Improving liquid metal

[Pagusi]

Hey,

I was wondering if it would be possible to create a mixture of some metals that would melt at a little over 100°C so you could heat it up, apply it to your Cpu and then let it harden? So it wouldn't move inside. Because I heard about some guys using liquid metal for their laptops and worried that the metal would move.

 

 

[r2724r16]

 

[Ja50n]

It’s important for it to be liquid because it is able to fill much more of the air pockets then. Thermal pastes generally degrade over time because they’ve dried out, and so air is now in place of where material used to be. Not to mention, a material that’s only liquid over water’s boiling point would be very difficult to apply... Better to look for superior containment methods for liquid metal.

[Ja50n]

3 minutes ago, Rasbir Singh said:

 

 

[WoodenMarker]

If the material was solid, it would either break from stress due to changes in temperature expanding and contracting the material or no longer provide good contact between the die and ihs or cooler. 

[Mira Yurizaki]

4 minutes ago, WoodenMarker said:

If the material was solid, it would either break from stress due to changes in temperature expanding and contracting the material or no longer provide good contact between the die and ihs or cooler. 

This is also why Intel stopped using solder.

[Ja50n]

14 minutes ago, M.Yurizaki said:

This is also why Intel stopped using solder.

Do you have any evidence of this, or any links you could provide? As I understand it, the solder basically bonds the die to the heart spreader, and remains stable until heated to a temp like 350c, well out of the range a CPU is expected to operate at. The only reason I’ve seen for not using solder is cost, as it’s been shown to function within margin of error of liquid metal, and 20c cooler than Intel’s TIM.

[Mira Yurizaki]

14 minutes ago, Ja50n said:

Do you have any evidence of this, or any links you could provide? As I understand it, the solder basically bonds the die to the heart spreader, and remains stable until heated to a temp like 350c, well out of the range a CPU is expected to operate at. The only reason I’ve seen for not using solder is cost, as it’s been shown to function within margin of error of liquid metal, and 20c cooler than Intel’s TIM.

There's no conclusive evidence because whatever Intel does is only known to Intel. But there was a really good post discussing it at: https://medium.com/@OpenSeason/soldered-cpu-vs-cheap-paste-59fb96a4fca7

 

If you want a summary of it:

• Intel's CPUs have concentrated hot spots relative to the die size, meaning uneven heating.
• Uneven heating is likely to cause higher thermal cycle stress on solder, which will create cracks/voids. Those will degrade the heat transfer performance over time
• Any large increases of temperature after delidding, especially with recent processors, may simply be due to removing the excess distance caused by the silicone sealant around the IHS.
• Author used a CFD simulation to test if higher thermal conductivity constant really mattered. They found out it doesn't after a certain point.
• Author made mention of having low thermal resistance should be a factor as well in looking at TIM performance over time. The TIM Intel chose is supposedly low enough.

[Mira Yurizaki]

Also to try to poke holes in the "Intel is saving money by using TIM", I picked a random lead-free solder from Amazon: https://www.amazon.com/gp/product/B01B61TWGY/ Use the original price of ~$16 and with a mass of about 100g, this works out to 16 cents per gram. Even if we used the sale price, it's 12 cents per gram.

 

Someone a long time ago found Intel sells the OEM thermal paste for $1.10 (https://www.techpowerup.com/forums/threads/thermal-compound-show-down-and-the-winner-is.26300/). They claim it was enough for just one CPU, and going by a small tube of Arctic Silver 5 containing 3.5g of the stuff and useful for say 3 applications, let's go with 1g. This works out to $1.10 per gram.

 

The manufacturing process cost is probably going to be irrelevant anyway since either application would be automated.

 

EDIT: Also if there's any doubt of Dow Corning's TC-1996's performance, here's a table from Hardware Secrets. tl;dr, it got within 1C of the fabled NT-H1 in their test.

[DarkStang]

9 minutes ago, M.Yurizaki said:

Also to try to poke holes in the "Intel is saving money by using TIM", I picked a random lead-free solder from Amazon: https://www.amazon.com/gp/product/B01B61TWGY/ Use the original price of ~$16 and with a mass of about 100g, this works out to 16 cents per gram. Even if we used the sale price, it's 12 cents per gram.

 

Someone a long time ago found Intel sells the OEM thermal paste for $1.10 (https://www.techpowerup.com/forums/threads/thermal-compound-show-down-and-the-winner-is.26300/). They claim it was enough for just one CPU, and going by a small tube of Arctic Silver 5 containing 3.5g of the stuff and useful for say 3 applications, let's go with 1g. This works out to $1.10 per gram.

 

The manufacturing process cost is probably going to be irrelevant anyway since either application would be automated.

I didn't think the added cost was a material cost. I thought it was an added cost for assembly line speed and complexity. Even AMD went with TIM on their budget APUs because of cost.

[Mira Yurizaki]

2 minutes ago, DarkStang said:

I didn't think the added cost was a material cost. I thought it was an added cost for assembly line speed and complexity. Even AMD went with TIM on their budget APUs because of cost.

Once you automate the process and fine tune it, the cost should be irrelevant between the two processes. The only reason I see soldering being more expensive is from lower yield due to exposing the processor to high heat to melt the solder. But taking that out of the equation I don't think manufacturing costs would differ significantly.

[Trik'Stari]

1 hour ago, M.Yurizaki said:

There's no conclusive evidence because whatever Intel does is only known to Intel. But there was a really good post discussing it at: https://medium.com/@OpenSeason/soldered-cpu-vs-cheap-paste-59fb96a4fca7

 

If you want a summary of it:

• Intel's CPUs have concentrated hot spots relative to the die size, meaning uneven heating.
• Uneven heating is likely to cause higher thermal cycle stress on solder, which will create cracks/voids. Those will degrade the heat transfer performance over time
• Any large increases of temperature after delidding, especially with recent processors, may simply be due to removing the excess distance caused by the silicone sealant around the IHS.
• Author used a CFD simulation to test if higher thermal conductivity constant really mattered. They found out it doesn't after a certain point.
• Author made mention of having low thermal resistance should be a factor as well in looking at TIM performance over time. The TIM Intel chose is supposedly low enough.

How do you explain people trying delidding with something like Arctic Silver 5, and then liquid metal, and still getting FAR better results with the liquid metal, and only marginal results with the "normal" enthusiast thermal paste?

[Mira Yurizaki]

1 minute ago, Trik'Stari said:

How do you explain people trying delidding with something like Arctic Silver 5, and then liquid metal, and still getting FAR better results with the liquid metal, and only marginal results with the "normal" enthusiast thermal paste?

See the third point. Distance is a huge factor with pastes that don't have as high of thermal conductivity.

[Trik'Stari]

3 minutes ago, M.Yurizaki said:

See the third point.

Yeah that's what I am asking about.

 

That same removal only yields marginal results when replacing Intels thermal paste with Arctic silver 5, and then replacing that with liquid metal yields far better results than either of the two.

 

Which makes me think that that point is not entirely accurate.

[Mira Yurizaki]

Just now, Trik'Stari said:

Yeah that's what I am asking about.

 

That same removal only yields marginal results when replacing Intels thermal paste with Arctic silver 5, and then replacing that with liquid metal yields far better results than either of the two.

 

Which makes me think that that point is not accurate.

The thermal conductivity constant for Thermal Grizzly's Conductonaut is 78W/mK. AS5 is 8.9W/mK.

 

The other part that everyone seems to forget is the long term viability of the TIM. I'm pretty sure you must replace liquid metal TIM more often than standard paste.

[Trik'Stari]

2 minutes ago, M.Yurizaki said:

The thermal conductivity constant for Thermal Grizzly's Conductonaut is 78W/mK. AS5 is 8.9W/mK.

 

The other part that everyone seems to forget is the long term viability of the TIM. I'm pretty sure you must replace liquid metal TIM more often than standard paste.

I can't speak to that to be honest. It's holding up over a year now on the little office PC I built for my mom. I delidded the I3 and put on liquid metal as a practice run before trying on my I5 and later my I7 (haven't done the I7 yet, the I5 I screwed up but luckily it wasn't fried). I do replace the silicon sealant with RTV sealant. Just 4 little dots in the corners to hold things in place then I close the cpu retention mechanism.

 

Only uses a small noctua cooler but damn if it doesn't run cool lol.

[WoodenMarker]

4 hours ago, M.Yurizaki said:

I'm pretty sure you must replace liquid metal TIM more often than standard paste.

If you mean for delidding, that's not the case. Most thermal pastes experience pump out issues and deteriorate to pre-delid temps in a few weeks compared to little to no change over the course of years for liquid metal.

[Mira Yurizaki]

20 minutes ago, WoodenMarker said:

If you mean for delidding, that's not the case. Most thermal pastes experience pump out issues and deteriorate to pre-delid temps in a few weeks compared to little to no change over the course of years for liquid metal.

Settling to pre-delidding temps but staying without a need to replace it is still indicative of long term reliability. But otherwise good to know about LM.

 

Overall, I feel like people complaining about Intel's decision to use TIM instead of [insert favorite replacement here] is looking at it with enthusiast blinders. A vast majority of Intel's customers don't care how hot their CPU runs or how far they can push the overclock, as long as their computer lasts like 5+ years.

[WoodenMarker]

10 minutes ago, M.Yurizaki said:

Settling to pre-delidding temps but staying without a need to replace it is still indicative of long term reliability. But otherwise good to know about LM.

I'm not sure how reliable it is if temps keep getting worse. I haven't seen too many results but it's likely that temps can worsen beyond pre-delid temps due to continuous pump out. 

[AngryBeaver]

LM is the only way to fly when it comes to a solution between IHS and DIE.

 

Liquid metal is much more resilient to the massive up and down swings in temp that you see on a CPU or GPU DIE. A normal TIM would slowly break down over time. LM does a much better job at filling the micro gaps and imperfections on the bottom of the IHS and to a lesser point the top of the DIE. It can move more heat away more quickly.

 

Now the OP was mentioning a higher melting point LM solution, these do exist in the way of pads.

 

https://www.ekwb.com/shop/ek-tim-indigo-xtreme-intel-2011-3-haswell-e for example.

 

These are more meant for IHS to Block though.  They are used much like you mentioned. You put them on, boot windows, unplug your cooling solution and run a stress test for about 3-5 minutes. Then you shutdown the PC and let it rest for 30 minutes. Then you can plug it all back in and get up and running.