Topic: Overclocking/Undervolting RX580 MSI Armor OC 8GB (Micron memory) (Read 477 times)

For ETH your 16384 should be enough, I didn't have issues, only for kernel 2 on phoenix i need double.

generally bigger page has no affect on hashrate  its just for the mining software requirement

Lunga Chung: I meant to ask you what a good Virtual Memory Paging min and max values are in 2019....

In the mining.help guide they say: In the “Initial size (MB)” type: 2000MB x GPU Count (12000MB for 6 GPU mining rig, 24000MB for 12 GPU). Since I have an 8GPU rig I ended up calculating 16000MB. However, this ended up being too low so I have been running on 32000. I am going to assume it is because the guide is out of date and the DAGs these days require more.

Here are my current settings:

Min 32000 (wondering if this value is too high? Does it really matter if it is too high and does it effect power usage)?
Max 64000 (this value doesn't really matter)

either add mV or lower OC what ever you prefer, it all boils down to you  el.power price. Your goal is to find best possible ratio mhs/W (unless you have free powr in which case just crank them up as much as possible )

As for the cards that work stable for more then 24h you can undervolt them by 5mV if they crash after some time add 2mV see if its stable. Let the first week of tuning be a testing week to measure what works best, don't worry about crashes...

Lunga Chung: I think I understand everything now. One last thing I am trying to resolve:

For a while I was running on the following settings with all 8 cards:
CORE:1170 / 850
MEMORY: 2150 / 850

However, GPU4 is still going to 0.00mh/s and failing eventually.

My first instinct was to switch GPU4 settings to the following (increasing the voltage on core and memory by 10):

CORE: 1170 / 860
MEMORY: 2150 / 860

So far things seem to be working... Should I have kept the voltage and decreased the core/memory mhz instead?

core and mem are not correlated in that sense, its entirety up to algo your mining, in this case ETH which is more memory dependent

Go up with mem if you can, and also HW memory errors don't have to be 0, up to 1k a day per card is fine

Lunga Chung:


You mean to keep HW running and check at the bottom section of each GPU and make sure errors are 0, 0, 0 or at a minimum right?

Additionally, are you saying if my cards allow for 2200mhz memory I should definately do it? I feel like without the core being increased beyond 1180 it doesn't take full advantage of it though. Maybe I should slightly increase the volts on the Core and go for 1200mhZ / 870mv.

ETH is all about memory, higher the better.

P.S. dont forget to check for HW errors

Lunga Chung: Great, so instead of 1170 / 850, you are saying to slightly increase and see if there is no crashing, something like 1175. I can guarantee that with my MSI Armor Core OC RX580 8GB cards that with 1180 / 850 and 2150 / 850 certain cards crash, not even sure if it is worth playing around with.

Does the memory make a difference at this point? I am currently on 2150 instead of 2200. I see a small increase when it is set to 2200, but only if the core is set beyond 1180 which requires more mV than 850.

I have Sapphire rx580 nitro+ 4gb samsung not all cards can handle 1180/837mV on core.
I'm sure that not all gpus are failing on 1180/837mV the on that can handle that voltage keep that OC

try with 1175/850mV core if its stable go higher by 5 mhz

Lunga Chung: I have learned a lot in the past 24 hours. I have all 8 cards running at the following settings with no problem:


GPU Core: 1170mhZ / 850mV (gpus are failing when I do 1180 / 850mV, I am curious if upping the memory to 2250mhz and running @ 1180 core would work, I am assuming no)
Memory: 2150mhZ / 850mV

Rig running 255mh/s @ 1053 watts which seems pretty solid when entering these numbers in WhatToMine versus other mhs / powerconsumption ratios.


By the way I found out my issue before with the crashing when setting the Memory to 2200, it is because I had Power Limit set to -30% which I believe overrides the voltage amounts causing certain cards to crash. Now that I have Power Limit setting set to zero for all cards I am able to easily do 2200mhZ / 850mV memory.


At this point I think I am going to keep the memory settings exactly where they are. I noticed it is all about slightly adjusting the GPU Core which will increase megahases but also increase wattage if I feed it a few more volts, for example: 1200 / 870mV did not seem profitable on WhatToMine as it brought my entire rig to 1130 watts.


Since my rig is currently running 255mh/s @ 1053 watts using GPU Core: 1170mhZ / 850mV, I am a bit nervous to lower the volts even further based on your last post.


Isn't 1180mhz / 837mV a bit dangerous for the card? I am sure it will be extremely profitable if it works. I was told you shouldn't go beyond 1180mhz @ 850mV in your other post, is this true?

I am seeing 648 WATTS for all 8 cards, the remainder is my Motherboard, CPU, RAM, SSD etc... I am using the typical celeron processor, 8GB ram, Asus B250 Mining Expert, kingston 120GB SSD.. I am wondering if it is the 7 fans I am running on the frame, maybe they are running too high, or it could be the tt setting in Claymore that I am running.

Here are my start.bat settings: -epsw x -mode 1 -mport 0 -rxboost 1 -tt 70 -ttli 75 -tstop 85 -wd 0

Not sure why my wattage is in the thousands and you are able to get yours down in the 900's. Let me know if you think there is anything better I can do. Thank you!

this is you OC on rx588?

honestly don't see the point in crancking core to 1240mhz the same effect can be achived with 1180mhz/837mV you gain only heat with that core, beside he is mining ETH not XMR

Piskeante: Thank you for the in depth post. I am a bit confused when you say 1mv? By this to you mean 1000mv? Can you try to explain one more time the exact steps I should take to get each of the 8 cards to a stable overclock state. Additionally, I would like to find the perfect power usage/hashrate balance. I am aware I could probably get 32mh/s per card but for another 200 watts I don't think it would be worth it profit wise.

I am currently getting 246mh/s and some change running the entire rig at 1037 watts. Each card is in the 30's and some change.

Is it worth trying to get any better than this? I feel like if I do the voltage will need to be increased from 850 and it will cause the wattage to increase wouldn't make it worth it. Let me know your thoughts. Thank you!

some tips here.

1º DO NOT UNDERVOLT MEMORY. Specially if you are mining ETH. My experience is that ETH is a highly intensive algo that needs to work with ram. If you undervolt errors and crashes can occur.

2º Moreover, some ram needs more voltage to work. If you overclock and at the same time undervolt ALL OF THEM JUST NOT CARING ABOUT DIFFERENCES IN QUALITY, YOU WILL GET CRASHES. Every one of my 12 cards was tested individually with the max stable clocks at 1mv on memory.

3º this action allowed me to push ram to the max getting in some of them 2250mhz on the memory, which makes a huge bump in performance. To be clear here. a card doing 2050mhz mem and 1150 mhz on the core, will do something about 30 ish on ETH. If you can do 2250mhz it will go to 31,4mh/s. If you go for 1240core and 2250mem you can expect around 32 ish.

So, the only way to squeeze more performance is either this two things:

1º bump ram voltage to 1mv and test individually your cards to see how much you can do without too many mem errors. You can allow to have errors, but not much. ideally 0. but you can get around 1000/day. if they are micron, i can almost assure you that with 1mv you may be able to do 2250 on almost all cards, since micron memory is the best for AMD polaris cards. All of my cards with Micron do 2250mhz at 1mv.

2º if you want to get extra performance at the cost of more heat and power consumption, you can try 1240mhz on the core at 881mv or maybe a bit more (depending on ASIC quality of the cards).

there is no other way.

rxboost 30 or 21 is just something people find most stable for rx580

no need for powerlimit if you mV them properly

Lunga Chung, awesome that is what I will do. What is the difference between -rxboost 1 and increasing the number like you are saying? Additionally, can you explain powerlimit?

just mark the cards that appear as 0.00 and put a little more mV on them

Also if using claymore try -rxboost 30 or 21 (no need for straps command if you used PBE for BIOSmod)
If you use PhoenixMiner download amdtweakCLI and use command --ref 10 (same effect as rxboost on Claymore)

Lunga Chung, thank you for confirming all of those things. This is the main guide I used to setup my rig on windows. As you can see the writer of the guide references RX580 w/ Micron memory which is where I got my initial settings that have been working steady. I have been using overdriveNT for overclocking.

GPU Core:
1150mhz / 850mV

Memory:
2050mhz / 850mV

I just feel like there has to be a way to squeeze between 31-32mh/s out of each card. Maybe I should try what you are mentioning: 1180 / 2150 both at 850mV still.

Ok, read this guide https://mining.help/ it basic stuff, also for AMD switch to OverDriveNTool its much better for OC then MSI AB (the guide explains in details how to set it up).

Also regarding BIOS modding this guide is pretty good

Core should no go over 1180 on 850mV
Memory at least on my Micron i see no gain above 2150 (also 850mV)

entire rig should not consume more then 960W

P.S. In my experience when cards hits 00.0 it usually goes back to default clocks and almost always indicates that Core is either OC-ed to much or undervolted to much

Lunga Chung: I was following the numbers in this post: https://www.reddit.com/r/EtherMining/comments/8m3yre/guide_overclocking_and_problem_solving_for_amd_rx/ There is a section that says [special: Micron memory].

These are the settings where all cards except for the 2 cards were 32mh/s and the two cards GPU 2 and GPU 4 were 0.000:
1200mhz / 850mV
2200mhz / 850mV
PowerLimit -30

Then it hit me, I am probably not supplying enough power for these settings. So I changed PowerLimit to 0, and upped to 900mV for both CPU core and memory, still no luck. Powerlimit is still a bit of a mystery to me if someone can explain. Does it cap off the amount of power that can be sent to the card? Doesn't make much sense to me because if 850mV is specified, then wont 850mV be used regardless of power limit?

hustleman: is it common to have 2 cards out of a batch of 8 of the same cards that will not work with the same overclock settings? I believe everytime I restart the computer or adjust the overclock/settings it isn't always the same 2 cardsd that give me probably. It is different 2 cards.

I would like to come up with a game plan to safely find the best sweet spot for all 8 cards. Not sure if I should start with my working settings 1150mhz/850mV and 2050mhz/850mV and slowly increase, or if I should start with the cards max default settings and slowly decrease. In overdriveNT I DISABLE all the GPU’s states except the last one (P7) where I enter the values and additionally for memory I disable all of the states except for the last one where I enter the values.

I know something is wrong when I see the two cards with 0.000, try to close Claymore and all of Windows freezes usually involving a restart. I am assuming this is normal when it comes to overclocking/undervolting if something is incorrect.

Maybe start by increasing your memory voltage to 900mv  Also if you have 2 particular cards that are having memory errors just leave those at 2050mhz asnd increase the other cards by 25 mhz until they start throwing errors, then back them down.