H100i Platinum - High temp delta between CPU and coolant

[tamag901]

Is a +55-60C difference between the CPU package temp and coolant temp normal?

 

Got this when running the CPU-Z stress test. When starting the test, the CPU instantly shoots up to 80-85C, with coolant in the 25C range. The coolant temp does rise by about 0.1C every 2-3 seconds, so there is thermal transfer. Once the coolant hits 32C, the CPU is at 100C and thermal throttling.

 

My chip is a i7-10700K overclocked to 5.1GHz @ 1.350v.

 

I'm using XTM50 thermal paste that I applied using the included applicator.

 

I've attached a screenshot of the temps from iCUE/HWMonitor.

Edited February 21, 2021 by tamag901

[c-attack]

That coolant to CPU delta a bit high for 1.35v. Before you go pulling anything apart, go the motherboard section of HWiNFO and check the Vcore reading there. It is not likely the actual Vcore matches the VID and if you are using auto voltage is may be pumping you up a bit more. I suspect the actual may be in the low 1.40s. You only need to run the CPU-Z bench for a few seconds to get the data you need here. All the subsequent rise is coolant temp bases (+1C H100i Temp = +1C CPU temp). That can be handled with fans, but the coolant to initial CPU temp is mostly about voltage and conductivity.

 

Also, be aware there is a bug with the current version of CPUID SDK that CUE uses for data and it makes Comet Lake CPUs read about a tenth under their actual Vcore settings. Don't use CUE voltage right now.

[Infin1tum]

whats your loadline? Telling your BIOS Voltage without LLC doesn't tell that much. Edited February 21, 2021 by Infin1tum

[tamag901]

Hi,

 

My LLC is currently set to Automatic - I was getting random crashes when trying out anything from Mode 3-8. Setting this back to Auto fixed the crashing so I believe this is either Mode 1 or 2.

 

My voltage is currently set to Adaptive + Offset mode in BIOS, with the offset at +0.025.

 

Here's a screenshot of the motherboard section in HWInfo while running CPU-Z stress - it says the Vcore is 1.356v.

[tamag901]

Small update - I did some tweaking around with voltages in BIOS and have set it to manual. I'm now doing 1.31v and Mode 4 LLC @5.1GHz stably.

 

With this setup HWInfo reports 1.29 vCore under load but the CPU still slams into the 80+C mark a few seconds after starting the test.

[c-attack]

So it looks like +50C coolant to CPU differential at 1.30v. That is more in line with typical overclock yields (vs the +60C) and where most people are forced to stop on voltage. At this point, there really are only two possibilities. One is the cold plate and CPU don't have perfect contact. Some washer missing in one corner, a post that didn't quite go all the way through, or some other fiddly aspect of mounting a CPU. It's possible it might show up on the TIM with a corner not fully spread like the others, but this is going to be subtle. You are over by a small number of degrees, not the typical +30-40C when you totally miss on the mount.

 

The horrible other possibility is this is the normal differential for that specific piece of silicon. Not all 10700Ks will have the same differential at the same exact load/voltage.

[tamag901]

I'll try re-mounting my cooler tonight and see if there's any improvement. I may have gone a bit heavy on my application of XTM50 when I used the applicator.

 

Apart from a better mount, is there any other way to reduce the differential? Better thermal paste?

 

160W doesn't seem like a lot for a H100i Platinum to dissipate and it feels like a waste if its the CPU at fault here.

[LeDoyen]

a better thermal paste can get you a few °C.

Coming from a really crappy one (but not dry) maybe 2 or 3°C typically.

I don't know the XTM50 but if it's a thick thermal paste and you had applied a lot, then using a more fluid one like kryonaut will improve further, because the excess will squish out and you'll end up with just the minimal thickness.

 

The differential will always be big with CPUs unfortunately. you'll have better results fine tuning the overclock than trying to soak excess heat. CPUs create very focused heat compared to GPUs, they are always harder to cool even on custom loop.

That's why dialing the overclock is so important. it yields better results than beefing up the cooling system.

[tamag901]

I think I've got my OC settings right on the edge of stable - 1.30v and Mode 4 LLC. Any lower and OCCT fails.

 

 

I just remounted and dropped my temps by quite a bit actually - I'm now hovering in the 70C range when running the CPU-Z stress instead of slamming straight into 80s.

 

I grabbed a shot of the IHS right after I removed the pump head, it looks like the thermal paste hadn't spread evenly?

[c-attack]

Hmmm... I am undecided if that is genuine thin spot or if it just came out that way when you removed the block. Either way, good to hear temps are more reasonable. I doubt you will be able to get under 1.30v for that frequency. I know I can't either.

 

It's really hard to make up temperature on the CPU block. In terms of thermal conductivity, it's very thin and not something you are going to see between a copper block of 95% vs 80%, etc. Most differences in CPU blocks related to cooling fin size/density and there is a trade-off in terms of liquid slow down in the block. More cooling fins means more resistance which in turns slows down the fluid passing through. That causes the fluid to heat up more while in the block, which is passes back to the CPU. This is not something you see much in AIO coolers because they are not going to make a dense block that needs a 4000 rpm pump to push fluid through.

 

The only really meaningful gain to be made here is delidding the CPU. My 10900K is delidded. I don't have a great sample and I shudder to think what it would be like if I had not done so. I would not be running 5.2x10@1.40v.