AIR vs AIO. 18c drop in temps

[Robert896r1]

I know this batted around a lot so I figured I'd share my findings. The results are below and everything is exactly the same in bios. The air cooler was cleaned. All fans are 100% I use realbench not because of some magic reason. Only because I knew I could run it just shy of throttling the CPU on air.

Settings:
9900k
HT OFF
5.2ghz
1.305v in bios
z390 aorus pro
Turbo LLC
46x core
Case = haf x with side panel off

Dark Rock 4 results:

Alphacool 360LT results:

 

[Robert896r1]

The AIO was already well warmed up with blender benchmark, time spy cpu loops and such.

The chip isn't binned or de-lidded. Just a retail unit I've said since launch.

Noise at 100% fans is well acceptable for benchmarking and waaay lower than my GB 2080ti when it warms up. In reality, it never gets past 60's during any gaming session such as BF5 which would put the DR4 with spikes around 80's.

Vcore is exactly the same. VR VOUT is what you need to look at on GB boards.

The 9900k even with HT off dumps a ton of heat and has thermal density issues which makes it difficult to cool. Most often, you'll be thermally limited on a 9900k more so than anything. I can easily make this AIO hit 100c by just enabling HT and bumping up the vcore to what's needed.

 

[Robert896r1]

Funny you talking about VR VOUT because this has been bugging me. I have set LLC to High, unstable on normal and Turbo to much for my liking, on GB 5 Gaming and IA AC 1/IA DC 1 to try and tighten Vboost. @1.34v in bios I will still see spikes of 1.39V idleling according to Vcore in HWinfo. So this a wrong value to follow and I should be looking at VR VOUT on GB boards?

I found little benefit in running adaptive voltage. The power savings is almost a rounding error if you do the following: I do the following: use a fixed vcore, disable c-states, enable speed step and in windows use a "balanced" power plan. This lets my core clock to drop down to 800mhz (and in between) while my vcore remains predictable. When I'm away or idle, power on CPU Package drop down to around 13w which is fine.

I'm don't know much about a Gaming 5 so can't comment what the right llc is but on the Aorus series, the vdroop on high is just too much and makes you run a decently high idle to make up for it. On my board, Turbo is only level 3 of max. Extreme and Ultra Extreme are above it. I don't use those.

 

[Robert896r1]

The Dark Rock 4 is not a good cooler when pushed: https://www.hardocp.com/article/2018/04/30/be_quiet_dark_rock_4_cpu_air_cooler_review/3

If you're not on a power hungry chip with thermal density issues, I'm sure it'd be fine but for this application, it isn't. I'm happy to see someone do their own testing. I was transparent in my settings above so feel free to duplicate and post results.

 

[Robert896r1]

Apparently its only a 3-4c difference between the Dark rock 4 and Pro 4 version

https://www.guru3d.com/articles_pages/be_quiet_dark_rock_pro_4_review,7.html

Here's the exact problem with that review and if you read it, they even kinda mention it but excuse it:

"We've been battling the question whether or not to actually use a Core i7 4790K on a Z97 motherboard. The Haswell processors all have poor heat transfer from the silicon die to the IHS."

So the reviewer is fully aware that most of the heat issues are between the IHS and Die meaning that coolers are only doing surface heat transfers *BUT* the reviewers doesn't state that will greatly limit cooling performance which should have been the next step.

This is not the reviewers fault but we're now comparing a 4c/8t thread that has much better thermal density than Coffee Lake Refresh. You can watch more on thermal density here: https://www.youtube.com/watch?v=uLVzRY27A-I&t=1046s

I ended up buying a dark rock 4 back when I built my PC by relying on poorly formulated reviews and being knowledge limited as I hadn't been tracking PC trends since Sandy Bridge days. It's ultimately my decision to buy it so I take full responsibility but I'm glad to have moved past it. For the record, I had the same AIO before and it died within 30mins. So all is not rosy on the AIO front either.

If you want to look at good reviews for coolers, look at their methodology. That's way more important than the results page.

 

[Robert896r1]

I thought it was the Pro lol. Would be useful to see temps on a Noctua NH-D15.

So, as to this, Prime AVX will push one of my cores above 90C on a 8600k @5Ghz 1.34v. I offset 1 AVX to keep it under 90C. This is on a Noctua NH-D15S, expected or to hot? I've reseated it twice using Noctua's included thermal paste. Airflow is not a problem with full NZXT Phantom that I have custom modded with 2 200mm Phanteks, 120mm+140mm

When I say custom modded I removed the HDD bay allowing for better airflow.

If you're not de-lidded, adding more cooling will have diminishing returns as most of the heat is trapped between the ihs and die and not being directly transferred to the IHS. De-lidding will net you a ton of temp savings. Way more than any air/aio/loop.

 

[Robert896r1]

Interesting data, but what is your case fan setup with each setup? Was air cooler getting air the same temp as radiator is? Below link is basic guide to airflow and optimizing case airflow by monitoring airflow temp into coolers. You might find it of intereest.
https://forums.overclockers.co.uk/t...-i-put-my-temp-sensor.18564223/#post-26159770

It's a haf x with the side panel removed. Rear exhaust fan and had a table fan blowing cooler air on the DR4 to avoid throttling. Without the fan, it'd throttle or crash out.

AIO had no fan assitance. Mounted top with top exhaust and what was the rear exhaust fan got flipped to rear intake to provide some air flow over the vrm's.

Ambient temps for both were within 1 degree in the same location.

 

[Robert896r1]

To show how the temps scale with voltage on the AIO, here is a 1 hr run of real bench with the same settings as OP EXCEPT:
- vcore is now 1.35v in bios
- Multiplier 53x



As I stated earlier, it's easy to even max out this AIO on a 9900k. Same settings as op except:
- HT ON
- Core 52x
- 1.35v is bios
*note the massive jump in CPU power package and a corresponding drop in vr vout minimum. We went from 118amps to 152amps thus the massive jump in heat. This is on the same voltage in bios! As a reference cinebench in the non HT setup above is around 180w"

A few things to note about cooling a 9900k. The good news is that with HT on, you'll never have to worry about too much vcore as you'll always be thermally throttled. Direct die can shave about 10c. After that you get into oversized and multiplier radiators, chillers, dry ice and LN2.

My particular chip hits a voltage wall between 5.3 and 5.4 with HT off. What does that means? The jump above you see going from 5.2 to 5.3 with HT off was 0.45v. However, going to 1.42v and 5.4ghz which is a .70v jump is barely stable booting into windows, never mind running a decent benchmark. All in all, my sweet spot is 5.3ghz. I also can't cool the chip at 1.42v/5.4ghz even with AIO on max.

If your biggest system stress is gaming, you can use use realbench as your test suite. I'd recommend finding the lowest voltage it'll run stable at then add .020v as a buffer to cover random scenarios. 1hr test should be ideal. If you're going do a lot of number crunching and scientific modeling, you need to use P95 as your baseline. It's more stressful so you'll need to raise your vcore or drop multiplier compared to realbench. If you're doing rendering, download blender benchmark and run the long test which takes about an hour.

Hopefully the info above helps those with more thermally limited chips such as the 9700k/9900k and the upcoming 9900ks.

 

[Robert896r1]

https://www.youtube.com/watch?v=gAy9uNnFr0I&app=desktop


Good video of high end air coolers dealing with an average oc of a 9900k.

 

[Robert896r1]

Comparing the OP's AIO to his air cooler is not exactly fair as someone said its not exactly high end.

In all good reviews it shows that the best of the high end air Coolers (Noctua D15 etc) are as good or better than AOI's esp for performance to noise ratio, custom water is another story though as obvious it going to be cooler

As i linked this video above
https://www.youtube.com/watch?v=gAy9uNnFr0I&app=desktop You can see how a D15 (and other high end air coolers) perform understress.

Here's a 15min run using his settings on the alphacool. 12-13c difference is a lot. If I could drop my temps by that much, I could run 53x daily.

The settings I copied:
50/43x
1.33v bios
15min loop of R20. File, preferences, minimum time duration = 900seconds



On an overclocked 9900k, a good AIO (swiftech, alphacool, ek) is a notable step above air coolers. I plan to do similar testing when I go custom and direct die as I want to have another 10-12c off my temps.

 

[Robert896r1]

I do not trust your readings as you are the only person including reviews to state this, none of those videos matter as you would need compare both on your system so would need one of the best air coolers then even if readings are weird they are apples to apples.

There's no reason to trust me. Simply, go find a 9900k being air cooled while pulling 150amps and 225w. Let me know what you find.

 

[Robert896r1]

No surprise to me https://youtu.be/RjRFXsjK414?t=305

 

[Robert896r1]

A 4770k has nothing in common with a 9900k. There have completely different thermal density, power outputs. Not to mention paste vs solder.

Here is something in 2019 that's directly relevant rather than digging artifacts: https://youtu.be/RjRFXsjK414?t=305

 

[Robert896r1]

While paste generally[] does not transfer heat from chip to IHS as well as solder, both have chips generating heat and moving that heat to IHS where it is transferred through TIM into cooler / waterblock, then into liquid that evaporates and moves through heatpipes to fins or is pumped through hoses to radiator.

There are many more variables involved in these reviews then the few differences between 4770k and 9900k. Very few even monitor air temp into cooler / radiator .. I don't know of any review test site that does anything like using pressure sensitive paper to determine how good / poor the mating of IHS to cooler / waterblock, and that alone can easily be 4-5c difference in recorded temps. So combine IHS to cooler / waterblock and a 2-5c possible difference between air actually entering cooler / radiator versus room air temp and we have a combined total 6-10c that can be accounted for as margin of error in test results.

I'm not saying it is that much, just pointing out 3-5c difference in CPU temps is about as close to '0c' as we can be sure we have .. meaning 65-67c for one cooler and 68-70c for another is both performing about the same.

There's nothing 3-5c in my example that I've posted or the techyescity video I posted or the craftcomputing video I posted.

This isn't a hard thing to solve. Simply demonstrate an air cooler getting 150a/220W off a 9900k without thermal throttling. It's a math problem.

In the same way that an AIO is not going to cool a 200a/300w 9900k oc. Each solution has it's limits. A good AIO has a higher threshold for heat dissipation than an air cooler.

Again, it's math. Not magic.

 

[Robert896r1]

Math yes, but it's not as simple as 200a/300w vs 150a/220w. Math is sometimes nice, but I've seen way too many time that physical application was radically different than math had calculated. Often math can explain things found by physical application after the fact better than it can predict what will happen. It is often extremely hard to make sure all the variables are included in original math calculations. The end result can be quite different from what your math says it will be.

Have you seen the detailed testing of custom loop, AIO, CLC and a few top tier air coolers done by [H]ard|OCP did on 2990wx? You might find it interesting.
http://m.hardocp.com/article/2018/10/09/noctua_amd_threadripper_air_cooler_roundup/8
http://m.hardocp.com/article/2018/09/28/threadripper_air_coolerssilver_arrow_tr4_vs_wraith_ripper/7

Again. That's not a 9900k. TR has a massive IHS and can dissipate heat better.

That's as relevant as saying "look how well a cheap AIO handles a 2080ti core"

You're having a hard timing finding those results because they don't exist so you keep trying to deflect. Come back with an air cooler dissipating 150a/220w on a 9900k. You won't.

 

[Robert896r1]

Top AIO's can use the their performance to allow higher thermal headroom than air coolers. Price, noise, longevity is something you'll have to determine for your use case.

Want to leave your machine overnight rendering for blender/handbrake etc? On a chip like the 9900k, the AIO will give you the thermal headroom as such programs will hammer the CPU continuously.

Want to only play games and use normal desktop apps? air cooling will be fine.

Your use cases and the thermal budget you need for your chip should determine what cooling you need.

 

[Robert896r1]

So equal and yet no one can point to an air cooler handling a 9900k dumping 150+ amps sustained.

Anyway, if/when I moved to an open loop, I’ll share those results using the same test suite and highlight the thermal limits of the 9900k there also.

Then we can have data points of air/aio/open on a 9900k

 

[Robert896r1]

I guess the AIO, as is expected from liquid cooling, would "tolerate" few spikes in temperature before the coolant increase its temperature, under normal use, clearly.
Under load, a fair test is always long enough to allow the heat to "saturate" the coolant, then see how efficient the radiator and fans are cooling the coolant down. But you know that.
Custom loop, quite often oversized, so quite hard to saturate and reach the limit of the radiator's capabilities.
It's quite challenging to push this CPU with air or AIO, as any unnecessary voltage increase would likely hold back few MHz. Custom would allow a bit more room.

Yeah. For me to run 5.2 daily and be comfortable, I’d want to drop between 8 and 10c off current. I can do that via direct die or custom loop.

 

[Robert896r1]

150+ amps isn't normal

Run blender. Run handbrake. Easily done. Do you have experience with a 9900k?

 

[Robert896r1]

it isn't normal if you overclock the **** out of the chip. The issue with that chip is getting the heat out to IHS not from the IHS to the cooler. If you are having to run such extreme cooling I'd suggest your OC is far too high for normal usage

Do you actually have experience with a 9900k?

What's extreme about the 360 AIO I'm using?