Log Playing with the team red

Kubunteando

Efficiency Noob
Feb 16, 2021
5
1
Good to see things worked out.
Very interesting this info about GamerOS, maybe it should be promoted more between the Linux community...

What drivers are you using for the graphic card? The Radeon open source that come with the Linux kernel?
 

Elaman

Cable-Tie Ninja
Original poster
Sep 13, 2020
155
119
Good to see things worked out.
Very interesting this info about GamerOS, maybe it should be promoted more between the Linux community...

What drivers are you using for the graphic card? The Radeon open source that come with the Linux kernel?

If memory serves well, GamerOS doesn't ask which drivers to use and dmesg at the time revealed the open-source (AMDGPU) drivers. As GamerOS is based on Arch, that is a safe assumption especially if we read what the [arch wiki] has to say about the proprietary drivers (AMDGPU PRO).
During last 5 years I have had dedicated (AMD) graphics in a laptop and I have alternated both proprietary and open-source, in my limited experience the difference was not noticeable. So I'd say it doesn't matter and I would even lean towards AMDGPU as default for ease of use.

But it gets better.



As you can see, the chosen driver is not making as big of a difference as the kernel and/or Mesa versions ([reference]). But look at what happens when we upgrade the kernel instead (violet line). And that was for 5.9; GamerOS at the time of this writing is using 5.11.

If we think really long term, we can't predict the future but we have seen driver updates in 2020 still improved performance significantly for RX580, 570 and the likes. These cards are aging like fine wine:



Launch date vs. January driver update (three months time). I mean... that jump in 6800 alone, plus taking into account that is the mean of results.
 

Elaman

Cable-Tie Ninja
Original poster
Sep 13, 2020
155
119

Back to the back to the kitchen​


As a gamer



with a gaming motherboard sitting in my gamer chair gaming a game in [game mode] in front of my gaming monitor using my gaming controller, I have spent a total of 65 minutes playing in this machine so far according to Steam.
In the same period time, since I started this build I have taken out the cooler from my CPU three times, cut down and rearranged my water circuit and spent three tubes of thermal paste. That tells you something I guess.
This is how my third time went.

Back to (school) Alphacool​



Now I have received some more extra fittings and adapters, 90º and 45º, which I can play with. Also got me a small tube cutter, which has made my life so much easier by orders of magnitude. This time I emptied and dried the loop, and screwed everything in place. Then I proceeded to add sections of tube step by step. The various articulated fittings give me more headroom as well.




It took a while but this time I got the correct tube sections for everything. The final result seems more sturdy and none of the tube sections is pulling or tense:






As I cannot do anything the normal way, I decided once again I was going to be smarter than everyone else and employ liquid metal coolant, or as I've come to call it:

Infernal drop from hell​




With the purpose of improving the cooling, while trying to follow on the instructions video I learned that this is a permanent solution. Meaning that, once the liquid metal heats, it cannot be removed like the thermal paste.

Immediately I decide to take the drop away.

Immediately also I learn that trying to catch this drop is like trying to catch mercury: it will not stick to paper, it will roll around the house and on the floor. You can barely catch it with two of those little plastic spoons that come to apply paste, but it slips rather easily.

Only two powerful weapons can defeat the infernal drop from hell. One is the syringe with which it was applied in the first place (barely visible in the photo):



The other is some special cleaning towels that come with the package, can be quite the lifesaver.
Now back to good-old thermal paste and abstract painting.



Overall temperatures are much better right from the start. Immediately after powering on I see roughly 50º (CPU) in the motherboard sensors while previously it has been ~60º on boot.



Nothing to do with the tubes, probably just the fact that I paid more attention to the Eisbaer-CPU contact this time. I cleaned up both surfaces prior, and used slightly less thermal paste this time. Also spread the paste with the spoon so that it would be uniformly distributed.

I have to look more in detail into cooling, but all these additional fittings meaning more metal in the loop, makes me think intuitively, that there should be more chances for heat to be dissipated. Is that so?

But more importantly, this means that tubes now are within the volume of the CPU compartment, and now we can take a look at how the power button is screwed in the case...



... and put on the cover ...




... and CASE CLOSED (I wanted to make that joke so much). And now, we can have the customary console-like photo:



Ta-daaa!!!

After waiting for a while for the loop to get hot, I can notice that cold air is coming upwards from the radiator and hot air is blowing towards the motherboard. Now I don't see very alarming CPU temperatures from the sensors, but switching the fan orientation is the kind of beginner mistake which has a rather easy solution.



The cherry on top would have been adding yet another fan on top of the radiator.
As for "real" problems, now that let's say "warmer" air is coming from the radiator to the outside, I don't find it quite hot enough to justify any additional cooling. I guess I have to get to proper testing before I get to any conclusion.
 
  • Like
Reactions: dealda

Elaman

Cable-Tie Ninja
Original poster
Sep 13, 2020
155
119

The benchmarkening​


With a better rearrangement of my cooling and more reasonable CPU temperatures, I became confident enough to start putting my new PC to the test.
For this purpose I will install two new OS'es. My original plan was to go with Intel's Clear Linux which seems to be a [benchmark powerhouse]. However the installer [doesn't come up in my PC] so sorry Intel, make it easier next time.

The test setups​


As there is no shortage of good options anyway, I am going with two I have never tried: one of them is [SalientOS] which is supposed to be among the best performers and I must say, I am liking a lot so far.

The other will be [Garuda]. Garuda is kind of the new hotness and comes with the possibility of a performance-oriented kernel out of the box. As an experienced user, I don't quite buy it and I plan to demistify this meme with actual testing (or maybe find the new hotness?).

Both are extremely easy to install and come with some kind of Steam. My plan is not to tinker with any of them as to show out-of-the-box performance.
Besides this, and since I (have not forgotten) am here to test the BIG1 case, I am going to perform tests with and without the case.



This GPU looks big but it's quite light, so I can get away with this.



Finally, I am going to also test the CPU in normal and ECO mode.

Expectations are of course, that temperatures will be different. Maybe it is not necessary to test without case and eco mode, but still this leaves me with 6 combinations for all my tests.

Also as a Linux user, my idea of testing is leaving my PC to work for 5 hours and go for a walk to the park. Last thing I want to do is babysitting my PC while it draws the exact same horse in 3D for the umpteenth time. All tests must be automated.



This gives me 8 possible combinations. Probably 5 or 6 will be enough though. Switching to/from open-air is rather annoying. For example if one configuration gives me lower CPU temperatures, it is obvious that I can get them lower yet by switching to eco mode.

The tests themselves​


For games, I am only testing Dota2. Counter-Strike runs on a potato, also the games I found automated in [Openbenchmarking] are quite uninteresting, Dota happens to be free and #2 most played game on Steam.
In SalientOS/Garuda as with the previous, I can configure the compositor to give me up to 240Hz, but this seems more than enough for a normal game:



And definitely more than enough FPS for the human eye, don't care what nVidia says.

For graphics I will go for the 4 unigines (sanctuary, tropics, heaven, superposition). They are not representative of a practical GPU scenario



but oh boy, are they good at putting the GPU power and temperatures to the limit...

For loading the CPU specifically I get to the kind of workloads that matter to me the most: gnupg for encryption, build-gcc for software building, glibc-bench, and for video encoding: ffmpeg, dav1d, libgav1, rav1e.
As for what the vast majority of our PC's around the world are actually doing most of the time, I chose a bunch of benchmarks for Java (sunflow, bork, java-scimark2, dacapobench) and a full suite for Python.

The observations​


- The biggest CPU burner turned out to be build-gcc:



This maximum of 90.75 in my sensor is consistent not only with the other non-eco mode runs, but also the max CPU temperature of all the test runs of all benches. This is probably the point at which Ryzen architecture puts the "stop" sign. Definitely not building GCC anytime soon unless I am on eco-mode.

- The ECO mode is [absolutely hilarious]. Sure, the "normal" mode gets slightly more performance out of the CPU here and there, the biggest difference being of course in build-gcc. But if best I can do is save 20 seconds out of 800, excuse me but I'll stay in eco-mode.

- The biggest GPU burner is clearly Unigine-superposition. It draws the most amount of power, maxing at 260 Watts except for 250W in open-air case. Which leads me to the next point.

- GPU temperatures were indeed better in open air. Which explains slightly less need for cooling which explains slightly less power draw. This happened across all Unigine tests, but best shown by Unigine-superposition:



On the CPU temperature side, still better temperatures with open case, but small difference: maybe 2 less degrees average.

- NVMe drive temperatures are always higher in Garuda across all tests. Spectacularly so, in certain cases: I got to see a +15ºC difference on averages. Garuda is using a copy-on-write filesystem called BTRFS which provides really good features such as snapshots and rollback to a working state. I had no idea about the higher temperatures though, but it makes sense to me. Anyway, this kind of filesystems are trending lately and price of storage is getting higher. Not to bash BTRFS, but something for people to take into account.

- Garuda is legit good, and wins in some cases. However, the performance improvement in DotA2 (~122 to ~133 fps) specifically in Vulkan led me to run a [dedicated Vulkan-versus-OpenGL bench], and there I saw the reality:


If anybody is even more data-hungry and graph-loving, here is a [summary comparison] including most of the test data.

The conclusions​


What can I say, I have seen many builds with big radiators while preparing mine. I did not really expect that a 80mm radiator could cool a Ryzen 5800 but took the leap of faith. Seems that @REVOCCASES was pretty on-point with his measurements.



I am obviously very impressed with this GPU and I have plans for it. On the one hand there weren't automated benchmarks for games that I'm interested in, and I don't feel like buying a game that I'm not interested in, only for a benchmark, if that makes sense.

Regarding Linux and new and flashy distributions, I wanted to be able to say certain things that I have now said, but with results backing them.

Plenty fun was had, guys.
 

Elaman

Cable-Tie Ninja
Original poster
Sep 13, 2020
155
119
Some nice person out there with roughly my CPU/GPU combo, had a thought about [the few people who still care about Cyberpunk].


The comparison is made between Windows 10 and Ubuntu.


I could try to include Ubuntu 21.04 in my performance comparisons, but I worry about how it could look.
In general both Windows10 and Ubuntu are operating systems targeted to be compatible with as many vendors as possible (much like Android is) so there is not a focus on performance and I don't know if it's a fair comparison.
 

Elaman

Cable-Tie Ninja
Original poster
Sep 13, 2020
155
119

Why drive temperature matters​


When talking about new hardware and drivers, we want the latest and greatest. Especially gamers, we want to always have as soon as possible the latest updates from Steam, GPU drivers (continuius improvement in performance), your gaming mice etc.

Latest and greatest is the enemy of stable operating systems. An operating system needs to integrate all these different elements and make sure they play nice with each other. So-called "Rolling releases" which get constant updates that are still in "testing" are famous for breaking often.

Garuda is one of so-called bleeding-edge, rolling release and their solution to the problem, now I see, is the copy-on-write filesystem: With each major update, you have your snapshot and this "super-power" of going back in time to when your computer still worked.

Now with more experience, it is clear to me that copy-on-write is the way to go. Sometimes Steam breaks, sometimes Vulkan, etc. It should not be a problem.
The question is, what can I do about those sometimes 15º extra on the SSD temperature? I have to find a way to improve ventilation in the case so that this filesystem is manageable.
 

Elaman

Cable-Tie Ninja
Original poster
Sep 13, 2020
155
119

The dirty prototypening​


Getting better temperatures in the SSD comes from getting better overall temperatures in the motherboard, and overall a better wind-tunnel thing going on in the two halves of the case: the GPU half and the motherboard/CPU half.

The first thing I can think of is that the air might be escaping through some other places and not the fan.
To prove my thesis, I only need to do a simple experiment:



All the small areas surrounding the fan are now covered by leftover foam that I had from one of the packages. Again, not a permanent solution but just to test my idea.

Now, let me run the exact same benchmark sequences as before:


[reference]

I have now (brown) scrapped away 2 degrees from what it seemed to be the maximum temperature (this is Unigine Superposition at Ultra settings, exact same test sequence.). Open case (pink-ish below) is still the winner, but I'm starting to see the light.

If I look at my earlier Vulkan(vs OpenGL) benchmark, which is more realistic scenario for an everyday GPU usage:

[reference]

The max is understandably higher, but average temperatures are almost on-par with open case, which is encouraging.

This is all I needed to see.
 

Elaman

Cable-Tie Ninja
Original poster
Sep 13, 2020
155
119

The optimizening​

- I can get a bit better temperature by covering the roof with foam.
- ...so how about covering the whole roof with fans?
- This is not a good idea, and it contradicts the whole point of why we are doing water-cooling in the first place. The idea is to be silent, in that sense fans are counter-productive and inefficient. Blah blah blah (...)
- ... but how about covering the whole roof with fans?
Right?



Right? How about that? MOAR FANS BETTER! AMIRIGHT?

From my measurements, I should have room for two fans of exactly 92mm, and really tight fit. I somehow had the expectation that maybe I would need some sanding here and there.
However, I take the first fan out of its package, and it fits in its corner

almost to measure:


I try to figure out how to work with what I have. Luckily the fit to the case cover and to the bracket below is so tight, that I will not need much help getting it well fixed.



I will tie the fans together with a ziptie so that it is still possible to get them in & out:



A third fan can land right on top of the radiator. My common sense tells me that this should help with CPU cooling somewhat. My common sense is about to be proven wrong.



The red areas show parts where the cables don't have an inlet and will be pressed between the fan's plastic and the case cover or chassis.

Therefore some of the fans will be



The black duct tape protects the cable cable and goes with the overall color somewhat.



The case does not support a 92mm fan on top of the radiator like this, but for my testing I can get it tightly fixed to the bracket below using zipties in strategic places:


And slowly, I become an adept in the art of zip-ties. This one ends in the space between fans, so it's not an obstacle for the grid (above) nor the chassis (below).



The duct tape can help explain to the air that it should go this way and only this way:


And there it is.
Look at it. LOOK AT IT.



It's the **Black Chimney**.
Look at it here, breaking my phone camera with its raw FPS power.

Countless frames per second are being burned away in its infernal furnace.


OK enough goofing (for now). I got four fans, but there is a possibility that you can get similar resuls with a three-fan setup:



again with the help of

And here is a four-fan setup:


This look is more aesthetically pleasing to me, because of the uniformity of black color and the symmetry of the four fans put together.
I will run tests for 3 and 4-fan setups, just to be methodical about it. I would rather keep the 4-fan setup for the looks, unless the 3-fan setup performs better for some weird reason.

Results​


The improvement in hard drive temperatures is nothing short of spectacular.



In the compilation test where we used to top close to 50 degrees, now Salient refuses to reach 31º. In fact Garuda with its busy filesystem will reach max. 35º.



Overall all SSD temperatures are down, be it min, max or average, across all tests.

Surprisingly (but very nice surprise) CPU temperatures also went down. If I choose again the most extreme case of CPU burn:



Especially on Salient, average temperature is a bit more reasonable and it seems to scrap some decimals away from the max. temperature. I'd still keep using ECO mode, though.
It is worth noting that in terms of performance, this particular run with Salient OS broke more records than any of the others.
So when CPU temperature is lower, the governor seems to be loading more work on to the CPU which brings more performance but temperature doesn't lower that much.
Overall CPU temperature went down in general a good 3 to 5 degrees, nevertheless.
[reference]

Finally for GPU temperatures I will also start the most extreme case which was Unigine Superposition on Ultra settings:



I get for the first time results that are more or less on-par with open case, performance same or better, temperature on Salient was actually better than open case.

We see this most clearly if we forget about the extreme Unigine temperatures and look at my compilation Vulkan benchmarks, showing more day-to-day use case of the GPU, the trend is obvious:


- (Cyan) Closed case with foam, worst case.
- (Lilac and red) Open case, better temperatures but not quite.
- (Green and pink-ish red) 4 fans, lowest temperatures all along.
[reference]

Conclusions

GPU turned out to be the most important part of the story: I have achieved a lower GPU temperature across all tests, min&max&average, compared to open-air. Which gives me an excuse to use the case. Because its little wind tunnel provides now better cooling than open air.

The PWM fans in the GPU compartment are guided by the motherboard temperature. This has probably been on-point regarding the SSD, but is not that useful for what is basically a GPU enclosure. The GPU itself doesn't have a direct PWM connector. Well, it sort of has this one:



Which goes to the GPU fans:



But is clearly a different type of connector than that of the case fan connectors:



It would be great to send out the information about RPMs that is coming from that green cable, directly to the Noctuas, however I can't think of a non-extremely-hacky way of doing that.

If I learned something from this forum, is that a build is never "finished". But I wonder if this is as far as I can go in this game, with the pieces I have to play with.
 

BaK

King of Cable Management
Bronze Supporter
May 17, 2016
930
931
Was fun and interesting to read! :thumb:
Great to see the wind tunnel is efficient!
But you don't tell us how is the noise with the added top fans?

As you do, I also like the look of the 92mm four-fans setup, I am however wondering what temps or noise you would get with a single 180mm like a Phobya G-Silent or a Silverstone Air Penetrator.
 

Elaman

Cable-Tie Ninja
Original poster
Sep 13, 2020
155
119
Great to see the wind tunnel is efficient!
I am learning as I go, but I wonder if fans below the case (especially in the GPU half) would help further with the wind tunnel?

But you don't tell us how is the noise with the added top fans?
My opinion on noise is to be taken with a grain of salt, because my laptop makes more noise (yes) and I am kind of in awe of the miracle of engineering that are the Noctuas.
I think in summary, the setup is only noticeably noisier when on the CPU is on high load (again: grain of salt). Also note, the radiator fan is even smaller (80mm) so the 92mm are not noisier than that.
If we go further and turn on the fans on the brick, which are very small fans, that noise will definitely "override" any other fan's noise (again the smaller the fan the louder).
And finally the two fans on the GPU half are very silent because they are "case fans". I'd actually like them to adjust to the GPU temperature, and then they would be more noisy, but if you think about it, that's the real deal. Sometimes the GPU fans are not even spinning, they only spin when required.

As you do, I also like the look of the 92mm four-fans setup, I am however wondering what temps or noise you would get with a single 180mm like a Phobya G-Silent or a Silverstone Air Penetrator.
100% and it would be more power-efficient also. And one would use a bracket.
In this case I am kind of using the fans (previously foam) to cover empty space, so they are right on top of the chassis, the CPU side they are almost touching the radiator save for those extra 3mm. This case is divided into two compartments, so I'd have to split the 180mm in two halves.
I was actually thinking to make a bracket for 180mm fans. Similar like the RAW S1 "supports" a top 140mm fan. But there wasn't much interest.

View attachment 1057
This reminds me of the MID1 but on vertical. I was not planning on adding fans, otherwise I think I'd go for the MID but then I started building, and one thing led to another...
 
  • Like
Reactions: BaK and REVOCCASES

Elaman

Cable-Tie Ninja
Original poster
Sep 13, 2020
155
119
As it turns out, as much as I dismissed the "software route" as convoluted, in practice it took me maybe 5 minutes to do this at the OS (Linux) level.
Turns out you just run [fancontrol] and it "takes you by the hand", first running the case fan at different speeds, so that you know which number is "silent", and then gives you a list of temperature sensors to control that fan, and I chose of course the GPU sensor (the GPU die sensor, the hottest one). Anyway this what I was after, it works as if you added two more fans to the graphics card.

And that's really all there is to it.
So it took me 3 tries until I got a really silent idle fan:



But all in all the average GPU temperature is lower in all three attempts.
[Full results]
It is not a fair comparison, because for these last three runs CPU was in ECO mode, but performance is better anyways. Also because I am present the whole time, I can tell that the extra fans are spinning at same rates as the GPU fans.

I don't have 3DMark but I got Ashes of the Singularity, for eh... research purposes, which is DX12, which has its own [benchmark]. I have modified the benchmark so that I can see DX11/DX12 results and I can tell the performance there is better as well.
I have to eh... keep studying of course. This is uh... very important research work.
 
  • Like
Reactions: BaK and REVOCCASES

BaK

King of Cable Management
Bronze Supporter
May 17, 2016
930
931
Thanks for the 'fancontrol' info, I will check it out!
Only done that on Windows so far with Speedfan.