Intel and the solder / thermal paste thing

[swaaye]

I did the delidding thing the other day with a 7600K and saw a ~20 °C drop in temperature while overclocking with the same cheap, 5 year old Cooler Master Hyper 212 EVO cooler. It was also fun to try Coollaboratory Liquid Ultra. Previously the 7600K was hitting 100 °C pretty easily at 4.8Ghz ~1.28v and throttling. It now stays under 85 °C during hours of logged Prime95 torture even with a slower, less annoying fan!

The thermal paste application under the IHS was interesting. There was a small gap between the IHS and the CPU die causing the paste layer to be fairly thick. Thermal paste is not a great conductor of heat, especially if there is any significant gap involved. I imagine that's why it performs so badly here.

I came across an interesting article about the chemistry and physics involved with soldering a nickel plated piece of metal to a silicon die. Maybe gives part of the picture as to why Intel stopped doing it.
http://overclocking.guide/the-truth-about-cpu-soldering/

 

[CSI PC]

Yeah,
isn't part of the issue the gap between IHS and CPU in this case with Kaby Lake?
Even without the gap, I assume Intel is looking at an envelope up to a certain threshold, maybe logically 1.2 to 1.24V *shrug* before hitting temp limits with their current die-TIM-IHS packaging.

Can you remember what your performance/temp was like before delidding and around 1.2V to 1.24V?
Fingers crossed Intel sorts the gap out on Skylake-X and their general TIM related packaging-design.
Cheers

 

[swaaye]

Yeah,
isn't part of the issue the gap between IHS and CPU in this case with Kaby Lake?
Even without the gap, I assume Intel is looking at an envelope up to a certain threshold, maybe logically 1.2 to 1.24V *shrug* before hitting temp limits with their current die-TIM-IHS packaging.

Can you remember what your performance/temp was like before delidding and around 1.2V to 1.24V?
Fingers crossed Intel sorts the gap out on Skylake-X and their general TIM related packaging-design.
Cheers

I think the gap has been an issue since Ivy Bridge switched to paste.

The 7600K here will do 4.7 GHz with +120mV of offset voltage and that was on the verge of throttling in Prime95 (100 °C ). I think that is slightly beyond 1.24v. Right now with the liquid metal in there it is at 4.8 GHz +160mV and the maximum temperature I saw in my 3 hour Prime run was 84 °C. If the die's power readout is accurate it's still using a few Watts less power than my only slightly overvolted 4.2 GHz 2500K machine during Prime and there's no AVX2 there.

 

[CSI PC]

I think the gap has been an issue since Ivy Bridge switched to paste.

The 7600K here will do 4.7 GHz with +120mV of offset voltage and that was on the verge of throttling in Prime95 (100 °C ). I think that is slightly beyond 1.24v. Right now with the liquid metal in there it is at 4.8 GHz +160mV and the maximum temperature I saw in my 3 hour Prime run was 84 °C. If the die's power readout is accurate it's still using a few Watts less power than my only slightly overvolted 4.2 GHz 2500K machine during Prime and there's no AVX2 there.

Thanks and yeah well remembered goes back that far and unfortunately rather hit/miss with Intel as some of their CPUs have been slightly better than others in terms of the die-TIM-IHS packaging/design but still not perfect regarding the gap.

So does seem plausible Intel is looking at an envelope with an acceptable ceiling around 1.2V (should be good enough for around 4.5GHz all cores on the 8C Skylake-X).
Quite interesting on another forum it was shown how Xeon Phi is also using paste and not solder, but it all comes down to the envelope and range.

Cheers

 

[Malo]

It was my understanding that the larger dies don't really suffer the issues with solder and that's why folks are saying the newer big Intel's should be soldered. I'm guessing Intel cares a lot more for longevity and warranties than overclock in.

 

[CSI PC]

Great analysis video by Der8auer regarding the Skylake-X TIM vs Delidding.

Looks like Intel has done a pretty good job.
The 10C 7900X with an AIO (Corsair 280) and with its default TIM hits 4.8GHz overclocked all cores; seems that was at 1.25V and so fits in with what was surmised earlier that potentially Intel was looking at an acceptable ceiling of 1.2V (on air) to 1.24V.
Note that is not delidded.

With delidding and using liquid metal he only managed 5GHz with the same AIO - hits 1.32V so we can see where the wall starts to be hit and very similar to Broadwell-E characteristic.

So looks like Intel did a pretty good job with the die-TIM-IHS packaging with Skylake-X.
Some may be critical of the temps (but that is with Prime95), but then Intel was looking at acceptable envelopes and limits for most users.
Cheers

 

[Clukos]

Looks like Intel has done a pretty good job.

84-94C across all cores on a non-avx load at 1.25vcore is far from a "good job". These should have been soldered from the get-go.

 

[CSI PC]

84-94C across all cores on a non-avx load at 1.25vcore is far from a "good job". These should have been soldered from the get-go.

It is still Prime95 that is one of the most stressful tests you can do.
A good job considering it is die-TIM-IHS and all cores 10C/20T and he is showing 4.8GHz with it.
Even Der8auer is pretty satisfied that the solution works fine for its purpose and he is a top league international overclocker.
Look back to how it was with 4C/8T products using die-TIM-IHS; in this thread swaaye said regarding 7600K:

The 7600K here will do 4.7 GHz with +120mV of offset voltage and that was on the verge of throttling in Prime95 (100 °C ). I think that is slightly beyond 1.24v

And remember my points that I have been saying for a little while is Intel has designed this probably for around ideal 1.2V ceiling for most users.
To put it into perspective here is the power draw for Broadwell-E with standard boost/4GHz at 1.2V and 4.3GHz at ~1.35V

That 4.3GHz OC (~1.35V) is done with a hefty water cooled rad solution and pretty much matches limit most other sites that overclocked on water the Broadwell-E 10C.

So my POV they have done a pretty good job, especially considering trying to set Ryzen to 4.1GHz that is usually unstable due to temp even on water and is soldered or looking at 4C/8T consumer SKUs with TIM., even with the paste solution you can get stable 4.8GHz with a moderate AIO with 7820X - Point being this is fine if one is not pushing to the extreme limits.

Edit:
To be clear I would also prefer soldered, but this is better than I expected.
Cheers