Hello Doktor
Thanks for the initiative. Single core performance on RandomXL seems good. However there are some error messages thrown up and multi-core performance is poor at the moment. Can you help? The test system is Dell R620 server with two Xeon E5-2620 V1 and 28 GBytes of RAM.
First the error messages:
Cannot bind database to node Cannot bind cache to node See https://drive.google.com/file/d/1o8AnMtIU3J9WstytBRktOa-L9f_C0BeN/view?usp=sharing
Cannot bind scratchpad to node See https://drive.google.com/file/d/18G_bTBlkV3hUXC-D9X0sMrui1dSmjkFe/view?usp=sharing
The errors appear if I use the wizard to create the BAT file or make my own BAT file.
How can these errors be fixed please?
The performance issue seems to be that the second CPU is not utilised even though SRBMiner identifies 12 cores and 24 threads.
See pics from HWInfo of the two cores
CPU 0 https://drive.google.com/file/d/1IhRbu9z1F1-DWaJpz7o3dyQvazSOe3VM/view?usp=sharing
CPU 1 https://drive.google.com/file/d/1C2s1PBm9hmzNAv91Jrsk2ZiWGlvW14Va/view?usp=sharing
Here is the command line used SRBMiner-MULTI.exe --algorithm randomxl --cpu-threads 0 --pool loki.miner.rocks:5005 --wallet L7f9SMpr43wNERBxCngFyqMoJvkMax8xSitiJY335ra12ucbiEFzgqKfHJg1Q9y58RGMi3TyGvPC94e 6w3zYKK9i9PHmDDr --cpu-threads
Finally I can't see documentation on how to use CPU affinity and --cpu-priority. Please can you add it?
Hi, first try removing the --cpu-threads 0 because that sets the number of threads to 1
Code:
SRBMiner-MULTI.exe --algorithm randomxl --pool loki.miner.rocks:5005 --wallet L7f9SMpr43wNERBxCngFyqMoJvkMax8xSitiJY335ra12ucbiEFzgqKfHJg1Q9y58RGMi3TyGvPC94e6w3zYKK9i9PHmDDr
You can also try to find the best --cpu-threads by hand for your system.
--cpu-priority is just a number from 1-5, the higher the priority level of a process, the more will be its assigned processor time.
--cpu-affinity is a mask (can be used hex like 0x.... or just the decimal number). An affinity mask is a bit mask indicating what processor(s) a thread or process should be run on by the scheduler of an operating system.
Hello Doktor
I tried SRMMiner 0.1.5 on a 2 core Celeron machine. It worked first time on the RX.MINEXMR RandomX TestNet. No errors. I followed your instructions for enabling huge pages. That also worked. Performance was 7% higher than XMR-RIG. Great.
Thanks for the feedback about the Dell R620 server. I started a fresh with your guided-setup.bat and used the RandomX testnet at RX.MINEXMR.
Unfortunately I got exactly the error codes as the first time + the second CPU is not being utilised.
I was able to utilise the second CPU by adding --cpu-affinity 0x555555 . That problem is solved. . Since the miner recognises all the threads and core automatically would it be possible to automatically set the miner up to use all CPUs?
Mining performance of SRBMiner 0-1-5 on RandomX at rx.minexmr.com was initially 92% of the hashrate achieved with XMRIG 4.5.0 Beta but when I ran it overnight hashrate exceeded XMRIG by 5%. Not bad for your initial release. I tried adding additional threads but it made no difference. Whereas XMRIG actually goes slower on these CPU when more than 1 thread per core is used.
The Celeron and the Dell R620 run Windows 10 Professional.
How about the binding errors? Do they matter? Perhaps additional performance could be achieved without these errors?
See links below. Thanks in advance for your help.
Screenshots of errors: Cannot bind to XXX errors
https://drive.google.com/file/d/1Mo6NM_wMCjEEB1nyF6Z3OSuPw2XItVFi/view?usp=sharing
https://drive.google.com/file/d/1XnHKjVxDX7qkFwbjIudWVE_0CyJbtUJo/view?usp=sharing
https://drive.google.com/file/d/1pas-ckr-JiVyvtJfA0HktwfhyE0WTZgX/view?usp=sharing
FOOTNOTE
I also have an R9 Fury on the Celeron machine. When I run SRBMiner 0.1.5 in parallel with Claymore 15 mining ETH I got a small speed improvement of 0.8% on Claymore 15! Claymore reduced the performance of SRBMiner on RandomX by 15% when run in parallel.