Thermosyphon Cooling Solution

[EarthDog]

Are you trying to insert a youtube video? If so change the https to http in the embedded link.

Why isn't this new video working?

I don't see an attempt at a link in the post above...where did you try to put it?

 

[Dr_Emmett_Brown]

 

[EarthDog]

Thanks for the vid.

One thing worth mentioning is that out of the box, all cores and threads run 4.7 GHz. It's a 300 MHz overclock for all core speeds. 3.6 GHz is the base speed which it rarely runs/typically only when throttled.

A couple of things i noticed in the video..

1. If your aida test was running at the beginning... there was almost no CPU use and there it shows thermal throttling already at that 'idle' load (or perhaps that is previous run remenents?). When I fire up AIDA64 stress test using your parameters (CPU and FPU only), my CPU use goes to 100%. What's going on in the vid that yours is at idle?
2. When you start Cinebench, you're thermally throttling almost immediately according to AIDA64.

You may want to have Coretemp or Realtemp running to confirm temperatures... maybe have Intel XTU running as well. At the bottom it will display throttling reasons, be it current, thermal, etc. But if AIDA is reading it throttling, chances are it is and you are at 100C.

 

[Dr_Emmett_Brown]

Thanks for the vid.

One thing worth mentioning is that out of the box, all cores and threads run 4.7 GHz. It's a 300 MHz overclock for all core speeds. 3.6 GHz is the base speed which it rarely runs/typically only when throttled.

This chip is 3.6 GHz out of the box. It's slow. It's not 4.7 GHz. I ran it without any cooling solution previously and it was fine.

1. If your aida test was running at the beginning... there was almost no CPU use and there it shows thermal throttling already at that 'idle' load (or perhaps that is previous run remenents?). When I fire up AIDA64 stress test using your parameters (CPU and FPU only), my CPU use goes to 100%. What's going on in the vid that yours is at idle?

Cinebench was run previously with the fans off. I was testing the condenser. With no fans and 5.0 GHz, it throttled back. Probably did not settle down when restarted.

2. When you start Cinebench, you're thermally throttling almost immediately according to AIDA64.

Yeah the VERY STRESSFUL test, the FPU test, was still on. I don't recommend anyone run FPU by itself. It really shoots the temps sky high.

 

[EarthDog]

4.7 GHz is the is the all core boost speed. 3.6 GHz is base speed as you mention at the end of the vid. So long as temps/mobo and such don't get in the way, it will run 4.7 GHz all c/t and 5 GHz 2t boost on on avx loads. If it's running at 3.6 GHz under heavy load, something isn't right (throttling or a board setting).

https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9900k-i7-9700k-i5-9600k-review

It was weird in the first part of the video that AIDA was running but there wasn't a load. Was it really running?

But yeah, you're throttling in that video from temperature. That cooling system isn't keeping the CPU under 100C it seems.


EDIT: Doc, I had a chance to watch the video a bit more closely. For certain, you were not running the stress test. I can tell by the temperatures that are just above ambient/typical idle, AIDA64 displays typical idle loads (low single digits), and last when you start a stress test in AIDA64, the elapsed time should be counting up - it isn't. You said you "stopped" the A64 stress test, but how did you do that when it wasn't running in the first place? You then point at the CPU use like you did something and it changed (nothing changed), lol. Hit start and watch the cpu load and temperatures skyrocket.

~3:20 you mention that temperatures are 'not too high'... but AIDA64 is showing two cores reaching 100% and throttling (cleary see the temps at 3:10). You can easily see the throttling in the CPU use area as well (red line). In fact, in this test you peaked at 20% throttling. I can cool an i9-10900K (10c/20t) at 5.2 GHz with 3x120mm cooler and 7960x with the same size rad (two different systems).

~6:00 in, you say that exact setup (meaning condenser size) cools an entire rack in a data center, yet it can't cool ~200W from a single i9-9900K and that idle GPU? What did I miss here? Does that magically scale with exponentially more wattage?

~6:20, you talk about the drop in cpu use with drop in temps. That is normal with air and water too.


Oddities of the video aside, how do you get the CPU use and temps to match 1:1? I can't seem to make that happen. I tried in preferences to set it to 1s, but then the temperatures stopped reporting altogether. I have screenshots showing what I mentioned above but want to look as pretty as yours and match the timing.

Last thing... with temperatures a whopping 15C different between cores, I'm betting that block isn't making great contact with the CPU or bad paste job (or bad solder job by intel).

 

[godevskii]

It was weird in the first part of the video that AIDA was running but there wasn't a load. Was it really running?

EDIT: Doc, I had a chance to watch the video a bit more closely. For certain, you were not running the stress test. I can tell by the temperatures that are just above ambient/typical idle, AIDA64 displays typical idle loads (low single digits), and last when you start a stress test in AIDA64, the elapsed time should be counting up - it isn't. You said you "stopped" the A64 stress test, but how did you do that when it wasn't running in the first place?

I got exactly the same impression.

Doc your setup is nice and cool to look at and make some experiments, i wont dispute that, but for some reason its not performing very impressive at all, in fact many custom water loops here in the forums outperform your solution.

 

[Dr_Emmett_Brown]

4.7 GHz is the is the all core boost speed. 3.6 GHz is base speed

KF

NOT K

NOT KS

KF

At about 1:07 into the video, if you squint, you can see it.

https://ark.intel.com/content/www/u...900kf-processor-16m-cache-up-to-5-00-ghz.html

3.6 GHz with a 5.0 GHz "boost."

I was running it at 5.0 GHz.

That's 1.4 GHz overclock.

 

[EarthDog]

KF

NOT K

NOT KS

KF

At about 1:07 into the video, if you squint, you can see it.

https://ark.intel.com/content/www/u...900kf-processor-16m-cache-up-to-5-00-ghz.html

3.6 GHz with a 5.0 GHz "boost."

I was running it at 5.0 GHz.

That's 1.4 GHz overclock.

Indeed. The difference between the K and KF is the latter doesn't have an integrated gpu, but all base and boost speeds are the same. It is 1.4 Ghz over the base clock, 300 MHz over the all core boost clock. 0 MHz over single core boost. The KS is 4GHz base and a flat out 5 GHz boost on all cores and threads. Intel stopped listing all core boost on the ark a couple of generations ago.

https://www.guru3d.com/news-story/i...close-multi-core-turbo-boost-frequencies.html
https://www.anandtech.com/show/14979/the-intel-core-i9-9900ks-review

Anyway, what happened that the system couldn't handle your 9900KF?

Any help on getting aida to match up like you did?

...or you going to correct a useless detail and nothing else?

 

[Dr_Emmett_Brown]

Indeed. The difference between the K and KF is the latter doesn't have an integrated gpu, but all base and boost speeds are the same. It is 1.4 Ghz over the base clock, 300 MHz over the all core boost clock. 0 MHz over single core boost. The KS is 4GHz base and a flat out 5 GHz boost on all cores and threads.

...or you going to correct a useless detail and nothing else?

The chip runs at 3.6 GHz out of the box. I have no idea why you are saying otherwise.

It clearly shows that on the box, on CPUz, and on HWMonitor.

Therefore the overclock was 1.4 GHz. The BIOS settings were such, not some "300 MHz overclock" which somebody mentioned.

The overclock metric is not the difference from the acclaimed boost speed. It's the difference between the base clock speed and the final stable overclock 24x7.

Fortunately, I did 3 takes of the video that day. I combed over the one I posted and found no immediate answers to the heat removal conundrum. However, the second video I shot that day solved the mystery.

1. The copper-finned condenser was NOT in place. It was aluminum.

View attachment 211302

2. Multiple instances of AIDA were running, or had been running. When you run AIDA with "FPU stress test" only, it spikes the temps. Severely.

AIDA has this odd feature in that "stopping" the process doesn't always stop it. Closing it down, for example, doesn't always shut off the run. Or at least that's the behavior I am observing on this setup.

The CPU utilization does not drop to 0% as you might expect once all other applications are deadened.

 

[EarthDog]

The chip runs at 3.6 GHz out of the box. I have no idea why you are saying otherwise.

It clearly shows that on the box, on CPUz, and on HWMonitor.

Therefore the overclock was 1.4 GHz. The BIOS settings were such, not some "300 MHz overclock" which somebody mentioned.

I mentioned it. I also linked you to how these CPUs work. You seem to prefer to base your overclock on the base clock, while most who do this base it on the all-core boost clocks. To each their own. I run AIDA64 stress test (CPU/FPU/Cache) at 4.7 GHz with 9900K right out of the box. Not sure what your issue is and why say its running at 3.6 GHz.

What motherboard are you using again (a signature with your hardware would be helpful. As was said, these CPUs run on boost unless a parameter gets in the way (be it power/current/temperature/motherboard limits). I suggest you pull up Intel XTU while running AIDA64 and see if there are other reasons it is throttling besides temperatures (current limit/power limit - imposed by the board) as it shouldn't be running at 3.6 GHz. So, yes, you overclocked 1.4 GHz from base clocks, 300 MHz for the DEFAULT/OUT OF THE BOX ALL CORE BOOST (which Intel does not list anymore in Ark), and 0 MHz for single/dual.

1. The copper-finned condenser was NOT in place. It was aluminum.

That may explain why it was throttling on a 9900KF... but I highly doubt changing the same size condenser to all copper will suddenly allow it handle a full rack of servers or 'quadruple' radiators as was claimed.

2. Multiple instances of AIDA were running, or had been running. When you run AIDA with "FPU stress test" only, it spikes the temps. Severely.

Yes, when you check off FPU only, it really works the CPU over (but not sure what that has to do with anything asked of you, lol). In your video it was abundantly obvious that no stress test was running, nor did you shut anything off when you said. Literally nothing changed, no window came up, you didn't do anything but (what I think) is realize it isn't running and continued on by it.

I'm not so sure you have an understanding how AIDA stress test works. When you bring up the application and the stress test, it will always bring up that monitoring window. That does NOT mean the test started or any stress test is running, though it does mean slightly increased idle CPU use (to run the monitoring). You have to hit the START button to start, and stop the test. Once it stops, utilization has always gone back to idle (whatever that may be on your PC).

I've never closed it without stopping it, so no idea... but AIDA sitting idle isn't your issue here.

I don't expect CPU utilization to drop to zero as the monitoring is running along with whatever other background processes the PC has. I've stripped OS's so bare for competative benchmarking you almost wouldn't know what the hell it was... and even then, there are still CPU cycles being used.


I'll ask a third time........lol.... how did you get AIDA to be in time like that? By default, I think the ping on sensors is 5s. When I change it to one to match the sensors, it doesn't seem to 'apply'. Is there something else you tweaked?

 

[Dr_Emmett_Brown]

That may explain why it was throttling on a 9900KF... but I highly doubt changing the same size condenser to all copper will suddenly allow it handle a full rack of servers or 'quadruple' radiators as was claimed.

If you would STOP jumping to conclusions, that would be great.

A copper condenser has about 1.7 times the heat removal capacity as an aluminum one OF THE SAME SIZE.

But as you are not in my lab, you have no chance at being able to make that call.

They were not the same exact size. Close in appearance at a glance from the exterior.

The copper condenser has about 50% more fin surface area than then aluminum condenser, with means it has 1.7 x 1.5 = 2.55 times the overall heat removal potential.

When viewing them from outside of the cages, they look the same.

The copper's cage is a little longer and a great deal thicker, usually in a push-pull configuration and not just a push configuration.

So 2.55 x the demo's cooling capacity with 8 fans and not just 2 fans makes for an easy 1 KW cooling solution.

The full rack cooler has a similar condenser but is 3-STAGE cooling, not 2-stage cooling.

A liquid cools the liquid-vapor, which is one stage.
The refrigerant undergoes phase change and cools the copper block, which is stage 2.
The vapor rises to the condenser, which is air cooled in stage 3.

Again, this was a RIG setup to SHOW what a thermosyphon looks like.

I was answering a question sent to me, not answering you.

I should have taken a closer look at it when I was in the lab. We swap stuff all the time. It's an awesome job. Paid to make awesome cooling solutions and play around all we want. But take a long lunch one day, and half your rig could end up on somebody else's bench.

I'll redo a video showing how it removes more heat when configured for max heat removal.

Not one where everyone can easily see half the cooling surface isn't being hit by the fans.

 

[EarthDog]

Good info.

I did jump to conclusions a bit. I assumed when you said server rack, you were talking data center or a more typical 42U rack. With that, wattage densities can reach 10K+ watts (loaded with blades) per. Clearly I wasn't thinking this solution would work that on that kind of scale, but a couple of thousand watts is EASY in a rack. Hell 1KW is nothing and can be just a couple of boxes...that is where my head was at (1KW-3KW).

I should have asked before assuming. Further, these details would be good to hear at the start so readers won't have much room to jump any conclusions. Please understand where we are coming from (no context mentions of cooling an entire server rack in the vid and better than quadruple radiators - however you define that).

Personally, I was excited to finally see video of it in action. I think we were all surprised to see it perform as it did, especially not knowing you were talking to someone else. Regardless, with all the talk I figured you'd put its best foor forward. The errors in the narration/describing what was going on just threw me off even further. I can't wait to see the new video and results as promised. Now go heist your peer's parts!



EDIT: Any help on the AIDA question I had?

how did you get AIDA to be in time like that? By default, I think the ping on sensors is 5s. When I change it to one to match the sensors, it doesn't seem to 'apply'. Is there something else you tweaked?

 

[MaddMutt]

Remember -> If you need addition cooling -> Delta is always heard above all others as a Overclockers best friend

 

[Dr_Emmett_Brown]

This was harder to debug than I thought. I was initially perplexed because the "test runs" I did prior to the video featured no CPU throttling at all. Then I was mad, mostly at myself, for missing this. Sorry if I took it out on any of you with my replies on here.

Precision. That was my first issue. No two runs of the dang thing were the same. So I went the full-tilt lab approach of extreme quality control. Every environment variable was recorded. Erecting an isobaric chamber finally identified the culprit.

A 6 parts per million vapor leak in one of the quick disconnects.

Each subsequent run it got a little hotter.

The leak was slow enough that one could not observe falling fluid levels by inspection when the system was shut down and 100% of the tubes were filled with the liquid form.

So that was it. Recharging the refrigerant after replacing the defective quick-disconnect cured the thrtottling.

Thanks to those who pointed it out to me.

 

[EarthDog]

So, it wasn't the condensor like you thought earlier?

I'd love to see it on video...we're left salivating over lofty claims and have yet to see it able to cool a stock 9900k without throttling. Let's get the train back on the tracks and see it cool that 9900k...we'll worry about server racks and it doing a better job than quad rads after the first step.

 

[Dr_Emmett_Brown]

So, it wasn't the condensor like you thought earlier?

Well having the larger copper condenser swiped definitely was a factor. But that still should have generated consistent results of lower performance. But after 5 runs it was even hotter. 10 runs it was almost too hot. 15 runs bordering on thermal shutdown.

The process of swapping condensers also involved changing the tubing and quick disconnects, which led to what I observed thereafter.

There will be more videos as time marches on, but the 9900KS lineup is being replaced by the 10900F which has 10 cores but is only 95W TDP. Supposedly it can hit 5.2 GHz without extraordinary measures. We'll see soon enough.

If anyone has a "burning hot chip" recommendation let me know. We're getting some new condenser samples later and I could always test a real firebrand on one.

 

[EarthDog]

Since you have the 9900KS, why not start there. When you get a 10900F (65W processor - i9-10900k is 95W 10c/20t), let's see that too. But what fun is that at 65W. I think you want in i9-10900K at 95W (and, like the 9900K(S) will easily eclipse that. )

I can hit 5.2 GHz on all 10900K's (all c/t, no avx offset) I've tested so far (4). That was with a 3x120mm AIO running AIDA (CPU/FPU/Cache) around 1.35V.


We just want to see it do any of what you said it does. While it is awesome to see this project, I'd like to see the money shot at this point!


We've been over the 'hot chip' recommendations before.... I don't think anything has changed since then.
https://www.overclockers.com/forums/showthread.php/796726-Hottest-CPU-Temperatures-Ever

 

[Dr_Emmett_Brown]

Well "the boss" wants us to look at the 10900F. Supposedly it's fsster than a 10900K by some small percentage. It will be our new overclocked entry level product.

Did you ever CineBench your 5.2 GHz rig? I'm guessing 6500 for an R20 score. I hope it's higher.

We don't have a 9900KS in shop anymore and I'd have to buy one. If I knew someone would purchase the whole system, thermosyphon and all, I'd build it live streaming and ship it when done.

 

[EarthDog]

I dont see how a 10900F is at all faster than a 10900k. Its slower by clocks all around (base, all core and dual core boosts), 65W chip vs 95W, and certainly it isnt faster per clock since its the same architecture (isnt it locked, too?). Not quite sure where "The boss" heard that information.

5.2 ghz is around that mark in cinebench.... why do you 'hope its higher'?

Someone may purchase it, however a proof of concept would be a good place to start. So far we've seen this unable to cool any cpu and you repeatedly dance around showing it working (since you joined?).

 

[Dr_Emmett_Brown]

I'm not sure why we're being asked to look at that chip (10900F) so I googled around and found this:

View attachment 211386

It looks like it gets the #1 spot on the PassMark single CPU core speed test.

My guess at the 6500 score was based on a linear extrapolation of the 9900KS architecture speed. A 5.0 GHz 9900KS gets a 5000 score. At 5.1 GHz it gets a 5100 etc. So I was hoping a 5.2 GHz 10-core score would be higher than 10/8 = 1.25 times as fast. That would mean the 10900 architecture is better per clock than the 9900KS.

We learn in Back to the Future III that "The Doc can dance," but the sentiment here does not apply. Let me show you something.

View attachment 211389

That's my arm with a typical copper condenser.

As stated before, copper convects about 1.7 times as much heat as aluminum. An aluminum condenser would need to be 1.7 times that size to do the same job. And condensers remove more heat than radiators. So imagine a radiator twice that size that also had a solid copper water block.

That's the kind of heat load the thermosyphon removes.