How high SHOULD you ACTUALLY overclock a 3990x?

[FL-Chan]

Premis:

The Threadripper 3990x has seen some amazing benchmarks, having been clocked above 5 GHz.
But these results are achieved with what seems at a glance to be unsustainable, potentially corrosive cooling if not done with absolute expertise; methods like liquid nitrogen, liquid helium, and mineral oil cooling seem too complex and risky.

 

Original Questions:
so what is the 3990x's limit with consumer solutions like custom cooling loops, radiators, fans- what are the best clock speeds achieved with sustainable consumer cooling methods?
are there safe and (relatively) simple options that I'm missing like some sort of chilled cooling loop? (like a safer equivalent of the one in this LTT video: https://www.youtube.com/watch?v=HMtvEbD2MQo&ab_channel=LinusTechTips )
or is a well thought out custom cooling loop the best choice?
To be clear, the question is, what is the maximum I can get out of my 3990x (or my 3090 strix for that matter) without having it wear out or fail prior to a decade of good use?

 

for context:
I'm building a near the best money can buy type build, a 3090, 3990x, with cost being saved on bare minimum storage, ram, and cooling - all of which I want to progressively upgrade over time to remove bottlenecking on the CPU & GPU and to allow for overclocking. I'll start with a 240mm aio for the TR with no overclocking, but want to mature the build to it's highest sustanainable output - specifically for 3D, from modelling and texturing, all the way to animation with particles and complex environments (blender, maya, Zbrush, substance painter, Mocap), for game dev (VR game dev, VR Game testing, 3d game dev, 3d game testing), for Music production (FL Studio, Ableton) and lastly for video editing (likely adobe premiere down the line.)

 

[Bombastinator]

I actually googled for a video for this one

not the real quote of course. The real quote is

 

Mineral oil cooling is sustainable, sort of. The problem is it’s a huge mess eventually, though that mess may be years off.  This also applies to sealed phase change exotic liquid stuff.  The most important part is going to be silicon lottery.  If you’ve got a lead chip there isn’t going to be much that can help it much, and if you have a gold chip it’s pretty easy.

Edited June 28, 2021 by Bombastinator

[FL-Chan]

Thanks for mentioning the silicon lottery, I wasn't familiar with that but now I know.

so what I gather is this:
1. optimize cooling, choose bios which allows necessary control, remove bottlenecks, but go crazy scraping out that little extra bit of performance.

2. incrementally raise single variables one at a time in your bios, testing performance along the way.
3. when you see thermal throttling, set back your overclock a few dozen Mhz, maintain at or below AMD's recommended 95°C, at or below thermal throttling, or play safely and keep temperatures even lower.

 

follow these steps and this should be the best performance you can get out of your CPU without noticably reducing it's lifespan (I.E, it would be obsolete by the time this kills it).
 

Can anyone correct me where I might be off or help me be more specific with this, or is this basically the right idea?

Edited June 28, 2021 by FL-Chan
Edited to make the solution highlight more concise.

[Bombastinator]

22 minutes ago, FL-Chan said:

Thanks for mentioning the silicon lottery, I wasn't familiar with that but now I know.

so what I gather is this:
1. optimize cooling, choose a good bios, remove bottlenecks.

2. incrementally raise single variables one at a time in your bios, testing performance along the way.
3. when you see thermal throttling, set back your overclock a few dozen Mhz, maintain in a safe 85 or 90 degrees Celsius.
 

this should be the best performance you can get out of your CPU without noticably reducing it's lifespan (I.E, it would be obsolete by the time this kills it).
 

Can anyone correct or help me refine this?

The bios thing is less important, or at least more accurately most bioses are reasonably decent.  There are a few actual bad ones though and if you simply avoid those it should be more or less ok.  As far as bottlenecks go most of the time they don’t matter very much.  Literally everything bottlenecks some.  A completely balanced system can only even be completely balanced in specific situations.  It’s a bit like the bios thing in that if one can avoid grotesque problems it’s more or less ok. Neither should be overthought too much.  As far as temp goes I’ll call that one a religious discussion.  You like 85-90c. That’s fine. It’s an opinion.  There are people who won’t do over 75c and there are people who think that a given chip’s throttle temp is the only true thing so one might as well run right up to it. I have no idea who’s right on that one. 

[FL-Chan]

Haha, right up near the throttle temperature I go then! 

so any bios that gives me the necessary control is good enough, some bottlenecking is inevitable, so just optimize what you have without going nuts about scraping out the remaining fractions of performance, and run at temperatures up to the level of risk you're willing to take, either well below for a more safe and conservative use, or right up to the thermal throttling if you're comfortable with that.

thanks for the input!      d(^ヮ^)

[Bombastinator]

1 hour ago, FL-Chan said:

Haha, right up near the throttle temperature I go then! 

so any bios that gives me the necessary control is good enough, some bottlenecking is inevitable, so just optimize what you have without going nuts about scraping out the remaining fractions of performance, and run at temperatures up to the level of risk you're willing to take, either well below for a more safe and conservative use, or right up to the thermal throttling if you're comfortable with that.

thanks for the input!      d(^ヮ^)

More or less.  As far as card balance and bottlenecking go, I remember mentioning once that my decrepit old lead 4770k would do 80+% load when the video card (a 580) was flat out in fallout4 people said that was unusually good balance and there wasn’t a lot of point in me adding much more gpu performance to the machine.   So my suspicion is anything above 50% is not particularly unusual. It’s when BOTH the cpu and gpu are running at low levels that one needs to look into things.  
 

As far as temps go my machine is unusually long lived.  It’s a 4770k that started at the very top of things (sort of a 5950 equivelant) in 2014 and today is somewhat slower than a 3100.  Not sure how much.  So it’s lost about 75% of its value in that time. I understand that the thing is well behind the somewhat creepy limit microsoft seems to be putting on win11, so in 5 years it will be totally obsolete and not even safe to turn on without an air gap.  Kinda like a machine running win7 is now. That chip never saw over 78c and was usually below 70c 99% of the time.  It’s gained about 1 degree of heat in that time. I suspect if I kept it going it would continue into full obsolescence when win10 runs out in 5 years and then probably go another 15.  Caps on the motherboard will probably give out before the cpu does. 
 

As far as silicon lottery goes my 4770k lost.  I can’t keep it really stable above 4.0ghz so that’s what it’s saying at pretty much since I bought it. It’s a lead chip.  Gold 4770ks could do above 4.7ghz.  It will run that fast, but it will crash every couple hours.

Edited June 28, 2021 by Bombastinator