Several important questions about overclocking regarding clocks, temperatures, etc.

[comanzo]

Hello guys. I have several important questions about overclocking gpu. I am currently doing so on gpu.
1. If you stress the gpu too much by adding either too much core clock, or memory clock, can that damage the gpu? If not, then why do people suggest going in small increments when raising clocks? Shouldn't it be from the top, down. Instead of from the bottom and raising incrementally?
2. I know that raising the clock of core/memory does raise heat mildly, but how much does overvolting raise it? I want to know this when I overvolt my 1070 ftw.
3.I heard if you raise the memory clock too much, you can fry the memory. Is this true? I recently got a checkered screen and had to do a hard reboot when raising the memory too much. I hope the answer to this is no, or else I am screwed.

Thanks for anyone who takes the time to help me with these questions.

 

[Tradesman1]

I'm one who favors taking things up slowly, you need to achieve the best balance between clock, memory and voltage OCs. As you go up check performance and compare to previous, you can get a setting in say clock or memory and raise the other and it may actually decrease performance, i.e. in round numbers say clock OC is at 100 and memory at 60 and things look great then you raise memory to 80 and performance drops a bit - that indicates you may need more clock to continue for true performance increase so raise the clock till it equals/exceeds what you had at 100/60 may take 125 to do better when memory at 80. It's all patience and experimenting.

 

[nikoli707]

1. on the stock gpu bios... no, you will very likely not damage anything. there are so many restrictions in place that the core will never reach dangerous voltage levels or temperatures to damage it. your welcome to try top down but bottom up is easier.

2. voltage will raise temps much faster than clocks.

3. again no. the memory will just error out. the memory can tollerate much higher temps than the rest of the card so damaging it is unlikely. since you cannot raise the dram voltage on anything except one or two top tier overclocking cards like the kingpin, damage is unlikely. in the real world memory overclock isn't going to amount for much performance gains. save the extra heat and stress for your core or lowering overall temps and fan noise.

 

[comanzo]

1. on the stock gpu bios... no, you will very likely not damage anything. there are so many restrictions in place that the core will never reach dangerous voltage levels or temperatures to damage it. your welcome to try top down but bottom up is easier.

2. voltage will raise temps much faster than clocks.

3. again no. the memory will just error out. the memory can tollerate much higher temps than the rest of the card so damaging it is unlikely. since you cannot raise the dram voltage on anything except one or two top tier overclocking cards like the kingpin, damage is unlikely. in the real world memory overclock isn't going to amount for much performance gains. save the extra heat and stress for your core or lowering overall temps and fan noise.

For answer #3, the answer surprised me. Several youtubers say the vrm's (the stuff responsible for memory) are very sensitive to temperature. This is probably why people say memory clocks have the potential to fry the memory. I guess they meant when you raise the voltage the vrm's are the most sensitive, not when you raise memory clocks. As memory clock hardly puts a dent for temperature. Can you verify if vrm's are the most sensitive when it comes to temps./voltages? Thanks for the quick response.

 

[nikoli707]

vrm is voltage regulating module and controls the vcore on the core. i guess it pertains to the memory as well but the core is vastly higher amperage and current than the memory and high vcore will be the cause of vrm failure long before gddr5 failure. again i wouldn't worry about memory clocks that much as they are already plenty high enough to keep the core saturated.

 

[comanzo]

I'm one who favors taking things up slowly, you need to achieve the best balance between clock, memory and voltage OCs. As you go up check performance and compare to previous, you can get a setting in say clock or memory and raise the other and it may actually decrease performance, i.e. in round numbers say clock OC is at 100 and memory at 60 and things look great then you raise memory to 80 and performance drops a bit - that indicates you may need more clock to continue for true performance increase so raise the clock till it equals/exceeds what you had at 100/60 may take 125 to do better when memory at 80. It's all patience and experimenting.

Ah. I see. So the reason why people take the clocks slow isn't due to safety reasons, but rather to find that balance more easily/quickly. Otherwise, if you did it from top to bottom, instead of bottom to top, you could miss that balance for best performance. Is this what you are saying?

Secondly, My temps with clocks adjusted, and temp./power limit to the max, is 70's mostly, with it never reaching 72. Do you think I should overvolt, or should I get better cooling first? Nvidia severely limited voltage to 1.093 volts with pascal, which isn't far from 1.050 volts. That is why I think I can do it without reaching dangerous temperatures. Can you confirm on this, as this is my first time overvolting as well.

 

[nikoli707]

i can tell you on the stock bios with the 1070 the card is going to boost itself up to the 2100mhz +/- range and automatically downclock when it reaches is limits. so toying around within a 25-50mhz range is going to yield you the same boosted max. i could be wrong but i think you will start thermally throttling before the core is starved of voltage... and again long before you are into dangerous vcore levels since you are capped with the stock bios. you could ramp up fan speed and get unnecessarily loud noise but high cooling performance to overcome this but its not worth it for a few extra mhz.

 

[comanzo]

i can tell you on the stock bios with the 1070 the card is going to boost itself up to the 2100mhz +/- range and automatically downclock when it reaches is limits. so toying around within a 25-50mhz range is going to yield you the same boosted max. i could be wrong but i think you will start thermally throttling before the core is starved of voltage... and again long before you are into dangerous vcore levels since you are capped with the stock bios. you could ramp up fan speed and get unnecessarily loud noise but high cooling performance to overcome this but its not worth it for a few extra mhz.

Ah. Ok. Yea the 70 degrees was at 100% fan speed. The case I am using is an ASUS M51AD. That's why my temps are quite high for 100% fan speed. My last question to you is, do you think I should buy 2 more fans for this OEM case, or buy an aftermarket case altogether for lowering temps.? You see, my case is a mid-tower, and has two more locations for fans. Or, the last option is that, since you said I am only missing out on a few mhz, that it may not even be worth it to get better cooling to then overvolt. Since you said that overvolting is only going to get you a few mhz, it may not even be worth it to invest in better cooling. Mind you, I have no budget, and if the temps. in any way is going to cause more wear and tear, then that will easily push me to get better cooling. So how would you go about this if you had this case with these temps.?

 

[Tradesman1]

Safety can play in also, I've seen people and threads here on the forums all too often where a person sees a video of someone OCing, takes it for gospel so they just set their GPU to very high settings and fire it up - and on occasion fry something, - often due to weak or cheap PSU, though other things can affect it also

 

[comanzo]

Safety can play in also, I've seen people and threads here on the forums all too often where a person sees a video of someone OCing, takes it for gospel so they just set their GPU to very high settings and fire it up - and on occasion fry something, - often due to weak or cheap PSU, though other things can affect it also

Yes there is a safety concern tooly a hard reboot. Are you talking about overvolting? If so, that makes a lot of sense. If talking about adjusting clocks, then there isn't much of a safety concern I believe.

On another point, what's your perspective on if I should buy more fans, or just buy a new case, or stick with what I got? Do you think I am not missing on much performance not being able to overvolt? I can't overvolt since I am being thermally throttled in the low 70's. If the answer is yes I am not missing on much, then I might not invest in better cooling. Otherwise, I may either buy more fans, or a new case altogether? Please note that with the case I have (ASUS M51AD), I am able to add 2 more fans.What's your opinion on this? Thanks for the help so far.

 

[nikoli707]

more case fans is almost never a bad thing. just get a decent oc out of your card. you dont need to max it out. in games there is only going to be a few fps difference between a moderate oc and a max oc.

 

[comanzo]

more case fans is almost never a bad thing. just get a decent oc out of your card. you dont need to max it out. in games there is only going to be a few fps difference between a moderate oc and a max oc.

Ah. I see. So you are in favor of getting more fans instead of a new aftermarket case to replace the OEM case. I was also leaning towards that option as well. It's also nice to know that I am not missing out on much in my oc (save for a few fps). One last question (and this time, I promise it is the last one). For future reference, I know that the danger levels of temperature is about 85 degrees. Anything lower and it's fine, even for extended periods of time. What I would like to know, is what should be the temperature target before overvolting? In other words, what is the highest temperature that is still safe to overvolt all the way? Would it be 60 degrees, 65 degrees, etc.? This question is only for pascal cards, as other generations behave differently to overclocks. Should it be forbidden to all air gpu's and only exclusively to water gpu's due to the sensitivity of vrm's? I do apologize for all these questions. Thanks.

 

[nikoli707]

well a new aftermarket case would be ideal overall. but you can get cheap case fans, sometimes for only a few dollars, that will definitely help. transplanting an oem motherboard into an aftermarket case might not be as easy as it sounds as sometimes these oem components are proprietary and mounting points might not line up. i would do research before going with new case.

 

[comanzo]

well a new aftermarket case would be ideal overall. but you can get cheap case fans, sometimes for only a few dollars, that will definitely help. transplanting an oem motherboard into an aftermarket case might not be as easy as it sounds as sometimes these oem components are proprietary and mounting points might not line up. i would do research before going with new case.

Yea. It might just be easier to buy fans, and once I need a cpu upgrade, I upgrade the motherboard as well, along with the case. I can then add the fans from the old case into the new case. I may just simply do that. What are your thoughts though on that last question about temperature target when overvolting? Like I said, I won't be overvolting on this gpu since you said it's not much of a loss. But I would like to know for future gpu's, what the temperature target should be? In other words, the highest temperature that is still safe to overvolt.

 

[nikoli707]

each different gpu architecture has different tollerances as far as temps are concerned. after that each individual board partner like msi evga sapphire gigabyte etc are going to have different temperature targets that they have decided to implement. in reality these modern gpu cores should be able to handle 90c without issue, but is this far higher than ideal and i definitely wouldn't recommend running the card that hot except maybe for a quick suicide benchmark run. most cards have a temp priority of 78-83c though and for the most part i would want to be below 80c. if you overclock and you are within 2-3c of the temp priority target while gaming then you will want to back off a little so you doing encourage throttling which will affect gaming performance and stuttering. 75c is plenty safe and far enough away from most temp targets that even if ambient temps rise on rare days you wont inadvertently hit the temp target.

 

[comanzo]

each different gpu architecture has different tollerances as far as temps are concerned. after that each individual board partner like msi evga sapphire gigabyte etc are going to have different temperature targets that they have decided to implement. in reality these modern gpu cores should be able to handle 90c without issue, but is this far higher than ideal and i definitely wouldn't recommend running the card that hot except maybe for a quick suicide benchmark run. most cards have a temp priority of 78-83c though and for the most part i would want to be below 80c. if you overclock and you are within 2-3c of the temp priority target while gaming then you will want to back off a little so you doing encourage throttling which will affect gaming performance and stuttering. 75c is plenty safe and far enough away from most temp targets that even if ambient temps rise on rare days you wont inadvertently hit the temp target.

Ah. Ok. So there is no set temperature to target before overvolting after all. For example, you can't say, "make sure your gpu temp. is 60 degrees or less before overvolting to 100%." The reason being becuase of the generation of the gpu, as well as the manufacturer(Evga,MSI, etc.). Ok, I got it now. Well thanks for the help, I appreciate it.