8 RX580 Hashrate Drop

[joemaio87]

Hello everyone,

Long time lurker, finally looking to join and become a master miner. Did some searching but didn't come across any answers. I've got an issue with my settings I believe, where I can get each card to individually mine Ethereum ~31.5-32.5 MH/s, but when I connect all 8 GPUs, only the card connected via HDMI consistently mines at that rate, and the rest are 27-29 each. Here's my setup:

Windows 10 Home
MSI Z270 SLI-PLUS
Pentium G4400
2 4gb Ballistix Memory
128 GB SSD
8 MSI RX580 Gaming X 8GB (5 Micron, 2 Samsung, 1 Hynix)
2 EVGA 850W Platinum Supernova P2 PSU

Just upgraded to the Adrenaline 18.5.1 (from 18.2) drivers and still having same issue.

Dual Mining Ethereum/Sia on Claymore with these settings
-dcri 18 -tt 67 -tstop 85 -fanmin 40 -fanmax 100 -cclock 1200 -cvddc 900 -mvddc 900 -mclock 2250

If I adjust the voltage in any direction, card 3 and 5 have insane memory errors. These are the only settings that have given me zero errors across all cards, they are running 130-135 watts each like this and I'd like to lower if possible. As it sits, I'm running 1485 at the wall, and I'd like to get to at least 1360w to keep the PSUs within 80% of total wattage.

I was able to run 4 cards at a consistent 31.5-32.5 and when I got around to getting all 8 cards put together everything dropped.

Anyone have any suggestions? Appreciate any help!

[SpceGhst]

Dual mining eats up the power, though you should have enough to handle it.  Try mining just eth without Sia and see if hashrate holds.

[joemaio87]

Dual mining eats up the power, though you should have enough to handle it.  Try mining just eth without Sia and see if hashrate holds.

With the 4 cards mining dual, they were all running at 32 MH/s each, so is that to say there's too much power being drawn with all 8?

[SpceGhst]

So many times when dealing with computer problems, you have to find out what’s not causing the problem to find out what is causing the problem.  By process of elimination, you eventually get to the problem.

To find out if it’s a power problem or not, reduce the power usage on the cards.  The easiest way to do that is stop dual mining and only mine Eth.  If mining Eth only works fine, then I’d think it’s a power problem.

[Rasanders22]

Why are you mining SIA? I thought ASIC's killed any profitability. Aren't there better coins to dual mine with? Have you tried re-installing the drivers with all 8 cards inserted?

[joemaio87]

So many times when dealing with computer problems, you have to find out what’s not causing the problem to find out what is causing the problem.  By process of elimination, you eventually get to the problem.

To find out if it’s a power problem or not, reduce the power usage on the cards.  The easiest way to do that is stop dual mining and only mine Eth.  If mining Eth only works fine, then I’d think it’s a power problem.

Same settings, and removed the dcri setting I originally had, Ethereum only has pushed hashrates back up to 32 MH/s with the exception of cards 3 and 5 (Now mining 15 and 10 MH/s) Cards 3 and 5 have insane memory errors again and gpu 2 with some errors. Total Hashrate has declined.

Why are you mining SIA? I thought ASIC's killed any profitability. Aren't there better coins to dual mine with? Have you tried re-installing the drivers with all 8 cards inserted?

Just did fresh DDU and installed most recent driver yesterday. Have been mining Sia because I like the coin, but I can go without mining it if it solves problems , since I'm only getting 5-600 MH/s and earning a few coins a week.

[jillscarbrough]

only the card connected via HDMI consistently mines at that rate, and the rest are 27-29 each.

That was AMD's cards issue. Sometimes one dummy output was enough to solve that. But in other cases, all of the cards or at least cards that troubled need the dummy output (I experienced on it with my AMD cards).

But may it will a bit different if you were running dual-mining. As other said, doing single mining will be better on your rigs stability.

[SpceGhst]

Reduce memory clock on cards 3 & 5 down to 2100 and see what happens.  All cards are not created equal.

[joemaio87]

Reduce memory clock on cards 3 & 5 down to 2100 and see what happens.  All cards are not created equal.

Same hashes as originally stated. 32 on GPU with HDMI, 27-29 on all others. Zero memory errors. I actually dropped card 2,3,5 to 2150 - these are the 2 samsung and 1 Hynix memory cards.

EDIT: Tried changing voltages of other cards and leaving card 2,3,5 at 900 and they go back to abnormal memory errors.

[SpceGhst]

Sounds like you need to play with each card individually to get best settings as you should get over 30 on all of them with Eth only.  Does each GPU hold the 1200 core setting, or do they bounce around?  If they drop lower, you may need to increase core voltage for that particular card.

[joemaio87]

Sounds like you need to play with each card individually to get best settings as you should get over 30 on all of them with Eth only.  Does each GPU hold the 1200 core setting, or do they bounce around?  If they drop lower, you may need to increase core voltage for that particular card.

What's the best way to go about this? Only leave one plugged in at a time? Using MSI Afterburner to make the adjustments or Claymore parameters. Currently getting my readings from HWinfo. Thanks for your input so far.

[jillscarbrough]

Reduce memory clock on cards 3 & 5 down to 2100 and see what happens.  All cards are not created equal.

Same hashes as originally stated. 32 on GPU with HDMI, 27-29 on all others. Zero memory errors. I actually dropped card 2,3,5 to 2150 - these are the 2 samsung and 1 Hynix memory cards.

EDIT: Tried changing voltages of other cards and leaving card 2,3,5 at 900 and they go back to abnormal memory errors.

Have you switched your cards to compute mode? Just in case you forget about that. You are using Adrenalin, right?

[joemaio87]

Reduce memory clock on cards 3 & 5 down to 2100 and see what happens.  All cards are not created equal.

Same hashes as originally stated. 32 on GPU with HDMI, 27-29 on all others. Zero memory errors. I actually dropped card 2,3,5 to 2150 - these are the 2 samsung and 1 Hynix memory cards.

EDIT: Tried changing voltages of other cards and leaving card 2,3,5 at 900 and they go back to abnormal memory errors.

Have you switched your cards to compute mode? Just in case you forget about that. You are using Adrenalin, right?

All set to compute, and update to latest adrenaline 18.5

[xxcsu]

its can be psu, power related , 8 of those cards can pull way more than 1700 watts , plus your system ...
Read this article about real power usage, this is my own personal experience
If you connecting 6 or 7 cards, then you get a good hashrate ?
If the ambient temp is high where your mining rig located, your cards running hotter , they pulling more power...
What are you using to measuring the power draw at the wall ?
What voltage that rig is running from ?

[joemaio87]

its can be psu, power related , 8 of those cards can pull way more than 1700 watts , plus your system ...
Read this article about real power usage, this is my own personal experience
If you connecting 6 or 7 cards, then you get a good hashrate ?
If the ambient temp is high where your mining rig located, your cards running hotter , they pulling more power...

Now that I've switched to Ethereum only, power usage dropped to 1085 watts at the wall, according to the KillAWatt. Rig is in a wide open air conditioned room 30x20ft, plenty of ventiliation and no furniture. They are running cool. I'm no longer convinced it is PSU related.

[xxcsu]

Its can be improper bios mod , or bad undervolting/underclocking.
Try to adjust -cvddc 900 -mvddc 900 a little bit higher
-mclock 2250 a little bit lower ?
What kind of bios mod you have ? paste and copy mem starps ?
8 MSI RX580 Gaming X 8GB (5 Micron, 2 Samsung, 1 Hynix) you can not use the same mem starp mod for all of those cards
Yes you have a power related issue for sure, its maybe not your psu , but the improper way to undervolt your cards, or its can be a bad mem strap mod.
Try to run your cards without any voltage adjustment... just with the bios mod and see your hashrate , you might can not run all 8 cards without undervolting so try 6-7 and see how stable is, how is the hashrate ... i also would remove the -dcri setting from the bat file , and let the miner run on default settings if you mining only ETH

[jillscarbrough]

All set to compute, and update to latest adrenaline 18.5

I just read back all of OP's post above, seem like all those troubled cards were only down a bit its hashing speed. Have you tried to give dummy output on to that troubled cards?

As you said:

Long time lurker, finally looking to join and become a master miner. Did some searching but didn't come across any answers. I've got an issue with my settings I believe, where I can get each card to individually mine Ethereum ~31.5-32.5 MH/s, but when I connect all 8 GPUs, only the card connected via HDMI consistently mines at that rate, and the rest are 27-29 each.

[androstan1234]

In my experience, use dummy plug only on igpu (for remote access).  Set igpu as default in mobo bios.

Make sure "above 4g decoding" is set in mobo bios.

Make sure paging file is at least 24GB, the more the better.

[joemaio87]

So after some tinkering, these are my results:

Switched to Ethereum Only
Removed -dcri option

Card 2=Samsung
Card 3=Hynix
Card 5=Samsung

-I connected 1 GPU at a time, flashed the original BIOS, removed and reinstalled atipatch.
-Restart.
-Reflash Samsung modded BIOS to each card, achieve 32 MHs each time.
-Same procedure with Hynix memory, except changed BIOS to "One Click" from Polaris.
-No Memory Errrors

-Connect 4 cards (including 1 Micron now) and Memory errors start pouring in with lower hashrate.
-Disable 1 GPU in Claymore at a time, and reference to the # when all 8 are plugged in. Alter voltages and memory, re-enable until no memory errors.
-Copy results over to 8gpu Batch file

-Total Hash 245.8MH/s now, 3 cards @ 32, 2 @ 30, and 3 @ 29.
-1180 Watts at the wall
-GPU5 2 Memory Errors, GPU3 4 Memory Errors in 15 mins

-cclock 1200,1200,1150,1150,1150,1200,1200,1200
-cvddc 900,900,900,900,925,900,900,900
-mvddc 900,900,900,900,925,900,900,900
-mclock 2250,2100,2100,2050,2150,2250,2250,2250

Took a while to wrap my head around 2 cards with the same memory needing different settings, doesn't make much sense to me.

Would REALLY love to get that extra 10-12 MH/s by gettin each card to 32MH/s, but if I spend anymore time on "That damn computer machine" my wife may divorce me  

[SpceGhst]

Hey... sounds like you’re getting there.  I’d still lower the mem clock on the two cards getting memory errors, by like 25... I don’t like errors, and it won’t change hashrate that much.  As for overall rig speed, don’t get caught up in trying to squeeze every little bit you can.  Most people say to find the max then back off 5-10%.  You’re equipment will last longer and won’t burn out so soon.  Besides, 245mhs isn’t so bad...