Question RTX 2060 memory overclock for mining

[Rajahspice]

Gpu - rtx2060. With default settings when i run miner i have approximately 26.5mh/s and gpu temp goes to 82C Once i i undervolted my gpu from 1.1V to ~850mV, decreased core clock by 500mHz and overclocked gpu memory by 800mHz (it crashes when i am trying to apply more) i got ~ 30.5Mh/s and the highest temperature is now 71C. Have i done something wrong so far?


As i know there is no temp sensor on gpu memory, can i somehow overheat it? I have bad airflow in my case, but as i said i managed to decrease gpu temp to 70 C

 

[Newtonius]

The memory most likely doesn't cause a full computer crash, at most it'll just spit out errors and cause you to start getting invalid shares the more unstable the memory gets. The issue is most likely with your core clock and voltage, it's probably not stable at that level. Up the voltage a little until it's stable at your desired core clock. This is a limitation with the physical silicon itself, nothing you can do about it. It's called the Silicon lottery.

 

[Rajahspice]

The memory most likely doesn't cause a full computer crash, at most it'll just spit out errors and cause you to start getting invalid shares the more unstable the memory gets. The issue is most likely with your core clock and voltage, it's probably not stable at that level. Up the voltage a little until it's stable at your desired core clock. This is a limitation with the physical silicon itself, nothing you can do about it. It's called the Silicon lottery.

Since i "played" with afterburner before your reply i only undervolted it a bit more and created 2 profiles
I've tested these profiles (in both i undervolted GPU to 700mV and it seems like im getting same hashrate as i had on 850mV)
_
1. Core clock (-500) Memory clock (+800) Power limit (73%) Fans: auto
avg GPU temp - 64.8°С avg hotspot temp - 81.7 °С
avg hashrate --- 30.40 mh/s
2. Core clock (-500) Memory clock (-500) Power limit (73%) Fans: auto
avg GPU temp - 62.3°С avg hotspot temp - 78.8°С
avg hashrate --- 27.4 mh/s
these temps from hwinfo and now i am mining with 2nd profile, but after i set custom fan spin speed (85%) temp went down to 56C and i think i am safe here.
So my qeustion is, should i overclock memory by +800mHz to get extra 3mh/s speed (gpu temp will be 59C with maxed out fans OR 65C with fans set to auto mode). Or should i just underclock everything (2nd profile) since i cant monitor my vram temperature in order to stay safe? Thanks in advance

P.S ik that its possible to get over 32mh/s with this GPU but as i said before i dont have good airflow in my case. Temps gonna rise up high if i will turn voltage and core clock back to defaults.

 

[Newtonius]

Since i "played" with afterburner before your reply i only undervolted it a bit more and created 2 profiles
I've tested these profiles (in both i undervolted GPU to 700mV and it seems like im getting same hashrate as i had on 850mV)
_
1. Core clock (-500) Memory clock (+800) Power limit (73%) Fans: auto
avg GPU temp - 64.8°С avg hotspot temp - 81.7 °С
avg hashrate --- 30.40 mh/s
2. Core clock (-500) Memory clock (-500) Power limit (73%) Fans: auto
avg GPU temp - 62.3°С avg hotspot temp - 78.8°С
avg hashrate --- 27.4 mh/s
these temps from hwinfo and now i am mining with 2nd profile, but after i set custom fan spin speed (85%) temp went down to 56C and i think i am safe here.
So my qeustion is, should i overclock memory by +800mHz to get extra 3mh/s speed (gpu temp will be 59C with maxed out fans OR 65C with fans set to auto mode). Or should i just underclock everything (2nd profile) since i cant monitor my vram in order to stay safe? Thanks in advance

P.S ik that its possible to get over 32mh/s with this GPU but as i said before i dont have good airflow in my case. Temps gonna rise up high if i will turn voltage and core clock back to defaults.

Keep overclocking the VRAM until you start seeing issues or inconsistencies in either hashrate or valid shares. If the VRAM starts getting too hot, you'll start seeing either a drop in hashrate or invalid shares. VRAM is very forgiving if you push it too far in overclock, the worst it'll do is just start getting fussy. It's the core (GPU) that's unforgiving, if you try to push it past what it's comfortable with it'll crash the whole system on you.

I pushed my 2080Ti to 1000+MHz on the VRAM and got it up to 65MH/s, though it gave me a few more errors per hour. Went up to 1100 and it started fluctuating wildly from 57MH/s to 63 MH/s. GDDR6 on RTX comes with memory correction, so the more unstable the overclock is the more work the memory has to put into error correcting. And everyone's card works differently, what might be stable for me may be unstable for you. So work around till you find that sweetspot on your card.

 

[Rajahspice]

Keep overclocking the VRAM until you start seeing issues or inconsistencies in either hashrate or valid shares. If the VRAM starts getting too hot, you'll start seeing either a drop in hashrate or invalid shares. VRAM is very forgiving if you push it too far in overclock, the worst it'll do is just start getting fussy. It's the core (GPU) that's unforgiving, if you try to push it past what it's comfortable with it'll crash the whole system on you.

I pushed my 2080Ti to 1000+MHz on the VRAM and got it up to 65MH/s, though it gave me a few more errors per hour. Went up to 1100 and it started fluctuating wildly from 57MH/s to 63 MH/s. GDDR6 on RTX comes with memory correction, so the more unstable the overclock is the more work the memory has to put into error correcting. And everyone's card works differently, what might be stable for me may be unstable for you. So work around till you find that sweetspot on your card.

it is getting clear for me now, when i was mining for 6 hours i had stable 30.4mh/s with +800 memory overclock, so it means i can run it normally without taking risks and with 58-66C GPU's temp?

 

[Newtonius]

it is getting clear for me now, when i was mining for 6 hours i had stable 30.4mh/s with +800 memory overclock, so it means i can run it normally without taking risks and with 60-66C GPU's temp?

Yes it's fine. That temp is mostly for the GPU die itself, not the VRAM. If you can't see the VRAM's temp then that means there's no sensor for it, but normally MAX temp for GDDR6 VRAM is 95C, anything more will cause damage, but these days manufacturers program these devices so if they do reach those temps the system automatically downclocks itself (thermal throttling) to prevent further damage. We live in a convenient era. Sort-of....

 

[Rajahspice]

Yes it's fine. That temp is mostly for the GPU die itself, not the VRAM. If you can't see the VRAM's temp then that means there's no sensor for it, but normally MAX temp for GDDR6 VRAM is 95C, anything more will cause damage, but these days manufacturers program these devices so if they do reach those temps the system automatically downclocks itself (thermal throttling) to prevent further damage. We live in a convenient era. Sort-of....

haha, i hope so)) Thanks for your answers, the last question for you, should i run it with fans set to auto (66% speed) or set custom speed to 85% There is 5°C difference(59C and 64C) I mean this is not such a big temp difference but i can decrease noise.

 

[Newtonius]

haha, i hope so)) Thanks for your answers, the last question for you, should i run it with fans set to auto (66% speed) or set custom speed to 85% There is 5°C difference(59C and 64C) I mean this is not such a big temp difference but i can decrease noise.

As long as the GPU is below 70C choose whatever fan profile fits that agenda. Generally fans can last years before needing replacement even if they're at 100% 24/7, so revving them up as much as you want 24/7 is fine. Cheaper to just need a fan replacement than a whole GPU replacement. But yea, just try to stay below 70 and you're golden.

 

[Rajahspice]

As long as the GPU is below 70C choose whatever fan profile fits that agenda. Generally fans can last years before needing replacement even if they're at 100% 24/7, so revving them up as much as you want 24/7 is fine. Cheaper to just need a fan replacement than a whole GPU replacement. But yea, just try to stay below 70 and you're golden.

Thanks, still a bit worried about this temp difference between GPU and it's hotspot https://prnt.sc/10roo90. But probably just gonna continue mining with this kind of paranoia. Big thanks for you, you significantly decreased my worries

 

[Rajahspice]

As long as the GPU is below 70C choose whatever fan profile fits that agenda. Generally fans can last years before needing replacement even if they're at 100% 24/7, so revving them up as much as you want 24/7 is fine. Cheaper to just need a fan replacement than a whole GPU replacement. But yea, just try to stay below 70 and you're golden.

By the way i just discovered, that whenever i run miner, my memory clock drops by 200 immediately. (whether its underclocked by 500 or overclocked by 800).
So if my default memory clock is 7000 and i underclock it by 500mhz, it will be 7000-500-200, and if i overclock it by 800mhz it will be 7000+800-200. And hashrate is stable on both 6300mHz and 7600mHz.
What can be the issue?

 

[Phaaze88]

Memory thermal throttling. What you see with the gpu core temp, the Vram is definitely going to be hotter than that while mining.
You can't see the temp of the Vram with the tools you currently have. Only way to see it now is with a thermocouple or something similar.

 

[Newtonius]

By the way i just discovered, that whenever i run miner, my memory clock drops by 200 immediately. (whether its underclocked by 500 or overclocked by 800).
So if my default memory clock is 7000 and i underclock it by 500mhz, it will be 7000-500-200, and if i overclock it by 800mhz it will be 7000+800-200. And hashrate is stable on both 6300mHz and 7600mHz.
What can be the issue?

Adding to what @Phaaze88 said, this is normal given mining or high intensity VRAM usage, as long as your Hashrate isn't drastically fluctuating or dropping below 27 MH/s then you should have no need to worry.

 

[Rajahspice]

Memory thermal throttling. What you see with the gpu core temp, the Vram is definitely going to be hotter than that while mining.
You can't see the temp of the Vram with the tools you currently have. Only way to see it now is with a thermocouple or something similar.

i get you., but i think i need to add some details.
-i was mining with those 6300mhz and 7600mhz for 3 days almost 24/7 and was getting stable hashrate without any rejected shares (27.4mh/s with 6300mhz mem clock and 30.05mh/s with 7600mhz mem clock ).
-with memory set to constant +800mHz when i started heaven benchmark instead od miner, (it didnt decrease my memory clock by 200mhz) there was some stutters and driver crashed in about 2 mins, however when i set it to +600mHz benchmark ran without driver crash (still with stutters, but i have underclocked everything else and it should be reasonable)
- I cant get the fact, that when i just launch my pc (temps should be ok), miner still decreases memory clocks (1 second after launching) for no reason, and if it decreases memory clock from 6500mhz to 6300mhz why it doesnt decrease memory clock from 7600mhz to 6300mhz?

P.S. I am sick now and maybe my questions are kind of dumb, but i will be grateful for your explanations

 

[Phaaze88]

It was a guess based on what you posted, but without a legit way to measure the thermals, who can say?

Heaven Benchmark isn't going to press the Vram hard at all; it's old, and the max APU it can run is DX11. It's not going to be an accurate representation of a mining routine in the slightest.

 

[Rajahspice]

Adding to what @Phaaze88 said, this is normal given mining or high intensity VRAM usage, as long as your Hashrate isn't drastically fluctuating or dropping below 27 MH/s then you should have no need to worry.

Yeah, as i said hashrate is stable on both memory clocks, 27.4 mh/s for 6500-200mhz and 30.5 mh/s for 7800-200mhz with an error of 0.01 maybe.
.

 

[Rajahspice]

Update.
With the latest version of GPU-Z i can finally see memory temperature, and as i expected it is was almost the same with hotspot temperature.
But the bad thing is that one of my fans just stopped spinning and avg temperature went to 72 C from 62 C. (Probably gonna do RMA so they can repair it)