Sunday, June 26th 2022
De-lidded AMD Ryzen 7 5800X3D Has Vastly Improved Thermals
An AMD Ryzen 7 5800X3D processor that's been de-lidded (has its integrated heatspreader or IHS removed), posts vastly better thermals, according to Madness7771 on Twitter, who succeeded in de-lidding their 5800X3D. The stock 5800X3D posts significantly higher CPU core temperatures than a regular 5800X, due to its 3D Vertical Cache (3DV Cache) chiplet design, in which heat from the CPU cores is conducted through structural silicon, to the surface of the die-stack, from where the STIM conducts heat onward to the IHS.
A de-lidded 5800X3D reveals the 8-core "Zen 3" 3DV chiplet (CCD) next to a blob of structural material in the vacant area meant for a second CCD. With the residual STIM cleaned off, Madness7771 used a Conductonaut TIM and a Noctua NH-D14 to cool the processor. Madness7771 also posted some before and after temperature numbers for the processor (using the same cooler). It sees a maximum temperature drop from 80 °C to 70 °C, and average temperature drop from 78 °C to 67 °C, tested with a Forza Horizon 5 gaming workload. They also note that the peak temperature of the 5800X3D no longer reaches over 90 °C. De-lidding processors with STIM is a very risky process, and will destroy your processor if not done right.
Sources:
Madness7771 (Twitter), VideoCardz
A de-lidded 5800X3D reveals the 8-core "Zen 3" 3DV chiplet (CCD) next to a blob of structural material in the vacant area meant for a second CCD. With the residual STIM cleaned off, Madness7771 used a Conductonaut TIM and a Noctua NH-D14 to cool the processor. Madness7771 also posted some before and after temperature numbers for the processor (using the same cooler). It sees a maximum temperature drop from 80 °C to 70 °C, and average temperature drop from 78 °C to 67 °C, tested with a Forza Horizon 5 gaming workload. They also note that the peak temperature of the 5800X3D no longer reaches over 90 °C. De-lidding processors with STIM is a very risky process, and will destroy your processor if not done right.
42 Comments on De-lidded AMD Ryzen 7 5800X3D Has Vastly Improved Thermals
Anyway, glad you got your temperature in check. You might want to try PBO2 Tuner for even more gains.
Seriously?
Im using a 420mm AIO (Arctic LF II) on my 5800X3D that is not much better than NH-D14 and i get around 50-60c in a hot room with a custom power plan that disables downclocking of cores and around 70-80c in games with 375W GPU (2080 Ti custom) dumping heat into the case.
I've seen 90c only in prolonged stress tests that use all cores and even then with added GPU load.
People here are complaining about temps shooting up to 90 degrees when starting up an app. Well thats obvious and PBO at work. It just boosts as high as it can while it's within power / temperature wall. I mean should have tested clock for clock, and not a boost functionality. Basicly the user just voided his warrantly completely, for a avg 14Mhz extra speed. Congrats.
Solder itself is well conducting. And AMD has bin using that on all it's (higher) end to midrange models for quite some years. There's no benefit really delidding these CPU's, other then 3 degrees max. You have more problems delidding it then gaining from it.The whole reason why IHS's existed is to protect the core(s) from being chrushed.
They did had cussion pads but they provided poorly protection. There where quite some instances where an edge of the core was simply broken off due to wrong application of the CPU's heatsink. By adding a (soldered) IHS with copper inside of it, most of the problems where gone.
Downloading steam games too. Hit 40MB/s and for whatever reason one core of the 5900x starts pounding and getting REALLY hot, maybe 70c, for such a basic task.
One sample cant account a thousands of 5800X3D.
Now, let me express this in simple terms.
1) The Ryzen processor was delidded, which in this case is difficult because the IHS is fused to the silicon by a solder interface.
2) What was found was that only one CCX was used, and the other had a silicon wedge that took the void area.
3) Whenever the solder was cleaned from the silicon, and a cooler was directly contacted to the silicon there was a drop in temperature.
None of the above should be rocket science...but it seems to be. Let me explain the rhyme and reasons that people seem to not get.
1) Why would AMD include a dead chunk of spacer in their processor?
Have you seen it? The CCX being left blank would create an area where if the same force was applied it would create deformation of the IHS rather than proper contact? This is mechanics 101. They have the spacer so that when you crank down the IHS there's no issue with the cooler and IHS making contact and distributing force on the underlying silicon evenly.
2) So, then why would simply removing the IHS cause such a change?
Well, two things. The thermal mass before the cooler decreases. That structural blob heats up to the temperature of the CCX by virtue of being in the same package...but as it's doing nothing it acts as a battery to store heat. Likewise, the change from CCX->solder->IHS->thermal compound->cooler being decreased to CCX->thermal compound->cooler removes bits from the system. By removing thermal mass, allowing for less steps in heat transfer, and thus increasing the delta between environment and energy being dumped into the cooler you get better performance. It's not hard to see why?
3) Then AMD must be stupid...right?
No. Most people don't delid. Most people don't run these processors to the redline. By protecting the processors with the spacer, the silicon with an IHS, and not cheaping out on straight-up thermal paste (looking at you Intel), AMD is doing the best for the 99% of the customers who are going to pop their shiny new processors into a shiny new motherboard and do nothing else. While AMD isn't exactly a bunch on angels, they are not doing wrong here.
4) Well, 10 degrees is a huge difference. How do you respond to that captain snark?
Why thank you, I get to do some self depreciation. Most heat transfer is conduction related. Conduction related heat transfer is entirely driven by deltas in temperature from one surface to another. This means that along with thermal mass, the more steps you've got between the point of heat creation and heat dissipation into the environment the less each step is capable of actually doing things. This is why you don't just see processors cooled with a hundred pound block of copper. It's also why going from an environmental temperature of 20 C to 80 C is pretty much negligible when the "poor" condition is 20 C environmental to 90 C source. The delta between them is 60 versus 70 , or about 17%.
But, 17% more energy transfer is huge, right? Well, no. Let's look at the difference in surface area between the cooler fins and the surface area of the CCX that actually produces the part. Maybe a factor of a few hundred, but let's just call it 200. 17%/200 = 0.085 of a delta of 0.085% per the differentiation in surface area. This means that, within reason, the results are basically negligible differentiation that is far outclassed simply by the calculated differentiation in surface areas of conduction.
So, let's wrap all of this up. There's room to improve processor temperatures, assuming that you are willing to decrease the amount of interfaces between the cooler and the CCX. There's compromises in the manufacturing process, so that most of the people buying these processors don't run into issues. Finally, all of this is in service to decreasing temperatures...which could quite easily be improved by changing to a slightly more exotic cooler, with functionally no chance of destroying the underlying processor. This is a fantastic exercise in finding a halo situation...but also an exercise in expending huge efforts for minimal gains with poor conditional definitions, that are easy for the uninformed to make sweeping statements about.
Personally, the 5800x3d is only viable for gaming situations...because basically every benchmark for other uses confirms it isn't significantly better than a 5800x. If you've got a processor that does only 1 thing slightly better, with a premium price tag, and can demonstrate that with another huge risk and premium expenditure, could be slightly better, assuming that with all of the premium expenditures already made you didn't simply buy a cooler of higher energy transfer potential...then I don't understand your thinking.
This is the red neck theory of slapping a jet engine onto a lawn mower. Technically it is faster and better, but in practice it's easier to buy a lawn mower with a wider deck to get the lawn trimmed faster. The 5800x is your current mower. The 5800x3d has the jet engine. Thread ripper is the mower with a bigger (ie, wider) deck. I...would have thought this was cool a decade ago...as I sit staring at a 3930k that is now outclassed in every way by something 1/4th the price I know otherwise. If only technology always worked like that.