When setting GPU Core and Memory voltage is it considered Max voltage or Exact

[dipepmining]

I am trying to understand when using overdriveNT tool on Windows 10 when setting the voltage of GPU Core and Memory does it mean Max voltage or the exact voltage that will consistently be sent to the card.

In Claymore I am currently getting this message for one of my GPUs:

GPU #2 got incorrect share. If you see this warning often, make sure you did not overclock it too much!

GPU: 1150 / 850
Memory: 2150 / 850

Could increasing the voltage for this card fix this warning that I am getting? Or is it strictly to do with Memory mhZ or is it both?

Should I start by taking the Memory down to 2145 / 850?

Is there a scenario where increasing the GPU speed could help prevent memory warnings at 2150mhz? Or is it strictly lowering values from this point forward for that card.

[Metroid]

This is not voltage, your memory chip can't stand 2150 for a long time without errors, increasing memory voltage will not help that, you must reduce memory clock and those errors will stop, memory is like core, lottery based if you are lucky.

Another way to help with that is making sure memory modules are not overheating, some graphics cards have memory cooling linked to core cooling, so setting fans at greater rpm speed will help with cooling the memory modules, try it, it might work for you.

[Piskeante]

there is a very spreaded fact that says that memory can work undervolted. My experience is that THAT DOES NOT WORK AT ALL.

To begin with, every card that i've seen is 1000mv stock on the memory (2000mhz on 580's , 1750mhz on 570's), so 850mv is an undervolt that can cause mem to not have enough power supply and crash. Moreover, this is going to happen if you overclock your memory.

So i bet if you leave it stock to 1000mv you will see less or nearly no errors.

You haven't told us your card,  asic quality, memory brand , etc, so not much i can tell you with this info, just what i said.

1150mhz for 850mv on the core, is also low. Most RX 580's will need around 887mv to work without issues though 850mv is possible for the core. I had like 6 of them around that voltage on the core (from 850-887mv). If the miner crashes, and reinitalises is a core that's too much undervolted. If you get too much errors , or your hash rate is low, or you get incorrect shares, that's normally mem issues.

go for 1150mhz on the core at 887mv
go for 2150mhz on mem with 1000mv

If with this parameters, you still get errors using HWinfo64 (or 32), this means your mem should go underclock. Then try 2125mhz and see. You must stay near 0. some errors, (small amount, can be accepted, less than 100 every hour). if it continues with errors , then 2100. This should be ok. all of my 12 cards could do at least 2100.

Normally, with micron mem you will probably do 2250mhz (all of my micron could do it, best mem for AMD). Hynix is tricky. Some will , and some will not get there. Samsung memory is really bad on AMD cards (the best on nvidia BTW).

i used custom bios + MSI afterburner to apply undervolt.

[Metroid]

The truth about "memory undervolt" is a myth,  people only undervolt the memory because claymore used or still use memory voltage linked to core voltage, so if you dont undervolt the memory, your core clock will be like the memory voltage and that means furnace will hit your house ehhe, as long as few software use the memory to be linked to core voltage then people will always think memory voltage is undervolted which is not true, setting your memory to 1.0v or 0.8v  or 1.2v will not have any effect at all in the memory itself. So the truth is memory undervolt or overvolt is useless on polaris. The right approach supposed to be the use of only -cvddc.

[WhyMe]

I always set core and mem voltage at the same value. If not, the higher voltage is applied on both.
And always starting at 900 and then adjusting per card.

[dipepmining]

Thank you for all the info guys.

I am using MSI Armor OC RX580 Cards with Micron memory.


7 of 8 cards have been running fine on the following OC settings:

1150mhZ / 850w
2150mhZ / 850w

Each card runs at about 80w, rig total is 1060w

The one card is giving me problems, I even lowed it to 2100mhZ / 850w and still giving me the red error.

As Piskeante is saying I think if I feed it a small amount of additional voltage the errors will go away.

Additionally, Piskeante there is absolutely no way gpu mining would be even the smallest amount profitable running at those high voltages you are talking about. If my wattage goes anything higher then it is the profits get destroyed. I have seen plenty of posts out there saying 850w is a common wattage for RX580s and it has been working for me just fine, besides this one card, so I am asking how you guys recommend I go about tweaking it. I do think adding a small amount for voltage will do the trick, I am going to try that now.

[Metroid]

I always set core and mem voltage at the same value. If not, the higher voltage is applied on both.
And always starting at 900 and then adjusting per card.

Sadly, that is the right thing to do.

[dipepmining]

Here is an update with my solution for anyone interested in this post:

In regards to "7 of 8" RX580 cards running well using:
1150mhZ / 850mv
2150mhZ / 850mv

The 1 card that was failing, not able to accept any shares, and causing this warning in Claymore:

"GPU #2 got incorrect share. If you see this warning often, make sure you did not overclock it too much!"

Solution: It simply can't run 2150mhZ / 850w without a small core boost.

It works perfect with:
1170 mhZ / 850mv
2150 mhZ / 850mv

I am now running all of my cards at 1170 / 2150 / 850mv.

There was one other card where the mh/s would drop to 0.000 after the first few minutes of running. This is usually voltage related, so I simply updated its settings to the following and it runs no problem:

Now I have 7 cards running @:
1170 mhZ / 850mv
2150 mhZ  / 850mv

1 card running @:
1170 mhZ / 851mv
2150 mhZ  / 851mv

CONCLUSION (could be extremely helpful to people using RX cards and Claymore):

1. If you are receiving the Incorrect Share error for a specific GPU without any memory errors, it is most likely because your GPU Core speed isn't high enough for your memory speed. For example (GPU: 1150 / 850mv Memory: 2050 / 850mv) might work, but (GPU: 1150 / 850mv Memory: 2150 / 850mv) might not work. So the solution is something like (GPU: 1170 / 850mv Memory: 2150 / 850mv).

2. If you are receiving the Incorrect Share error or seeing actual memory errors for a specific GPU, it is most likely because your Memory speed is too high, try lowering it.

3. If you are seeing a specific card's mh/s drop to 0.0000 and die, this is most likely a voltage issue, usually core voltage. Try adding 1-5mv to both the GPU and Memory to that specific card just like I did above. Now if more than one card is failing, then that means your speed/voltage ratios are probably way off and you need to increase/decrease items by larger amounts.

[Metroid]

Here is an update with my solution for anyone interested in this post:

In regards to "7 of 8" RX580 cards running well using:
1150mhZ / 850mv
2150mhZ / 850mv

The 1 card that was failing, not able to accept any shares, and causing this warning in Claymore:

"GPU #2 got incorrect share. If you see this warning often, make sure you did not overclock it too much!"

Solution: It simply can't run 2150mhZ / 850w without a small core boost.

It works perfect with:
1170 mhZ / 850mv
2150 mhZ / 850mv

This makes no sense. The incorrect will continue till you decrease memory clock but it will happen sometimes. You think what you did worked but that is not true. Every time you reset your computer memory behaviour also changes, maybe up to now you did reset your computer in times where that gpu showed no incorrect memory because its memory were fit perfectly, you might not have the same luck next time you reset your pc.

Anyway, you will find out sooner or later hehe, It's a learning process hehe

[io8621]

I proceed like this, I don't know if it's okay but I can have quite stable results.

Example with one Sapphire Rx570 Nitro 4 gb Hynix, vga bios coming with 2 different chip strap, Elpida and Hynix but my card is equipped only with Hynix H5GC4H24AJ.

I tested a lot of strap but memory limit without error is in range 2045 - 2050 mhz, acceptable mem error until 2070 mhz more than 2070 i can not go without thousand of error.

Overclock and undervolt from Phoenix Miner or with OverdriveNT with the same result.

Ref parameter is set to 30.

1100 - 2050  [900 - 900]  30.65    86w
1110 - 2050  [900 - 900]  30.7      86w
1120 - 2050  [900 - 900]  30.8      86w
1130 - 2050  [900 - 900]  30.95    86w
1140 - 2050  [900 - 900]  31.3      86w
1150 - 2050  [925 - 900]  31.55    90w
1160 - 2050  [925 - 900]  31.6      92w
1170 - 2050  [925 - 900]  31.65    92w
1180 - 2050  [925 - 900]  31.65    92w

Clock > 1160 with ref 30 dont give me more hs so i set to 1150 - 2050 [ 925 - 900 ] for ~ 90w and go stable for days without error.

I can go down with core to ~915mv but after some time go to 0 h/s and restart.

Some question:

• for cvddc i made test with increase of 10mv each step, but effective cvddc in hwinfo is always a few mv less, exist a precise way to increase or decrease the core voltage? 6 mv step or 8 mv step or other?
• plus or minus mv to memory affect memory error?

[Metroid]

The more you decrease the memory clock the more likely you will get no errors, that does not mean you will never get memory errors if your memory clock is greater than manufactures set speed, which for polaris, 1750mhz, elpida and hynix, 2000mhz for samsung. Also errors depend on many factors, one means luck. You memory could have millions of memory errors and still the miner shows no memory error shares and another event is no memory errors at all and the miner says memory errors shares.

If miner shows memory errors shares then just reset it, if the same gpu keeps showing memory errors after few resets then decrease the memory clock.

[WhyMe]

I proceed like this, I don't know if it's okay but I can have quite stable results.

Example with one Sapphire Rx570 Nitro 4 gb Hynix, vga bios coming with 2 different chip strap, Elpida and Hynix but my card is equipped only with Hynix H5GC4H24AJ.

I tested a lot of strap but memory limit without error is in range 2045 - 2050 mhz, acceptable mem error until 2070 mhz more than 2070 i can not go without thousand of error.

Overclock and undervolt from Phoenix Miner or with OverdriveNT with the same result.

Ref parameter is set to 30.

1100 - 2050  [900 - 900]  30.65    86w
1110 - 2050  [900 - 900]  30.7      86w
1120 - 2050  [900 - 900]  30.8      86w
1130 - 2050  [900 - 900]  30.95    86w
1140 - 2050  [900 - 900]  31.3      86w
1150 - 2050  [925 - 900]  31.55    90w
1160 - 2050  [925 - 900]  31.6      92w
1170 - 2050  [925 - 900]  31.65    92w
1180 - 2050  [925 - 900]  31.65    92w

Clock > 1160 with ref 30 dont give me more hs so i set to 1150 - 2050 [ 925 - 900 ] for ~ 90w and go stable for days without error.

I can go down with core to ~915mv but after some time go to 0 h/s and restart.

Some question:

• for cvddc i made test with increase of 10mv each step, but effective cvddc in hwinfo is always a few mv less, exist a precise way to increase or decrease the core voltage? 6 mv step or 8 mv step or other?
• plus or minus mv to memory affect memory error?

Your watts are just impossible with these voltages, if we are talking about ETH

[Metroid]

Your watts are just impossible with these voltages, if we are talking about ETH

That is more like 92 + 50 watts = 140w or around 170w or so, remember that many apps dont show the true used power. The 92w is just the core watts. The rx 580 uses a lot more than the rx 480 at same voltage, rx 480 around 68w - 75w at 900mv, rx 580 85w - 95w 900mv.

[io8621]

Yes, i know,  this is only core watt like hwinfo report.

[WhyMe]

Your watts are just impossible with these voltages, if we are talking about ETH

That is more like 92 + 50 watts = 140w or around 170w or so, remember that many apps dont show the true used power. The 92w is just the core watts. The rx 580 uses a lot more than the rx 480 at same voltage, rx 480 around 68w - 75w at 900mv, rx 580 85w - 95w 900mv.

Strange. Same model for 480 and 580 ?

Talking about partial power has no sense. Just power at wall is a true value.

[Metroid]

Your watts are just impossible with these voltages, if we are talking about ETH

That is more like 92 + 50 watts = 140w or around 170w or so, remember that many apps dont show the true used power. The 92w is just the core watts. The rx 580 uses a lot more than the rx 480 at same voltage, rx 480 around 68w - 75w at 900mv, rx 580 85w - 95w 900mv.

Strange. Same model for 480 and 580 ?

Talking about partial power has no sense. Just power at wall is a true value.

I'm talking about at the wall. The rx 5xx series are much more stable because at same voltage they use more power and most users think that is not the case, seeing is believing ehhe

[dipepmining]

Metroid, I have to admit you were right. After running for 24h thinking I was in the clear, what do you know:

GPU2 not accepting any shares (zero men errors though).

I have 8 of the same exact RX580 cards..

7 of them are running fine at

1070 / 850
2150 / 850

I wonder how much I am going to have to lower the clock on GPU2 so it doesn't show that warning.

I started at 2100 / 850, when I was doing 1150 / 850 for core and was still getting the warning. But without a PC restart so maybe I had invalid results.

I guess I will lower by increments of 10, restart the PC and start the miner again until it goes away. If all the GPU models are the same I shouldn't need to lower more than 50mhz on this specific card should I?

You guys seem positive the issue has nothing to do with Core or Voltage when this warning appears with no memory errors so I will focus on Memory speed only.

[Metroid]

Metroid, I have to admit you were right. After running for 24h thinking I was in the clear, what do you know:

GPU2 not accepting any shares (zero men errors though).

I have 8 of the same exact RX580 cards..

7 of them are running fine at

1070 / 850
2150 / 850

I wonder how much I am going to have to lower the clock on GPU2 so it doesn't show that warning.

I started at 2100 / 850, when I was doing 1150 / 850 for core and was still getting the warning. But without a PC restart so maybe I had invalid results.

I guess I will lower by increments of 10, restart the PC and start the miner again until it goes away. If all the GPU models are the same I shouldn't need to lower more than 50mhz on this specific card should I?

You guys seem positive the issue has nothing to do with Core or Voltage when this warning appears with no memory errors so I will focus on Memory speed only.

That is the right thing to do, if 2150 is giving errors then 2140 and so on until you dont see that error anymore, also lower your core clock back to 1150mhz and try lower voltage, increments of 6, 844mv, if it worked without problem at 1170 850mv then at 1150mhz even 838mv will not be a problem, lower votage less heat and less heat even if is 3 watts is a lot for cooling and will improve memory temps and improving that 2150mhz for memory might work just fine.

Also like I said before, increase fan speed will also improve temps which will in turn make memory error shares for that gpu to disappear. There are many things to try.

[dipepmining]

Metroid: Should I always keep gpu core voltage inline with memory voltage?

For example if I set GPU Core voltage to 844 and it works should memory voltage be 844 as well or keep it at 850?

Additionally, if memory at 2200 / 850 works for some of the cards without any warning messages or mem errors since they are all Micron would you suggest I keep it at that or was 2150 / 850 better?

One more thing... Here are the results when running at the following settings:

When I run:

1150 / 844
2200 / 850
I am getting 253mh/s steady (at slightly lower wattage than when using 850 which is great)

When I run:

1150 / 844
2150 / 850
I am getting 252mh/s steady (at slightly lower wattage than when using 850 which is great, but not sure if it is worth the 3mh/s sacrifice, will need to check whattomine)

When I run:

1160 / 844 (this is probably a bad idea)
2150 / 850
I am getting 253/4mh/s steady

When I run:

1170 / 850 (not sure if this will work at 844, is it worth trying, I am nervous)
2150 / 850
I am getting 255mh/s steady (I still think these are my most solid settings, even the PC restarts faster with them, if that is a thing)

When I run:

1180 / 850
2200 / 850
I am getting 258mh/s steady (but at the cost of higher wattage and risk of a gpu mh/s dropping to 0.000 crashing due to not enough voltage)

I am almost positive without a higher core set, the 2200 can't take full advantage based on the results above.
----------

PLEASE let me know your thoughts on finding the sweetspot based on the above tests if you dont mind. (taking into account that my electricity costs are .1333, which I think is on the high end)

SIDE NOTE: After increasing minfan speed to 50% the gpus all run a lot cooler and GPU 2 that was having problems is accepting shares, even at 2200 / 850 memory. Is this actually a thing or am I most likely going to get warnings again resulting in me to go back to 2140 / 850?

[Lunga Chung]

don't be afraid to experiment  

I'm running 1180/837mV on core, the worst thing that can happen to GPU is to stop mining

Every card that can handle 2200 keep it OCed like that....

I see you use Claymore maybe you could try Phoenix with AMD MemTweak tool just for comparation.

Config Phoenix and run AMT with --REF 90 i get better stability and hashrate then with Claymore (even lower consumption)



Regrading cooling, you are not mistaken most of Polaris cards have temp. sensors on  core only (and some cards don't even have a aluminium plate for additional cooling of ram), memory can get pretty hot to try to keep the rig in the range of 60C you'l get more stable hashrate