What is a "good" delta T for a GPU + CPU loop?

[sveniat]

I've been having some trouble finding how to rate a loops thermal performance. I've read that 5C is amazing, 10C is good, and 15C-20C is sub-par (all of those numbers are deltas above ambient), but it didn't specify if that was CPU only or CPU+GPU. I've also read that for "a GPU loop", that 15-20C used to be considered good, but it doesn't specify what changed and how it affects rating a system. I guess that increase power efficiency means that things dont run as hot = you shouldnt see as high of temps? but if that's the case, how much did the "good" standard drop by? And that still doesn't specify if "gpu loop" means a full CPU+GPU loop, since I know some exotic custom loops have a seperate loop for each.

 

Also, I see a lot of "stay below x If you want to do a lot of overclocking". But if you already have an overclocked system, how do you evaluate your delta without intentionally underclocking everything just to bench (not to mention GPU boost makes underclocking a guessing game anyway). 

 

I realize that practically speaking, as long as you have good component temps you should be happy, and I do have decent temps, but i'm also a fan of getting the most out of my system just because I can and want to know how my loop compares to a global custom water loop standard. 

 

Right now I've got a i7-7700k overclocked at around 4.8ghz (it's sadly not a great overclocker), and a single 1080ti running at about 2063mhz which doesn't get higher than ~52C at sustained full load. I've got a 420mm rad and a 360mm rad, and i'm seeing peak water temperature deltas of about 15-16C. Is that good relative to my radiator space given things are already overclocked? should I be looking to "fix" something if they are not as low as expected?

 

I want to re-iterate that I'm not trying to overkill my loop for the sake of performance, because I already have "adequate". I'm doing this because I simply want to make an objectively great loop, and learn about making good loops in the process. 

[Cyracus]

Watercooling has evolved from people slapping things together to actual components being designed and produced for the purpose of watercoooling a pc, so that's one reason things have changed. But your loop sounds great, nothing to fix from what I can see, especially if you're running your fans slow to keep things quiet

[sveniat]

Just now, Cyracus said:

Watercooling has evolved from people slapping things together to actual components being designed and produced for the purpose of watercoooling a pc, so that's one reason things have changed. But your loop sounds great, nothing to fix from what I can see, especially if you're running your fans slow to keep things quiet

good to hear that my loops sounds good, but just for my own knowledge what range would you say that is? In other words, what temp delta would start going "hmmm that seems a bit high"? 

[sveniat]

oh, also how much does this info change if you add SLI into the mix?

[RadiatingLight]

1 hour ago, sveniat said:

good to hear that my loops sounds good, but just for my own knowledge what range would you say that is? In other words, what temp delta would start going "hmmm that seems a bit high"? 

maybe 30-40c delta would be concerning, depending on how much radiator space you have and what fan speed you're running at.

 

1 hour ago, sveniat said:

oh, also how much does this info change if you add SLI into the mix?

another GPU really heats things up more. if you were to SLI, I'd expect temps across the board to go up a few degrees. an OC'ed 1080Ti is nothing to sneeze at.

[For Science!]

4 hours ago, sveniat said:

-snip-

I'm not sure if I understand what you mean by "peak water temperature deltas" whether this is referring to the change in water temperature under load, or the difference between temperature between the water temperature and the measured component temperature.

 

However,

- If you are confident your mounting of waterblocks is good (i.e. correct spreading of thermal paste)

- and there is adequate water flow in your system (can be quantified using a flow meter)

- and there is enough radiator space coupled to airflow that will not heat soak (not overcapacity of radiator)

 

The former would be influenced heavily by air flow (fan properties, radiator properties, case properties), whereas the latter would be influenced by the interface (how good the mount is, how rapidly the temp readings are updated, where the thermistor is).

 

Then the only consideration is keeping the water as cold as possible without annoying your ears, while maintaining safe temperatures for the components. Delta's above ambient is not really a good measure for anything in watercooling since it is the water temperature that ultimately dictates how effectively heat is moved away from the said components generating heat.

 

How close you can get the water temperature to ambient temperature under a sustained heat generated is determined solely by air flow and radiator area. If you cannot reach a sensible temperature (<65 degrees technically, but I prefer <50 degrees) then you need more airflow (stronger fans or colder air). But if the noise annoys you, you have to increase radiator space.

 

How close you can get your component temperatures to water temperatures under a sustained load, is kinda a stupid thing to worry about in my opinion since the component temps are affected by things like how rapidly the temp readings are made, where the thermistor probe is, whether there is an insulating layer of TIM between the IHS and CPU die, etc. If your waterblock is mounted correctly then one should just focus on component temps alone.

 

For example, despite having a water temperature of about 36 degrees or so (at near idle) once I start Prime95 (v26.6, non-AVX as I understand) then my core temps (7700K @ 4.9 GHz) will hit close to 80 degrees. Technically that is a 44 degrees above water temperature, and something like 55 degrees above ambient. But I am sure nobody in the audience would necessarily fault my loop for an overclocked KabyLake i7 hitting 80's using small FFTs in P95. 

 

During a slightly longer RealBench stress test, my GPUs (1080 in SLI) get to max of 47 degrees (~5 degrees or so hotter than the water temperature) whereas my CPU is in the mid-70s (~30 degrees or so hotter than water temperatures).

 

On an marginally related note, HWMonitor and HWinfo report temperatures 10 degrees hotter for the CPU compared to AI Suite (ASUS's own thing). I've been quoting the hotter temperatures to be on the safe side.