i9-8950HK intantaneous core temperature spikes

Apparently it also happens on the MSI GT75 Titan in the same way that it happens on the R5, and one of the reasons I returned mine. I have recently been on the MSI forums and read a review of the GT75 Titan with i9-8950HK and GTX 1080. He mentions instant 95C spikes, and I cannot assume that his laptop was also dodgy. I therefore conclude that the temperature spikes are a feature of the i9-8950HK. We need more users to come up with more real life reviews and stress tests to be able to tell for certain. It seems to me that Intel has pushed the envelope a bit too far with their Speedstep and Short Turbo implementation on the i9-8950HK.

 

it's something that has been present many generations already. it's just burst for a micro second. allmost all K processors have that, and it's just a lag of the fan not ramping up it's speed fast enough. it won't thermal throttle or power throttle, because that's just how Intel CPU's work.

https://communities.intel.com/thread/119878

https://communities.intel.com/thread/121022

Intel investigated this matter already. you can read their answer here:
https://communities.intel.com/message/471425#471425

 

I do not think that the fan/heatsink could ever catch this, the heat does not even have time to spread to the other cores, let alone reach the CPU's case and the heatsink. Because of that behaviour Intel should have programmed the profile not to trip whatever it does trip and then cause massive UC. Maybe it might be better to disable this "short turbo" and find a constant speed that the CPU can run at within acceptable temps. Especially for desktop replacements we do not care about battery but stable performance.

 

Like I said, it's not something that causes concerns. It's just how K processors burst to their boost speeds and back to idle again faster.

If you wanted the temperature to be less, then you should've bought a laptop with a 8700K. It has the same "issue" but because it's a more "pure" piece of water, the temperature and voltage it needs to get to it's boost speeds are much and much less. On the desktops it burst for example to 60/70C and back to idle fornexample 30/35C. But because you have a binned 8700k that has become a HK, you have the lesser quality wafer. So temperatures will be naturally higher.

As long as your cpu runs within it's Tjunction, it is within acceptable temps. The Tjunction of the 8950K is 100C so you are within acceptable temps (according to Intel).

You can disable it, by disabling speed shift (in TS). But do not forget, windows 10 now manages speedshift, so it could have no effect.

You can read more about it here:
https://www.anandtech.com/show/9751/examining-intel-skylake-speed-shift-more-responsive-processors

 

As @Rei Fukai said, I'd go a step further as Kevin said when I asked the usage of C1E in TS. He said TS was meant to work on older CPUs from Intel and also on modern Intel chips too, he also suggested me to disable C1E to decrease micro-stutters. In fact it did work.
I'd recommend you use TS and uncheck C1E and Speedstep or EIST and check off HWP_EPP aka SpeedShift v1(sky)/v2.0(kaby and above).

 

I've seen "spikes" on the 7820HK, but never spikes that excessive. I've seen spikes where the CPU is idle at 35C then jumps to 55C, 60C or even 65C for half a second then drops back down. These spikes are always because something signals a load on the CPU.

However, at heavy load, let's say at 80C for example with most cores evenly loaded, unless another application suddenly decided to steal CPU cycles, I NEVER see a spike higher than 2 or 3C from anything with a balanced load, ever. So no 80C to 100C jumps to worry about.

One thing I *DID* notice and this could be a bug in the Bios or it could be real, is that basic windows and even some applications like Chrome use AVX instructions. AVX instructions cause extra load on the CPU, and when an AVX instruction is executed, CPU voltage is increased by 30mv. (That's why when you run an AVX version of prime95, you notice the CPU voltage 30mv higher than if AVX is disabled in local.txt or an old version of prime95. This is brainless and and simple to see with desktops that have true "CPU VCORE" sensor monitoring, but with laptops, at least on Alienware and MSI books, this 30ms "spike" is impossible to see because of how the default IA AC DC Loadline is tied into the CPU VID response. The only way to see this on a laptop is to set the IA AC DC loadline values to "1" (0.01 Mohms), and then you will see it. And this 30mv AVX voltage boost was mentioned a long time ago over on overclock.net, not here. In that case, if IA AC DC Loadline is set to 1, you will always see the "CPU VID" boosted by about 30mv if you attempt to run AVX prime95, and it will remain there and not fluctuate.

I noticed that when I set an AVX offset in the MSI Bios, the CPU would randomly downclock by that offsets during basic windows idle operations and in some games. It is beyond my patience to test whether this is a bug or not, or whether something is pulling an AVX operation in the game or the basic windows desktop.