Open Isuress opened 7 years ago
Your GPU (AMD 6870) isn't supported in the latest AMD OpenCL drivers anymore, so it won't be listed via clinfo without a driver. You have to install the corresponding open source mesa driver to get your card working, but you will have to look what kind of performance the drivers will have. On Fedora/CentOS/RHEL this would be mesa-libOpenCL but for Ubuntu I don't know the exact name without a running Ubuntu. Maybe this wiki can help you.
It depends on the L3-Cache of your CPU which performance your output will be. Running the initial config.txt will post you in the right direction with a good suggestion. E.g. if there are only 4 of 8 cores listed you will have a too low L3 cache to run all 8 cores together for cryptonight mining. Cryptonight does need exactly 2MB per running core. If you are not quite sure how many L3 cache your CPUs have, you can take a look in the Intel documentation.
No kidding? Damn. I'll have to google the MESA driver. I could potentially use an older version of Ubuntu maybe? I'm using 16.04.2 at the moment but I could switch to Ubuntu Server 14.04.5? Would that have the older driver pre-installed that would all OpenCL to work? I'll maybe give it a try tomorrow. I'll have probably tried it before you have a chance to respond but any and all advice is welcome. (Thank you by the way)
From a quick google search, it seems that my Intel Xeon E5630 CPUs (DL320G6) have 4 cores (with 2 logical each) with12MB of SmartCache; while my Intel Xeon L5630 have (ML370G6) have the same as above. If what you're saying is true, then there should be more than enough L3-Cache to go around? If I have x2 L5630s in a dual-proc server, then it should have 24MB of L3-Cache; which would be enough. Unless the logical cores also need their own share of the cache? Maybe I'm not doing a setting correctly? I still get the NMAP errors with relation to the pool memory and whatnot even though I made the change the developer suggested in the page file. Unless I did it wrong?
I wonder if the other miners that are available out there would give me less issues for equal to or greater than hash/s.
@Isuress I using ubuntu 14.04 for my old amd gpu, this should work for you too. If you are using the old ubuntu you need to google and add package source for the compiler gcc 5.1+ because the system compiler is not supported. For the cpu version use the current dev branch with hwloc and xmr-stak will show you the optimal config at the first start.
@Isuress
Ubuntu 14.04 should go, but 16.04 should also work with the right drivers installed.
As you can see your bottle neck is your L3-Cache. You have 12 MB for 8 cores so you need 8 x 2MB = 16MB L3 cache to run all cores at optimum performance. But your cache size is only 12MB, so you can only run 6 cores (6 x 2MB) at the maximum performance just like you already discovered. The 2MB is from the crypto algorithm specification and is not a requirement of xmr-stak-cpu, so other miners could not give you (much if any) more performance. And no, with 2x CPUs you don't have automatically enough L3 cache to run 8 cores. You have to make sure that at least 4-6 cores run on the first CPU and the remaining 2 run on the other processor. L3 cache is different than RAM and is only available and access-able from the corresponding CPU.
And for the NMAP problem: do you use a graphical environment and start the cpu-miner there? If so, you also have to edit the file /etc/pam.d/common-session
and add the line session required pam_limits.so
. After that you have to logout and login again or reboot.
Hello,
$1 I installed the AMDpro driver following the link you posted and their instructions. I cloned and compiled your program. I had to copy the
config.txt
from the source directory into the bin directory because it wasn't created. Here's my config:When I go to run the program, I get the error:
I've read through other people's issues and googled the errors. I've tried some of the things people suggested but didn't get a fix. Out of curiosity, I ran
clinfo
(had to apt-get it first) and got this:From the looks of it, my CPU has an integrated GPU. I attempted to go to the BIOS and disable it but there was no such option. Instead I opted to change the "primary display device" to PCIE and hoped that would work. It didn't. So it seems that for whatever reason, my Ubuntu has decided that the Intel GPU is the first and only platform to be used. I ran the
lspci -vnnn | perl -lne 'print if /^\d+\:.+(\[\S+\:\S+\])/' | grep VGA
command and got:So the card is there and the OS sees it, but for whatever reason it's not using it accordingly. What should I do? I'm currently using your CPU miner on some other servers and it's working just fine.
$2 Well, sort of fine actually... I might as well post it here 'cause hopefully you'll be reading this anyway. So I have an HP DL320G6 with a Intel Xeon E5630, running ESXI and in Ubuntu Server 1604.2 in a VM. The E5630 has 8 Logical Processors so I set the VM to have 8 cores. When setting up the
config.txt
, I created 8 CPU_AFFINITY related entries and then ran the program (after doing the page pool size, etc - which actually still gives me errors btw). I only got like 120hash or something like that. I went to theconfig.txt
and changed the 8 CPU_AFFINITYs into 6 of them; and then ran. My hash then went to 220hash. How come using 6 gets a better has than 220? (This is only the first part of this question)With all that said, I have an HP ML370 with Dual-Proc Intel Xeon L5630 (2 CPUs). This means I have 16 Logical Processors instead of 8 (as noted by ESXI). I changed the VM's CPUs from 8 to 16. I did all the same setup and config as mentioned prior; except I changed the CPU_AFFINITY from 6 CPUs to 16 CPU entries. My hash was only like 160? I changed my VM's CPUs to 8, and then used the same config as my DL320G6. Now I get 220ish hash. What is going on? How come it's not making use of the 10 extra cores available to it? What should I do? Is there any other information I can provide you to help possibly?
P.S. - If you'd like me to move the second part of this question to the issues section of your CPU miner; I can.
NOTE
CPU_AFFINITY = { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 0 },