-
Notifications
You must be signed in to change notification settings - Fork 3.4k
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
v2.99.4-beta Terrible RandomX hash rate on quad Opteron 6348 server #1099
Comments
Please specify option |
In addition please run |
I had "algo": "rx", in the config which I assumed was RandomX. Changed it to "algo": "rx/loki," and am currently testing. It now does show RX but is taking 60-65 seconds to initialize the Dataset (memory) for each NODE. miner@R815-1:~/xmrig-2.99.4-beta/build$ ./xmrig
Is there some way to speed up the init? EDIT: If I do: ./xmrig --randomx-init=6 Then INIT time is reduced to 6-9 seconds from 60-65 seconds. Which makes sense as each NODE in an Opteron 6348 has 6 threads. However if I change the "init" in the config.json from -1 to 6: "randomx": { The INIT time goes to 12-14 seconds. The "-1" or "6" for "init" in config.json is not optimal and takes much more time to INIT than doing the --randomx-init=6 on the command line. Why? |
My results for RandomX test are: [2019-08-04 11:46:44.753] speed 10s/60s/15m 9479.1 9477.9 n/a H/s max 9482.1 H/s The 9482 H/s is 10.6% lower than the 10612 I was getting with randomx-benchmark. seq 0 7 | xargs -P 0 -I node numactl --localalloc -N node ./randomx-benchmark --mine --largePages --jit --nonces 100000 --init 6 --threads 6 |
Attached file topology.xml for the Dell R815 Server with quad 12-core Opteron 6348's. |
InitInit don't use NUMA specific bindings, I use this machine:
for verify NUMA bindings, but it much simpler that Opterons, I don't have access to such CPUs right now. For Intel I found, use more threads is better (all 40) init each dataset take less than 3 seconds. Bug with long initialization (about 60 seconds) was fixed in v2.99.2, init was start from one of mining threads, but if affinity used, all 40 threads use same core, (it Linux specific) now init started from main thread. If change somehow affinity of main thread it will not work correctly. if you run You can use tools like perfomanceI can't confirm noticeable difference between randomx-benchmark and the miner, both do about 8240 H/s (+/- 20 H/s) on CPUs above. Difference between miner and benchmark, miner use affinity and benchmark don't. |
I believe a lot of the issues I am having in performance INIT timing being over 60 seconds and performance being 10.6% low has to do with the libraries that get used when compiling from the source. Going back and testing the compiled version from the source and the pre-compiled version I have found that the pre-compiled version is both faster doing INIT and in hash rates. With the pre-compiled version I see these libraries being used: libuv/1.24.1 OpenSSL/1.1.1a hwloc/2.0.4 whereas if I compile from the source it uses these libraries: libuv/1.8.0 OpenSSL/1.0.2g hwloc/1.11.2 I have done the: sudo apt update but that does not get any new updates to these libraries. Where might I obtain the same libraries used in the pre-compiled version? |
Using the Pre-compiled version of XMRig my results for RandomX test are: [2019-08-04 13:46:55.611] speed 10s/60s/15m 10193.6 10192.3 n/a H/s max 10196.4 H/s The 10196 H/s is only 3.9% lower than the 10612 I was getting with randomx-benchmark so I believe that this hash rate of 10196 is very acceptable as this is a REAL miner and probably has some overhead compared to just the randomx-benchmark program. Well Done. |
I use custom static libraries builds for simple reason, reduce external dependencies, you can verify it by Can you try make a build with hwloc 2.0.4?
Then go to xmrig build directory and:
|
Still
Segmentation fault (core dumped)
Alexandre Naverniouk
…On Sun, Aug 4, 2019 at 10:19 PM xmrig ***@***.***> wrote:
Reopened #1099 <#1099>.
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#1099?email_source=notifications&email_token=AAQHPJD4BUSDDGSNE5DY5PLQC6Z43A5CNFSM4IJFWGD2YY3PNVWWK3TUL52HS4DFWZEXG43VMVCXMZLOORHG65DJMZUWGYLUNFXW5KTDN5WW2ZLOORPWSZGOS3YEQ5Y#event-2532329591>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AAQHPJDNARS6IUMSDALVZWLQC6Z43ANCNFSM4IJFWGDQ>
.
|
I try compile, but I failed
|
Okay I updated manual, |
Hey what speed ram and size and amount of sticks did you use? DDR3? 1600mhz? 16 4gb sticks? |
Hi, what wrong? I have 2 times less hashrate than I can see in test from others.
|
System is:
Dell R815 Server with quad 12-core Opteron 6348's. 32GB of PC3-10600R memory is installed. 8GB per Opteron 6348 which is 4GB for each of the two NODEs in each Opteron 6348.
http://www.cpu-world.com/CPUs/Bulldozer/AMD-Opteron%206348%20-%20OS6348WKTCGHK.html
v2.99.4-beta is only getting 1644 H/s for RandomX whereas testing with randomx-benchmark produces 10612 H/s for RandomX
seq 0 7 | xargs -P 0 -I node numactl --localalloc -N node ./randomx-benchmark --mine --largePages --jit --nonces 100000 --init 6 --threads 6
It appears that v2.99.4-beta has broken NUMA/Thread Affinity and/or Memory Assignment rules.
This is the output of the run:
miner@R815-1:~/xmrig-2.99.4-beta/build$ ./xmrig
L2:48.0 MB L3:64.0 MB 48C/48T NUMA:8
[2019-08-04 11:01:10.908] use pool donate.v2.xmrig.com:3333 159.89.38.204
[2019-08-04 11:01:10.908] new job from donate.v2.xmrig.com:3333 diff 1000225 algo cn/r height 1893504
[2019-08-04 11:01:10.908] cpu use profile cn (48 threads) scratchpad 2048 KB
[2019-08-04 11:01:17.248] cpu READY threads 48(48) huge pages 48/48 100% memory 98304 KB (6339 ms)
| CPU THREAD | AFFINITY | 10s H/s | 60s H/s | 15m H/s |
| 0 | 0 | 33.7 | n/a | n/a |
| 1 | 1 | 34.0 | n/a | n/a |
| 2 | 2 | 33.7 | n/a | n/a |
| 3 | 3 | 33.7 | n/a | n/a |
| 4 | 4 | 33.7 | n/a | n/a |
| 5 | 5 | 34.6 | n/a | n/a |
| 6 | 6 | 34.0 | n/a | n/a |
| 7 | 7 | 33.6 | n/a | n/a |
| 8 | 8 | 33.6 | n/a | n/a |
| 9 | 9 | 33.9 | n/a | n/a |
| 10 | 10 | 33.6 | n/a | n/a |
| 11 | 11 | 34.0 | n/a | n/a |
| 12 | 12 | 33.7 | n/a | n/a |
| 13 | 13 | 33.9 | n/a | n/a |
| 14 | 14 | 33.9 | n/a | n/a |
| 15 | 15 | 34.0 | n/a | n/a |
| 16 | 16 | 34.0 | n/a | n/a |
| 17 | 17 | 33.9 | n/a | n/a |
| 18 | 18 | 33.9 | n/a | n/a |
| 19 | 19 | 33.9 | n/a | n/a |
| 20 | 20 | 34.6 | n/a | n/a |
| 21 | 21 | 34.0 | n/a | n/a |
| 22 | 22 | 34.4 | n/a | n/a |
| 23 | 23 | 34.2 | n/a | n/a |
| 24 | 24 | 34.5 | n/a | n/a |
| 25 | 25 | 34.0 | n/a | n/a |
| 26 | 26 | 33.9 | n/a | n/a |
| 27 | 27 | 34.0 | n/a | n/a |
| 28 | 28 | 34.0 | n/a | n/a |
| 29 | 29 | 34.2 | n/a | n/a |
| 30 | 30 | 34.1 | n/a | n/a |
| 31 | 31 | 34.4 | n/a | n/a |
| 32 | 32 | 33.9 | n/a | n/a |
| 33 | 33 | 33.6 | n/a | n/a |
| 34 | 34 | 33.9 | n/a | n/a |
| 35 | 35 | 33.7 | n/a | n/a |
| 36 | 36 | 33.7 | n/a | n/a |
| 37 | 37 | 33.9 | n/a | n/a |
| 38 | 38 | 34.0 | n/a | n/a |
| 39 | 39 | 34.6 | n/a | n/a |
| 40 | 40 | 34.0 | n/a | n/a |
| 41 | 41 | 33.9 | n/a | n/a |
| 42 | 42 | 34.5 | n/a | n/a |
| 43 | 43 | 34.1 | n/a | n/a |
| 44 | 44 | 34.6 | n/a | n/a |
| 45 | 45 | 34.2 | n/a | n/a |
| 46 | 46 | 34.4 | n/a | n/a |
| 47 | 47 | 33.7 | n/a | n/a |
[2019-08-04 11:01:39.776] speed 10s/60s/15m 1632.5 n/a n/a H/s max 1644.0 H/s
[2019-08-04 11:01:42.260] Ctrl+C received, exiting
[2019-08-04 11:01:42.290] cpu stopped (31 ms)
This is the output of randomx-benchmark:
Running benchmark (100000 nonces) ...
Calculated result: d6660144e9a2e68bf47d7cc8afc206672e72f82dfff69fe0d974531e85f7504f
Performance: 1329.85 hashes per second
Calculated result: d6660144e9a2e68bf47d7cc8afc206672e72f82dfff69fe0d974531e85f7504f
Performance: 1328.99 hashes per second
Calculated result: d6660144e9a2e68bf47d7cc8afc206672e72f82dfff69fe0d974531e85f7504f
Performance: 1327.96 hashes per second
Calculated result: d6660144e9a2e68bf47d7cc8afc206672e72f82dfff69fe0d974531e85f7504f
Performance: 1326.96 hashes per second
Calculated result: d6660144e9a2e68bf47d7cc8afc206672e72f82dfff69fe0d974531e85f7504f
Performance: 1327.62 hashes per second
Calculated result: d6660144e9a2e68bf47d7cc8afc206672e72f82dfff69fe0d974531e85f7504f
Performance: 1326.37 hashes per second
Calculated result: d6660144e9a2e68bf47d7cc8afc206672e72f82dfff69fe0d974531e85f7504f
Performance: 1324.89 hashes per second
Calculated result: d6660144e9a2e68bf47d7cc8afc206672e72f82dfff69fe0d974531e85f7504f
Performance: 1324.93 hashes per second
Total: 10617.57
The text was updated successfully, but these errors were encountered: