News Ryzen 9 7900X Delidded: Lower Temperature and 5.50 GHz on All Cores

[Admin]

Derbauer radically lowers AMD's Ryzen 9 7900X temperature with direct-die cooling.

Ryzen 9 7900X Delidded: Lower Temperature and 5.50 GHz on All Cores : Read more

 

[TR909]

There was a time there was no IHS on CPUs but "accidents" were happening from enthusiastic but inexperienced DIY'ers.
I believe INTEL and AMD should make advancements in that part of the CPU so that we are able to get the maximum performance from modern processors, while keeping temps at normal levels.

 

[-Fran-]

Thanks for making this its own article, as I hope AMD understands the problem is not running the CPUs at 95°C because we fear it'll melt or break, but because the IHS is so bad it's leaving a lot of performance on the table hurting efficiency.

I've heard and read other people say/write about delidding being just an extreme OC'er thing, but look at the data from a regular user perspective: the power draw to reach the same speeds goes down significantly without sacrificing performance; in fact, it goes up a lot! AMD just needs to improve how horrible bad the IHS performs to get some decent efficiency gains in Desktop, as in laptops we already have delidded CPUs an GPUs, so this is already seen there (where you could argue matters more?) and it shows how big the improvements are.

Maybe sell special delidded CPUs with a special mounting getting help from der8auer or something? There's a huge potential here for good press and AMD should capitalise on it. Unlike Intel and their complete avoidance of this topic, which keep in mind, also improves Intel's efficiency and they've been completely ignoring and saying "nope, not an issue". Don't be the same big headed moron, AMD. Come on.

Regards.

 

[who_farted]

i wish direct die cooling was mainstream. cpu's with a built on cooler.

 

[rluker5]

I think running a CPU really hot is a terrible inconvenience from an acoustic perspective. My pcs at home are tuned quiet. The ones at work that are cranking up the fans for trivial tasks give me feelings of disdain and anxiety.

I know these things can run a lot cooler if you undervolt/underpower/underclock them and still have great performance. If AMD had a suitable IHS and HONEST clock tuning this wouldn't be a problem. Maybe the worst thermal performance in a human generation is out of desperation, but I think it is a prepared selling point for future products that will offer better, cooler performance because they don't have that IHS that is becoming an icon for bad IHS designs.

You can literally get half a generation of improvement with a little clock tuning and a normal IHS. Seems like such a low hanging fruit that it was placed there on purpose. It isn't like AMD is new at this.

 

[wifiburger]

fun...

way too many things that can go wrong here with caps/resistors everywhere
somebody on reddit already killed his 7900x trying this

 

[helper800]

fun...

way too many things that can go wrong here with caps/resistors everywhere
somebody on reddit already killed his 7900x trying this

And that was the risk he took trying to do something cool. It is what it is. One mans failure is another mans fun.

 

[Phaaze88]

OK. How about how much more performance does this add though?
... and without the all core OC, letting the algorithm boost on its own?

 

[Kamen Rider Blade]

The main issue with the current IHS compared to the previous one is you lost ALOT of flat surface area to transfer heat between the IHS and the CPU cooler.

Yes, the THICC-er heatsink helps, Intel already did this before as well, it gives you extra heat soaking capacity.

But you still need more flat surface area to contact your CPU cooler for heat transfer and I hope AMD redesigns the IHS to fill out those large gaps by next gen Zen 5.
Losing that much Flat Surface Area is incredibly bad for Heat Transfer and that's why I think we need a improved IHS.

Spoiler: This is how much Surface Area you're missing.

By my rough estimates, the current design lost ~30% of it's flat contact surface area on the outside to transfer heat between the IHS & CPU Cooler.

That WILL have an affect on temperature for the entire CPU.

 

[btmedic04]

Optimum tech achieved a 30c and 50w reduction just with curve optimizer, setting a lower ppt and changing the thermal limit. All while maintaining stock out of the box performance on a 7700x.

 

[dk382]

The main issue with the current IHS compared to the previous one is you lost ALOT of flat surface area to transfer heat between the IHS and the CPU cooler.

Yes, the THICC-er heatsink helps, Intel already did this before as well, it gives you extra heat soaking capacity.

But you still need more flat surface area to contact your CPU cooler for heat transfer and I hope AMD redesigns the IHS to fill out those large gaps by next gen Zen 5.
Losing that much Flat Surface Area is incredibly bad for Heat Transfer and that's why I think we need a improved IHS.

Spoiler: This is how much Surface Area you're missing.

By my rough estimates, the current design lost ~30% of it's flat contact surface area on the outside to transfer heat between the IHS & CPU Cooler.

That WILL have an affect on temperature for the entire CPU.

The AM4 IHS fit entirely within the solid part of the AM5 IHS. The AM5 IHS is thus bigger and has more surface area than the previous IHS. It's not the surface area that's the problem here, it's the thickness. More thickness isn't better. You don't want the IHS to have a lot of thermal mass because then it becomes harder to dissipate that heat. You want it to have very little thermal mass but be highly efficient at transferring the heat to a cooler.

 

[Kamen Rider Blade]

The AM4 IHS fit entirely within the solid part of the AM5 IHS.

The AM4 & AM5 IHS is the same foot print.
They both have a identical 40 mm x 40 mm base PCB Substrate that they sit on.
So no, that's not what is happening.

The AM5 IHS is thus bigger and has more surface area than the previous IHS.

The base of the IHS between AM4 & AM5 occupy the same area / foot print on the PCB.
Thus the AM5 IHS ISN'T bigger than the AM4 IHS. They are identical in foot print at the base.

It's not the surface area that's the problem here, it's the thickness. More thickness isn't better. You don't want the IHS to have a lot of thermal mass because then it becomes harder to dissipate that heat. You want it to have very little thermal mass but be highly efficient at transferring the heat to a cooler.

The new Cut-outs reduced the surface area and IHS volume, but because the Z-Height is taller thanks to LGA, they made it "THICKER" to match the Z-Height for AM4 to maintain backwards compatibility.

How much of that Surface area is contacting the CPU cooler? ~30% less is contacting the CPU cooler's flat bottom plate.

Intel did the same thing previously with it's CPU and made their IHS thicker while maintaing the same contact surface area between the IHS & CPU Cooler base.

Spoiler: Previous Intel IHS change by making it THICKER.

AMD cut the amount of Surface Area of the IHS making contact with the CPU cooler base.

That's why it's less efficient at transferring heat to your CPU cooler.

There's ~30% less surface area coming into physical contact with your CPU cooler's bottom flat plate.

The trade off was a THICKER IHS to compensate, but that can only do so much.

Derbauer made in entire video about this issue.

 

[wifiburger]

The IHS is one trash design, it's 5.8mm height worth of copper

You should be able to sand this down with 150grit and remove 3mm from that ihs
The limit being 3.8mm before reaching the LGA bracket / ihs legs.

It should net you the 20c cooler with regular paste without de-lid if you can lower your cooler by 3mm.

 

[The Historical Fidelity]

AMD, bring back the athlon XP design!!!
https://www.legitreviews.com/images/reviews/554/athlonxp_agoia.jpg

 

[jp7189]

I'm not understanding the problem with a thick IHS. From what I understand, it would only a problem if the heatsink on top is vastly better at thermal conductivity, but thats not likely as the IHS is nickel clad copper. Sure you want TIM as thin as possible because it sucks compared to copper. If the IHS is copper and your heatsink is copper what difference does it make where the thickness is? If the heatsink is aluminum (which most are), then I would think a thicker copper IHS is a good thing.

 

[The Historical Fidelity]

I'm not understanding the problem with a thick IHS. From what I understand, it would only a problem if the heatsink on top is vastly better at thermal conductivity, but thats not likely as the IHS is nickel clad copper. Sure you want TIM as thin as possible because it sucks compared to copper. If the IHS is copper and your heatsink is copper what difference does it make where the thickness is? If the heatsink is aluminum (which most are), then I would think a thicker copper IHS is a good thing.

Anything thick degrades thermal conductivity. You want it as thin as possible due to the formula: C thermal = W / (m*K) which means the number of watts (W) of thermal energy transfer decreases with an increase in material thickness (m). So even though the IHS is made of the second highest rated thermally conductive metal, if it’s too thick then performance degrades.

 

[_dawn_chorus_]

Silicon Lottery sadly closed their doors too soon...

I'd never attempt to delid a CPU myself but I love my delided 8700k I got from them years ago. Keeps my rig running quiet when it never reaches over 65c.

 

[Kamen Rider Blade]

AMD, bring back the athlon XP design!!!
https://www.legitreviews.com/images/reviews/554/athlonxp_agoia.jpg

Not going to happen, direct die was a RMA nightmare for those who implemented it.

 

[Kamen Rider Blade]

Anything thick degrades thermal conductivity. You want it as thin as possible due to the formula: C thermal = W / (m*K) which means the number of watts (W) of thermal energy transfer decreases with an increase in material thickness (m). So even though the IHS is made of the second highest rated thermally conductive metal, if it’s too thick then performance degrades.

I find doubt in that claim.
Intel did the same thing with their IHS and made it "THICK-er". But they didn't lose any IHS Contact Surface Area and it helped them with their temps.

Spoiler: Intel made their IHS Thicker in the past

The major issue is that while AMD made their IHS THICKER to compensate for loss in mass in the cut-out, they lost ~30% of the Contact Area between the IHS & the CPU Cooler Base.

Losing 30% of your Flat Contact Area will affect performance, having the extra Thick IHS is a way to compensate, but I think it wasn't enough.

AMD made those cuts for automated manufacturing reasons, but I did the estimates and having a AM4 like IHS with only just enough cut-out area for the tabs would lead to a 2% loss in Flat Surface Area.

Spoiler: AMD IHS and how much contact area is missing

Notice how all the cut-outs that AMD chose for automated manufacturability assembly, that leads to a ~ 30% loss in Contact Surface area.

If AMD can make a new IHS that still had the cut-outs for the SMD Capacitors & Transistors at the feet, but filled out the top end of the IHS and left only just enough room for the contact frame's tabs, then they would only lose ~2% of Contact surface area. Yes that comes out to ~1% on the Left & Right side each. I did the napkin math and it would be great for AMD if they can pull it off in time for Zen 5. A side benefit is that they would have a IHS with even more Thermal Mass to act as a buffer before the CPU cooler transfers the heat away.

 

[-Fran-]

I find doubt in that claim.
Intel did the same thing with their IHS and made it "THICK-er". But they didn't lose any IHS Contact Surface Area and it helped them with their temps.

Spoiler: Intel made their IHS Thicker in the past

The major issue is that while AMD made their IHS THICKER to compensate for loss in mass in the cut-out, they lost ~30% of the Contact Area between the IHS & the CPU Cooler Base.

Losing 30% of your Flat Contact Area will affect performance, having the extra Thick IHS is a way to compensate, but I think it wasn't enough.

AMD made those cuts for automated manufacturing reasons, but I did the estimates and having a AM4 like IHS with only just enough cut-out area for the tabs would lead to a 2% loss in Flat Surface Area.

Spoiler: AMD IHS and how much contact area is missing

Notice how all the cut-outs that AMD chose for automated manufacturability assembly, that leads to a ~ 30% loss in Contact Surface area.

If AMD can make a new IHS that still had the cut-outs for the SMD Capacitors & Transistors at the feet, but filled out the top end of the IHS and left only just enough room for the contact frame's tabs, then they would only lose ~2% of Contact surface area. Yes that comes out to ~1% on the Left & Right side each. I did the napkin math and it would be great for AMD if they can pull it off in time for Zen 5. A side benefit is that they would have a IHS with even more Thermal Mass to act as a buffer before the CPU cooler transfers the heat away.

You analysis is very unidimensional.

There's way more to an IHS than just how chonky it is: material, TIM used, contactarea, flatness and polish. You're using a cheap "cop out" by just focusing on the contact area. As for your "Intel made it better", sorry but that's laughable. No solution is perfect and Intel has had most of the most dumb solutions around in different gens. Or are you forgetting their stupid bending issue that affects contact area?

Again, not uni-dimensional.

Regards.

 

[The Historical Fidelity]

I find doubt in that claim.
Intel did the same thing with their IHS and made it "THICK-er". But they didn't lose any IHS Contact Surface Area and it helped them with their temps.

Spoiler: Intel made their IHS Thicker in the past

The major issue is that while AMD made their IHS THICKER to compensate for loss in mass in the cut-out, they lost ~30% of the Contact Area between the IHS & the CPU Cooler Base.

Losing 30% of your Flat Contact Area will affect performance, having the extra Thick IHS is a way to compensate, but I think it wasn't enough.

AMD made those cuts for automated manufacturing reasons, but I did the estimates and having a AM4 like IHS with only just enough cut-out area for the tabs would lead to a 2% loss in Flat Surface Area.

Spoiler: AMD IHS and how much contact area is missing

Notice how all the cut-outs that AMD chose for automated manufacturability assembly, that leads to a ~ 30% loss in Contact Surface area.

If AMD can make a new IHS that still had the cut-outs for the SMD Capacitors & Transistors at the feet, but filled out the top end of the IHS and left only just enough room for the contact frame's tabs, then they would only lose ~2% of Contact surface area. Yes that comes out to ~1% on the Left & Right side each. I did the napkin math and it would be great for AMD if they can pull it off in time for Zen 5. A side benefit is that they would have a IHS with even more Thermal Mass to act as a buffer before the CPU cooler transfers the heat away.

You are right that the 30% loss in thermal interface area is correct, however my assessment on increasing IHS thickness is still true and affects ryzen 7000’s ability to transfer thermal energy to the cooling solution way more than losing 30% IHS contact area. Physics formulas do not lie, if C thermal = W / (m*K) then yes 100% of the time if you increase the thickness of the material then you lose thermal conductivity. This same formula also demonstrates that losing IHS contact area doesn’t affect temperatures as much as you think because the shortest distance between die and top of IHS (m) is directly above the dies themselves. The cutouts are so far away from the dies that (m = drawing a line from edge of top of die to top of IHS at location of cutout) the (m) thickness results in severely diminished thermal conductivity.

And your example that Intel increased their thickness of their IHS is missing crucial information. They increased IHS thickness (copper = 401 W / (mK)) to make up for them shaving off excess silicon thickness on the die itself which they did to increase thermal conductivity of the die itself which has a much lower thermal conductivity (silicon = 148 W / (mK)). When doing the math, decreasing the silicon thickness while increasing copper thickness results in a higher effective thermal conductivity of the die-IHS system vs before. Since zen 4 dies are the same thickness as zen 3 dies, increasing IHS thickness can only reduce thermal conductivity.

 

[jp7189]

Anything thick degrades thermal conductivity. You want it as thin as possible due to the formula: C thermal = W / (m*K) which means the number of watts (W) of thermal energy transfer decreases with an increase in material thickness (m). So even though the IHS is made of the second highest rated thermally conductive metal, if it’s too thick then performance degrades.

I appreciate the explanation, but I'm still not getting it. The thinner the IHS, the faster you get to to air which is relatively high heat resistance. Why is getting to high resistance materials faster a good thing? As far as I can imagine, you're idea is only valid if the ultimate destination has no resistance. If the IHS was a foot thick, wouldn't that do a fantastic job of keeping the CPU cool?

 

[Kamen Rider Blade]

You analysis is very unidimensional.

There's way more to an IHS than just how chonky it is: material, TIM used, contactarea, flatness and polish. You're using a cheap "cop out" by just focusing on the contact area. As for your "Intel made it better", sorry but that's laughable. No solution is perfect and Intel has had most of the most dumb solutions around in different gens. Or are you forgetting their stupid bending issue that affects contact area?

Again, not uni-dimensional.

Regards.

So far, AMD hasn't changed their "Flatness Profile" and AMD's IHS is flatter than Intel's slight convex IHS.
As far as Polish, I can't detect a difference from pictures as to any textural difference between AM4 & AM5 via photos.

Yeah, as far as I know. AMD hasn't had a MoBo bending issue because their CPU's are still square instead of rectangular.

The TIM used underneath the IHS is still Solder. As for TIM on top of the IHS, that's a end user choice.

In the end, the major difference is the mass & thickness of the IHS & the Contact Area difference.

You are right that the 30% loss in thermal interface area is correct, however my assessment on increasing IHS thickness is still true and affects ryzen 7000’s ability to transfer thermal energy to the cooling solution way more than losing 30% IHS contact area. Physics formulas do not lie, if C thermal = W / (m*K) then yes 100% of the time if you increase the thickness of the material then you lose thermal conductivity. This same formula also demonstrates that losing IHS contact area doesn’t affect temperatures as much as you think because the shortest distance between die and top of IHS (m) is directly above the dies themselves. The cutouts are so far away from the dies that (m = drawing a line from edge of top of die to top of IHS at location of cutout) the (m) thickness results in severely diminished thermal conductivity.

Backwards compatibility with AM4 coolers was a desired/designed/required feature. You have some Give & Take.

And your example that Intel increased their thickness of their IHS is missing crucial information. They increased IHS thickness (copper = 401 W / (mK)) to make up for them shaving off excess silicon thickness on the die itself which they did to increase thermal conductivity of the die itself which has a much lower thermal conductivity (silicon = 148 W / (mK)). When doing the math, decreasing the silicon thickness while increasing copper thickness results in a higher effective thermal conductivity of the die-IHS system vs before. Since zen 4 dies are the same thickness as zen 3 dies, increasing IHS thickness can only reduce thermal conductivity.

Then maybe we need to shave the Die Package to make up for the thicker Copper IHS layer on top.

Maybe add in Zinc Coated Copper Shims and use Liquid Metal to bond the shim to the bottom of the IHS and top of the Thin Silicon Package.

I know this guy used Thin Copper Shims and got dramatic temperature improvements on his RAM package.

Copper Modding Helps Cool A Toasty GPU

[DandyWorks] had an NVIDIA RTX 3070 Ti GPU, and found it was running incredibly hot, with the card’s memory hitting temperatures of 110 °C. He decided to try “copper modding” to s…

hackaday.com

 

[TerryLaze]

I appreciate the explanation, but I'm still not getting it. The thinner the IHS, the faster you get to to air which is relatively high heat resistance. Why is getting to high resistance materials faster a good thing? As far as I can imagine, you're idea is only valid if the ultimate destination has no resistance. If the IHS was a foot thick, wouldn't that do a fantastic job of keeping the CPU cool?

Because if you are an overclocker you have super strong cooling, often liquid nitrogen...so the thinner the ihs the sooner the LN2 can start working.
You are thinking the same way the companies or a normal customer would think about it, the thickker the ihs the more slack there is for your normal cooling.
The people here are discussing it from the point of view of pro overclockers...

 

[jp7189]

Because if you are an overclocker you have super strong cooling, often liquid nitrogen...so the thinner the ihs the sooner the LN2 can start working.
You are thinking the same way the companies or a normal customer would think about it, the thickker the ihs the more slack there is for your normal cooling.
The people here are discussing it from the point of view of pro overclockers...

Point well made. I was not considering ln2.