Watercooled GPU for mining - max memory overclock

[Margott247]

Hello, I am running at these settings, and i get around 41-44mh on phoenixminer/ethermine - it is around 41-44 when im on +1500mhz memory too, and core has no effect on the hashrate.

my gpu is an RTX 2080S Waterforce by gigabyte, that cools the memory too.

 

 

My question is, is it safe to run my gpu like this, with the memory overclocked so high, or will my gpu die sooner?
We have talked about this with my mates, but i realized i dont really know whether its the heat or the high clocks that damage the memory over time.

Thanks.

[tikker]

Yeah core won't affect ETH much. In the end heat is the biggest threat. The unsafe aspect of high memory clocks is the lack of temperature sensors on the memory chips, so it's impossible to judge safe operating ranges. It might be running extremely hot which could damage it over time.

 

Also when I was reading up on my 1080 Ti I saw it mentioned that GDDR5 and newer have error correction built in which could lead to higher apparent hash rates, but lower effective rates as it has to correct the errors it made due to higher clocks and take longer to finish a share. Can't find it anymore though, so take this with a grain of salt for now. Might be visible if your pool reports your effective hashrate alongside your reported hashrate.

[Mick Naughty]

Don’t see any setting killing something. 
 

Pointless to use settings that don’t directly effect the hash rate.  
So if you’re testing shows it working, then keep doing it. 

[AngryBeaver]

The 3080 does have a memory temp sensor due to it being common on gddr6x.

 

Now that said my 3080 can run at +1500 memory without errors when on water... my temps still hit 88c-90c. If I run my memory at +1000 then those temps are 80-84c. Which makes me a little nervous, but apparently anything under 95c is pretty safe. Now cards that are on air constantly throttle due to 110c or higher temps. 

 

Now I just need a water cooled backplate to help memory temps further.

 

Oh and since we are talking settings and hash rates. My 3080 runs at 68% plimit, -250 core, and since the difference isn't worth thr risk +1000 on memory for about 105 average MHs. At 1500 it will land around 110, but I would prefer seeing memory at 80-82c

[MGKompressoR]

Hello guys, I have a palit 2080s gaming pro non OC model, and at least with the GPU-Z it shows a Memory Temperature, that is normally so close to another measurement called Hot Spot, I wonder how much can trust in those. 

http://www.palit.com/palit/vgapro.php?id=3470&lang=en

Will resume my experience. All with custom fan curve - stable temps

Stock settings                    Core: 66 °C  Hot Spot: +75.5 °C     Mem: + 74.8°C    36MH/s

Only mem +800                 Core: 66 °C  Hot Spot: +76.5 °C     Mem: + 75.7°C    40.4 MH/s

My custom                         Core: 60 °C  Hot Spot: +72.2 °C     Mem: + 71,4°C    42,5MH/s

My custom with mem +0   Core: 60 °C  Hot Spot: +71.9 °C     Mem: + 71,0°C     36 MH/s

 

My custom is -50% PL. -200 Core clock.  +1200 Mem Clock.

As you can see, when the PL it's on 50% the memory speed barely affects the temperature, i can even run it at +1500 and doesn’t change more than .2 °C.

I’m open too your opinion about, how safe it is to run at which speed, and how trusty is that memory temp number. I really don't want my GPU death for mining 2 months.. 

 

Sorry for my english it's cearly not my language. 

 

 

[Sarra]

On 1/31/2021 at 2:10 PM, tikker said:

Yeah core won't affect ETH much. In the end heat is the biggest threat. The unsafe aspect of high memory clocks is the lack of temperature sensors on the memory chips, so it's impossible to judge safe operating ranges. It might be running extremely hot which could damage it over time.

 

Also when I was reading up on my 1080 Ti I saw it mentioned that GDDR5 and newer have error correction built in which could lead to higher apparent hash rates, but lower effective rates as it has to correct the errors it made due to higher clocks and take longer to finish a share. Can't find it anymore though, so take this with a grain of salt for now. Might be visible if your pool reports your effective hashrate alongside your reported hashrate.

Now I'm actually glad I ended up with a Vega 64 instead of a 1080TI. My Vega has HBM temp sensors, which is why my card is still functional right now (I had a HBM overheat, so I backed my memory clocks back).