[SOLVED] i9-10900k core temp difference

[ccqueen]

Hello all. Just completed a build running a 10900k (turbos up to 5.1GHz) and Noctua NH-D15. Everything went well during the install; put the right amount of thermal paste using Noctua's included NT-H1 and eyeballed it the best I could when it came to lowering the heatsink onto the CPU. Once I started to catch the screws to secure it I noticed it slid around a bit as it was trying to catch but did not actually come up 😬 I then tightened the screws evenly and all the way. Fast forward to now and the temps idle around 28c-31c evenly for the most part. The only discrepancy (or maybe it's just me) is when I start to play games and I go back to hwmonitor, core 4 & 5 seem to have the highest recorded max temp out of the rest of the cores (average 10c difference). It's not a constant thing as when running prime95 the temperatures stay fairly close to each other but get an occasional temperature spike on a random core. The attached pic is playing BFV for around 45 minutes. Another thing to mention is I adjusted the cpu fan curves in MSI's bios as the stock ones were geared more towards keeping things quiet and not cool. My biggest concern when building a pc is installing the heatsink and applying the right amount of grease to get the best temperatures possible given the hardware and was just curious as to if this is normal and okay or maybe I would benefit from reseating it? Thanks for the help.

 

[Karadjgne]

It's pretty normal to have upto a 10° difference between hottest and coolest cores. Some cores in the center are landlocked, have multiple cores surrounding them, adding to heat soak, and some cores (especially the corners) are almost heatsoak free with 2 free edges.

Add in the differences with silicon and voltages necessary per core and 10°C difference isn't a stretch. The advantages most get from delidding is better heat transfer, which negates much of the heatsoak, but even then there's still sizable differences between most cores.

 

[ccqueen]

It's pretty normal to have upto a 10° difference between hottest and coolest cores. Some cores in the center are landlocked, have multiple cores surrounding them, adding to heat soak, and some cores (especially the corners) are almost heatsoak free with 2 free edges.

Add in the differences with silicon and voltages necessary per core and 10°C difference isn't a stretch. The advantages most get from delidding is better heat transfer, which negates much of the heatsoak, but even then there's still sizable differences between most cores.

Just ran Cinebench R23 on Multi Core and then Single Core (first pic multi, second single). I noticed on Single core, 4 and 5 were constantly hotter than the rest and has about a 20c difference @ max values. I know 10c is fairly common but 20c.. Maybe I should just reseat the darn thing and see how "good" of a job I did on the previous TIM. A bit nerve racking to say the least.

1.

2.

 

[Karadjgne]

Probably couldn't hurt. At the very least you'll get to see what kind of spread you got out of the paste and can adjust. Being new, won't hurt to reuse it, if you have enough, but using a credit card just get a good thin even coat on the cpu.

According to HWInfo above, the max per core is showing just a 9° spread, which is about right, 81°-90°.

If it's worrysome, use Prime95 Small fft torture test (not smallest) and disable any AVX technologies. That's a straight 100% load, no deviation unlike many others. That'll show you exactly what each core is doing, with the same load.

Your cpu does have favorite cores, those are cores the cpu prefers to use first, so often times you will see certain cores taking the brunt of any workloads, which can adversely affect temps. Is why Prime95 small is as good as it gets, same load all cores.

 

[ccqueen]

Thanks for the reply! Ran Prime95 small with AVX disabled for around 5 minutes and was around a 7 degree difference. With 80 being the lowest and 87 being the highest. I probably won't mess with the heatsink unless I get bored and feel like tinkering with things 🤓

 

[Karadjgne]

With a 7° difference total, you are pretty close to having a golden chip. 5° median is about norm, so with 3½ median, you are well above the curve. I'd not mess with the heatsink at all, at this point it'd be a waste. You are going to see differences in cores, they are not identical, each is unique, plus the TIM you applied is a second layer, there's also one under the IHS, whether it's paste, metal or solder, and add in the concave nature of the IHS to begin with.

So you aren't going to get perfect exact temps across cores, 7°C is better than normal for a non-delidded cpu.

What I normally do for temp readings is use CoreTemp, it has a choosable polling core and an output that sits next to my clock on the Taskbar. You can choose whichever core you want as the output core I choose a median temp, not hottest, not coldest. Kinda averaged. But thats me. If 4 is generally your hottest, you could use that, and know all the other cores are Less. It'll give an idea of exactly how the apps are cooking the cpu and CoreTemp is as accurate for Intel or Ryzen as it gets.