Author

Topic: Does system memory speed matter for Cryptonight? (Read 189 times)

newbie
Activity: 14
Merit: 0
February 01, 2018, 09:06:17 AM
#3
My understanding - which has been borne out by testing on 2 different machines - is that the only memory that matters with CPU mining Cryptonight is the L3 cache; the faster it is, and the more of it, the faster Cryptonight can be mined.

FYI, the ideal number of threads for mining Cryptonight is L3 size in MB / 2. For example, the Ryzen 5 1600 has a 16MB L3 cache so the ideal number of threads is 8.

Also, avoid the use of Core 0 / Thread 0 if you run Windows 10 and need to use the computer for other things.

I highly recommend XMR-Stak in CPU-only mode and posted below is the relevant part of my cpu.txt file that sets up the thread configuration for my Ryzen 5 1600 and which gets around 440 H/s on average with minimal impact to usability:

Code:
"cpu_threads_conf" :
[
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 1 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 3 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 4 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 5 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 7 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 9 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 10 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 11 },

],

This configuration was figured out with the help of PM conversations between me and fellow user @Set Ready Go.



Thank you for the great information.  I was hoping someone would respond who has experience with the Ryzen 5.  I was planning on trying it as a dedicated CPU miner to get a benchmark hash rate and then try it with windows and a few GPU's mining something other than Cryptoknight (Not sure how much impact GPU mining will have on CPU mining).  I will read up on XMR-Stak as i noticed on monerobenchmarks.info that most of the "good" benchmark entries were using it.
full member
Activity: 420
Merit: 184
My understanding - which has been borne out by testing on 2 different machines - is that the only memory that matters with CPU mining Cryptonight is the L3 cache; the faster it is, and the more of it, the faster Cryptonight can be mined.

FYI, the ideal number of threads for mining Cryptonight is L3 size in MB / 2. For example, the Ryzen 5 1600 has a 16MB L3 cache so the ideal number of threads is 8.

Also, avoid the use of Core 0 / Thread 0 if you run Windows 10 and need to use the computer for other things.

I highly recommend XMR-Stak in CPU-only mode and posted below is the relevant part of my cpu.txt file that sets up the thread configuration for my Ryzen 5 1600 and which gets around 440 H/s on average with minimal impact to usability:

Code:
"cpu_threads_conf" :
[
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 1 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 3 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 4 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 5 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 7 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 9 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 10 },
    { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 11 },

],

This configuration was figured out with the help of PM conversations between me and fellow user @Set Ready Go.

newbie
Activity: 14
Merit: 0
I just purchased a Ryzen 1600 and have a BIOSTAR TB350-BTC motherboard on the way from Newegg and want to try cpu mining (I currently have two GPU rigs and if this does not work out I can use the MB to make a 3rd rig).  I can't seem to find a definite answer to whether the type of system ram makes a difference in hashing rates.  Will a cheap 4GB DDR4 2400 Kingston stick produce the same results as say a faster dual channel low latency pair?
Jump to: