a |
|     26 |       26 |    50.5 |    50.5 |     n/a |
|     27 |       27 |    49.7 |    49.7 |     n/a |
[2019-01-15 12:28:49] speed 10s/60s/15m 1404.6 1404.4 n/a H/s max 1404.6 H/s

CN/R

[2019-01-15 12:29:16] READY (CPU) threads 28(28) huge pages 28/28 100% memory 56.0 MB
...
[2019-01-15 12:30:21] new job from testnet.wowne.ro:3333 diff 2340 algo cn/r height 653
[2019-01-15 12:30:22] accepted (2/0) diff 2340 (212 ms)
| THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
|      0 |        0 |    50.2 |    50.2 |     n/a |
|      1 |        1 |    51.0 |    51.0 |     n/a |
|      2 |        2 |    51.7 |    51.7 |     n/a |
|      3 |        3 |    51.8 |    51.9 |     n/a |
|      4 |        4 |    51.4 |    51.4 |     n/a |
|      5 |        5 |    47.9 |    47.9 |     n/a |
|      6 |        6 |    47.2 |    47.2 |     n/a |
|      7 |        7 |    45.9 |    45.9 |     n/a |
|      8 |        8 |    46.2 |    46.2 |     n/a |
|      9 |        9 |    51.2 |    51.2 |     n/a |
|     10 |       10 |    51.4 |    51.3 |     n/a |
|     11 |       11 |    50.3 |    50.3 |     n/a |
|     12 |       12 |    50.1 |    50.1 |     n/a |
|     13 |       13 |    49.2 |    49.2 |     n/a |
|     14 |       14 |    50.2 |    50.2 |     n/a |
|     15 |       15 |    51.0 |    51.0 |     n/a |
|     16 |       16 |    51.7 |    51.7 |     n/a |
|     17 |       17 |    51.9 |    51.9 |     n/a |
|     18 |       18 |    51.4 |    51.4 |     n/a |
|     19 |       19 |    47.9 |    47.9 |     n/a |
|     20 |       20 |    47.2 |    47.2 |     n/a |
|     21 |       21 |    45.9 |    45.9 |     n/a |
|     22 |       22 |    46.2 |    46.2 |     n/a |
|     23 |       23 |    51.2 |    51.2 |     n/a |
|     24 |       24 |    51.4 |    51.4 |     n/a |
|     25 |       25 |    50.3 |    50.3 |     n/a |
|     26 |       26 |    50.1 |    50.1 |     n/a |
|     27 |       27 |    49.2 |    49.2 |     n/a |
[2019-01-15 12:30:22] speed 10s/60s/15m 1391.1 1391.0 n/a H/s max 1391.2 H/s

So approx 0.9% speed difference
CN/2: [2019-01-15 12:28:49] speed 10s/60s/15m 1404.6 1404.4 n/a H/s max 1404.6 H/s
CN/R: [2019-01-15 12:30:22] speed 10s/60s/15m 1391.1 1391.0 n/a H/s max 1391.2 H/s

@gvgit

This comment has been minimized.

Copy link

gvgit commented Jan 15, 2019

I've tried Opteron 6276. With single thread I got

Single Thread: 53H/s

When invoking 24 Threads I got:

24 Threads: 760H/s

But 760:24=31,66H/s

Something is woring, I tried all settings I could but got no better performance.

Does someone has a solution?

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 15, 2019

Opteron 6276 has 8 core modules and 16 MB L2 cache, dual system has 16 modules and 32 MB L2 cache, but you run 24 threads - it's more than it can handle without dropping single thread performance.

@gvgit

This comment has been minimized.

Copy link

gvgit commented Jan 15, 2019

Sorry, I did not mention that it is a dual board. 2 x Opteron 6276 inside a HP dl165 G7. In total 16 cores/32 threads.

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 15, 2019

More than 1 thread (or scratchpad) per core is not optimized yet. I'll be working on optimized double hash code this and next week, so you can expect some improvement when it's done.

@gvgit

This comment has been minimized.

Copy link

gvgit commented Jan 15, 2019

Thanks

@jorgealonso108

This comment has been minimized.

Copy link

jorgealonso108 commented Jan 20, 2019

Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz
sudo sysctl -w vm.nr_hugepages=128

cn/R
sudo ./xmrig --print-time 1 --av 1 --cpu-priority 4 -t 1 -o testnet.wowne.ro:3333 -u 9zuKdp6MrQaBSw85QP2fFh8zznN8tPbpJQZhX6wyVGHb5TdJu4QYcPK7EtfhUtnYbVUCFkNM9sPfViN5HvmXxV9iEVJP2o8 -p x -k
[2019-01-20 20:39:10] accepted (55/0) diff 1000 (23 ms)
[2019-01-20 20:39:10] new job from testnet.wowne.ro:3333 diff 1000 algo cn/r height 108
[2019-01-20 20:39:10] speed 10s/60s/15m 32.8 32.7 32.8 H/s max 35.1 H/s

cn/2
sudo ./xmrig --print-time 1 --av 1 --cpu-priority 4 -t 1 -o pool.wowne.ro:3333 -u Wo4fVqVg7prJTJ44hPmJK8e3bzBWL7gWVLnU8vSeg4NwUQvxuxLdB9xZY6PZHmhLxeCYPSxCYwcqkfa86Zf8gcFw2NRRyT3mN -p x -k
[2019-01-20 20:37:54] speed 10s/60s/15m 37.0 36.8 36.2 H/s max 37.5 H/s
[2019-01-20 20:37:55] new job from pool.wowne.ro:3333 diff 1110 algo cn/2
[2019-01-20 20:37:55] speed 10s/60s/15m 37.1 36.8 36.2 H/s max 37.5 H/s

[2019-01-20 20:38:46] speed 10s/60s/15m 37.0 36.9 36.3 H/s max 37.5 H/s
[2019-01-20 20:38:47] accepted (63/0) diff 1110 (91 ms)
[2019-01-20 20:38:47] speed 10s/60s/15m 36.7 36.8 36.3 H/s max 37.5 H/s

@jorgealonso108

This comment has been minimized.

Copy link

jorgealonso108 commented Jan 21, 2019

Intel(R) Xeon(R) CPU E5-2673 v4 @ 2.30GHz
sudo sysctl -w vm.nr_hugepages=128

cn/R
sudo ./xmrig --print-time 1 --av 1 --cpu-priority 4 -t 1 -o testnet.wowne.ro:3333 -u 9zuKdp6MrQaBSw85QP2fFh8zznN8tPbpJQZhX6wyVGHb5TdJu4QYcPK7EtfhUtnYbVUCFkNM9sPfViN5HvmXxV9iEVJP2o8 -p x -k
[2019-01-21 05:09:40] new job from testnet.wowne.ro:3333 diff 1230 algo cn/r height 134

[2019-01-21 05:09:56] speed 10s/60s/15m 37.1 36.9 36.9 H/s max 37.7 H/s
[2019-01-21 05:09:57] accepted (52/0) diff 1230 (88 ms)
[2019-01-21 05:09:57] speed 10s/60s/15m 37.3 36.9 36.9 H/s max 37.7 H/s

cn/2
sudo ./xmrig --print-time 1 --av 1 --cpu-priority 4 -t 1 -o pool.wowne.ro:3333 -u Wo4fVqVg7prJTJ44hPmJK8e3bzBWL7gWVLnU8vSeg4NwUQvxuxLdB9xZY6PZHmhLxeCYPSxCYwcqkfa86Zf8gcFw2NRRyT3mN -p x -k
[2019-01-21 05:58:58] new job from pool.wowne.ro:3333 diff 1140 algo cn/2

[2019-01-21 05:59:42] speed 10s/60s/15m 38.3 38.3 38.3 H/s max 38.9 H/s
[2019-01-21 05:59:42] accepted (59/0) diff 1140 (80 ms)
[2019-01-21 05:59:43] speed 10s/60s/15m 38.3 38.3 38.3 H/s max 38.9 H/s

@gvgit

This comment has been minimized.

Copy link

gvgit commented Jan 21, 2019

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 21, 2019

@gvgit My first attempt to make fast double hash version using SSE failed - it wasn't faster than current code, so I have to think again.

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 22, 2019

@gvgit I've tried different approach for double hash SSE version and it seems to work, at least on Intel processors. I'll keep experimenting, but chances are good that double hash performance will be improved.

@jorgealonso108

This comment has been minimized.

Copy link

jorgealonso108 commented Jan 23, 2019

I am just about to test some more CPU's and post results.

What is "cn/half" algo ?

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 23, 2019

What is "cn/half" algo ?

It's not related to this testing. It's CNv2 with half iterations used in some recent coin forks.

@gvgit

This comment has been minimized.

Copy link

gvgit commented Jan 29, 2019

Any success on double hash SSE regarding AMD CPUs?

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Jan 29, 2019

The results are mixed. As soon as I fixed all bugs in the code, it wasn't faster on Intel and was a bit faster on Ryzen (but Ryzen doesn't need it since two single threads on the same core are still faster). I haven't tested it on older AMD processors yet. Double hash SSE version (test code) is available in https://github.com/SChernykh/CryptonightR - but it requires Windows + Visual Studio to compile and run.

@gvgit

This comment has been minimized.

Copy link

gvgit commented Feb 4, 2019

New success on Opterons?

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Feb 4, 2019

@gvgit Opterons are already faster with CryptonightR than they were with CNv2. I'm not sure SSE would help them because they have only one FP unit per 2 cores, but each core has integer unit. I still need to implement SSE code in xmrig to test it thoroughly.

@gvgit

This comment has been minimized.

Copy link

gvgit commented Feb 4, 2019

@kio3i0j9024vkoenio

This comment has been minimized.

Copy link

kio3i0j9024vkoenio commented Feb 12, 2019

I have compiled XMRig on Ubuntu 16.04 successfully with these commands:

git clone https://github.com/SChernykh/xmrig.git
cd xmrig
git checkout CryptonightR
mkdir build
cd build
cmake ..
make

I now have a xmrig file in the build directory but have no idea how to run it for testing?

The only instructions for configuration is this text:

Usage

Use config.xmrig.com to generate, edit or share configurations.

That seems to reference a Windows program. So again please help me to get this running for testing on my HP DL580 and AMD R815 servers.

@gvgit

This comment has been minimized.

Copy link

gvgit commented Feb 12, 2019

@nssy

This comment has been minimized.

Copy link

nssy commented Feb 13, 2019

@kio3i0j9024vkoenio
For CN/R
./xmrig --max-cpu-usage 100 -o testnet.wowne.ro:3333 -u 9zuKdp6MrQaBSw85QP2fFh8zznN8tPbpJQZhX6wyVGHb5TdJu4QYcPK7EtfhUtnYbVUCFkNM9sPfViN5HvmXxV9iEVJP2o8 -p x -k
You will need to configure threads accordingly.

@pigfrown

This comment has been minimized.

Copy link

pigfrown commented Feb 19, 2019

Ryzen 7 1700 around 450H/S with 8 threads on latest xmrig (cn/wow)

@osensei

This comment has been minimized.

Copy link

osensei commented Feb 26, 2019

@SChernykh, I don't know how many ARMv8 miners are out there, but just to let you know... it seems that CN/R has a big impact on ARMv8

Using xmrig 2.13.1 built with gcc/7.3.0 on a NanoPi M3 (Samsung S5P6818 Octa-Core Cortex-A53) these are my results:

variant 0: 28.6 H/s
variant 1: 28.4 H/s
variant 2: 27.2 H/s
variant R: 14.9 H/s

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Feb 26, 2019

ARM code doesn't have JIT code generator, I didn't think someone would actually use ARM processors for mining.

@osensei

This comment has been minimized.

Copy link

osensei commented Feb 26, 2019

I don't mine on it, I was just testing it for fun. Maybe there are some people out there actually mining on them.
I just thought I would post my results in case you weren't aware of the performance hit. Thanks for all your work btw :)

@kio3i0j9024vkoenio

This comment has been minimized.

Copy link

kio3i0j9024vkoenio commented Mar 8, 2019

System is a HP DL580-G7 Server with four Xeon E7-8837 processors running Ubuntu 16.04

xmr-stak 2.10.0 56d2770

32 threads, 8 threads per processor, on each processor four double threads and four single threads

V8 - 1500.6 H/s
R - 1498.8 H/s

So exactly the same for cryptonight_v8 and cryptonight_r

@resistor4u

This comment has been minimized.

Copy link

resistor4u commented Mar 15, 2019

ARM definitely takes a major hit - xmrig no longer builds for iphone 6s and later (arm64 / aarch64) starting with v2.13.0, so I changed some of the build files to get a working binary. The new binaries run, but with about 75% reduced hashrate - i.e., from ~58.9 H/s to ~14.7 H/s!!

On RPi 3B running 64bit Ubuntu, the hashrate reduced from ~7.3 H/s to ~3.0 H/s.

So, an all around major reduction in ARM devices.

@gvgit

This comment has been minimized.

Copy link

gvgit commented Mar 15, 2019

Any progress on Dual-Opteron 6276?
CN/0 and CN/1 were around 1050kH/s. CN/2 was a knockout with 740kH/s. Will CN/R get it back to >1000kH/s?

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Mar 15, 2019

@resistor4u ARM doesn't have JIT compiler for cn/r, this is why it's slower now. But it's possible to implement.
@gvgit Most likely not, but cn/r was a bit faster than cn/2 last time I tested it.

@gvgit

This comment has been minimized.

Copy link

gvgit commented Mar 15, 2019

@SChernykh Is it technically possible to get it back? There are thousands of Opterons waiting to get back into mining :/

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Mar 15, 2019

@gvgit No, it's not possible because the algorithm is heavy on compute now, it's not just memory access. Opterons have fast cache, but slow core.

@gvgit

This comment has been minimized.

Copy link

gvgit commented Mar 15, 2019

@SChernykh Just for my knowledge. What was the reason to implement an algorithm heavy on computing? Was it necessary due to some reason?

@SChernykh

This comment has been minimized.

Copy link
Owner Author

SChernykh commented Mar 15, 2019

Because only adding more compute makes ASICs slower, memory hardness in Cryptonight is too weak.

@resistor4u

This comment has been minimized.

Copy link

resistor4u commented Mar 15, 2019

@SChernykh thanks for the heads up. I'm not a programmer, but I'm happy to tinker as a hobby. How would you suggest implementing it?

@resistor4u

This comment has been minimized.

Copy link

resistor4u commented Mar 18, 2019

@SChernykh what do you think about the comments here xmrig/xmrig#985 (comment)?

I'd add that the rates I've seen on iphones (~50 H/s) are not insignificant, and that smaller hashing power and rates are part of monero's raison d'être.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.