Stresstest a fanless overclocked Pi 3 B (no +)?

thelostsoul · 27 Mar 2018, 17:25

Hello guys. I finally got my new case and did some small overclocking the Pi 3 B (no plus) and managed to stresstest it a little. It is fanless cooling. My setup:

Pi Model or other hardware: Raspberry Pi 3 B
Power Supply used: Official 5.1v 2.5A psu
RetroPie Version Used: 4.3.16
Built From: Pre made SD Image on RetroPie website
USB Devices connected: 2 Buffalo snes gamepads, 1 arcade stick, wireless adapter for keyboard
Guide used: https://core-electronics.com.au/tutorials/stress-testing-your-raspberry-pi.html

The case I am using without a fan: BIQU Aluminium Raspberry Pi Modell B

My overclocking settings are:

arm_freq=1300
core_freq=500
sdram_freq=500
over_voltage=2

From the guide, I used this line

while true; do vcgencmd measure_clock arm; vcgencmd measure_temp; sleep 10; done& stress -c 4 -t 900s

The 15 minutes stresstest with stress application is over and successful. It started with following first 30 seconds:

frequency(45)=1300020000
temp=42.9'C
frequency(45)=1300000000
temp=56.9'C
frequency(45)=1300000000
temp=60.1'C

However, around 4 minutes later or so, it started to throttling? Like this:

frequency(45)=1300000000
temp=70.9'C
frequency(45)=1299998000
temp=69.8'C
frequency(45)=1300002000
temp=70.9'C

I think, it ended with this, If I am not mistaken (my last screenshot is messed up, I should have used a log file):

frequency(45)=1300000000
temp=74.1'C
frequency(45)=1299998000
temp=74.1'C
frequency(45)=1300000000
temp=75.2'C

Would you call this stable? Is throttling happening? My main usage would be playing 8- and 16-bit consoles with shaders at 720p and a little PSX and even less N64 (no priority here). Edit: Btw, I just played Wave Race 64 for about 15 minutes and then after quitting N64 emulator, I exited EmulationStation as fast as possible to check the temperature.
It says CPU: 52°C/125°F GPU: 52°C/125°F, which is quite impressive, isn't it? The game was playing nice without hiccups or slowdowns. Why is there so much difference between realplay and stresstest? What is the matter of stresstesting then?

quicksilver · 27 Mar 2018, 19:06

@thelostsoul
The stress test that you performed is not enough to confirm stability. And no, your CPU was not throttling otherwise your ARM frequency would drop and your temp would be much higher. But 15 minutes isnt near enough to determine stability. You really will want to see how your overclocked CPU, GPU and RAM handle all being pushed hard at the same time. When I overclocked my pi I ran the same stress test and I passed 15 minutes at 1400mhz. But when put to a real world test it failed miserably. I was even able to get memtester to give me a passing grade with my RAM set to 600mhz. But this was not actually stable either.

I found that Quake 3 is a much better REAL WORLD test of stability. It will push the CPU to the max and also work the GPU and RAM very hard. Bad over clocks will often fail within a few minutes although occasionally will take a few hours. Here is a little test you can perform to test your overclock:

Install Quake3 from optional packages. The freeware version of quake 3 should work fine. Easiest way to test is to load a skirmish, load a bunch of bots (8 seems a good number). Set kill count and time limit to 0. Once you get into the game press ~ key to open the console. Type /team spectator

Then press enter. Press ~ again to close the console. You are now spectating. Press the fire button and it will place you in a first person view of one of the bots. Now the game will run until you stop it or it crashes. Make sure you have good cooling because it will get your pi very warm. I would also recommend using SSH to monitor your temps while the game is running. If the game crashes, your overclock was not stable.

Not every game or emulator will work your pi to the max. The reason we stress test is so that when your pi suddenly is pushed to the max, your pi doesnt crash and you dont corrupt your SD card. If your pi has an unstable overclock, everything may be fine for weeks until suddenly you do something that makes your pi work very hard and then it crashes and you lose data. Would you make a modification to your car that made it faster but when you pushed the accelerator 100% your car fell apart? No, of course not that wouldnt be reliable. Overclocking is pointless if it makes your pi unreliable.

thelostsoul · 27 Mar 2018, 20:01

@quicksilver Ok I understand this now, makes sense. Currently, I think I don't have enough space for Quake 3. How much space will it "cost". I will upgrade from 32gb to 64 or 128 soon.

I will lower the overclocking settings until then. But is Quake 3 really a realworld testing? I mean, when will I get full load on cpu, gpu and ram at once if I only play those systems I mentioned. None of them should stress like Quake, or will it?

I just wanted overclock slightly, so I am on safe side. And thank you, for taking your time.

quicksilver · 28 Mar 2018, 13:22

@thelostsoul said in Stresstest a fanless overclocked Pi 3 B (no +)?:

I will lower the overclocking settings until then. But is Quake 3 really a realworld testing?

Yes it is. There are quite a number of things that will stress your pi, building emulators from source, some N64 games, Mame, dreamcast can all work your pi pretty hard. If you run an unstable overclock, even if it runs for days, it may crash when you least expect it. You could corrupt your SD card and lose your data.

I mean, when will I get full load on cpu, gpu and ram at once if I only play those systems I mentioned. None of them should stress like Quake, or will it?

Again, look at my earlier analogy. Would you make a speed modification to your car, knowing that in an emergency it may cause your car to stop responding? If an overclock is unstable then your whole system is unreliable. Feel free to do what you want, it's just a bad idea.

thelostsoul · 28 Mar 2018, 14:43

@quicksilver Please don't get me wrong. I mean, if Quake 3 isn't the only thing stressing the pi, I would like testing it with real realworld stress. You said in example some N64 games will do this. I would like testing it with those things I actually do. Or I build an emulator from source.

What is the most stressing thing I actually do and own on my pi? Hope you know what I mean. I didn't say that I don't want to test. I get yout point.

quicksilver · 28 Mar 2018, 15:35

@thelostsoul said in Stresstest a fanless overclocked Pi 3 B (no +)?:

@quicksilver Please don't get me wrong. I mean, if Quake 3 isn't the only thing stressing the pi, I would like testing it with real realworld stress. You said in example some N64 games will do this. I would like testing it with those things I actually do. Or I build an emulator from source.

What is the most stressing thing I actually do and own on my pi? Hope you know what I mean. I didn't say that I don't want to test. I get yout point.

I feel like I already answered this ;)

You never know what is going to suddenly be too much for an unstable overclock. Quake 3 is an easy testing method to use and if it doesn't crash after running for hours you can be reasonably certain that nothing else on your pi will. Quake 3 is recommended as a testing tool by the raspberry pi foundation.

thelostsoul · 28 Mar 2018, 15:53

@quicksilver i understand about Quake 3.
My question is an alternative testing, which I will use on my pi. I mean if building emulator is the highest load I can do, why shouldnt I use this? Is there nothing else doing same load else Quake 3? My point is, if not, then its not relevant testing tool. I want test tze highest load I actually want to use on my pi.

I respect what you tell me, but I feel you don't understand what I am trying to achieve.

mitu · 28 Mar 2018, 15:56

@thelostsoul said in Stresstest a fanless overclocked Pi 3 B (no +)?:

I mean if building emulator is the highest load I can do, why shouldnt I use this

Compilation only stresses the CPU/disk, but not the GPU.

quicksilver · 28 Mar 2018, 15:57

@thelostsoul said in Stresstest a fanless overclocked Pi 3 B (no +)?:

@quicksilver i understand about Quake 3.
My question is an alternative testing, which I will use on my pi. I mean if building emulator is the highest load I can do, why shouldnt I use this? Is there nothing else doing same load else Quake 3? My point is, if not, then its not relevant testing tool. I want test tze highest load I actually want to use on my pi.

I respect what you tell me, but I feel you don't understand what I am trying to achieve.

Maybe Im not understanding then. Building something from source isnt going to work your GPU. If you want the most thorough testing method I would recommend Quake 3. You can also use the stability tests you have also been using like Stress and Memtest. But Quake 3 really is the best.

thelostsoul · 28 Mar 2018, 16:10

@mitu that is my point. If nothing doing on my pi will stress like Quake 3, then there is no point in doing it.

thelostsoul · 29 Mar 2018, 09:11

@quicksilver I am running Quake 3 like you suggested. Still not a big fan of such high load testings, if I never need it in "realworld". Maybe its debatable, but whatsoever, at least, if it is stable with Quake 3, then everything I do should be stable too (in theory). (Btw, my native language isn't English and I don't want sound like I would want degrade your statements.)

It runs for about 10 minutes now and the case got hot, but still no crash or sign of slowdowns. How long would you suggest to run the test, to call it stable?

Darksavior · 29 Mar 2018, 09:41

@thelostsoul You really have to try something for like an hour or os. You can check temps while you're playing if you ssh to your pi and run the command /opt/vc/bin/vcgencmd measure_temp
Though, those oc settings you're using are the similar to mine and my pi3's stable since day 1. I'm using the example settings from the retropie overclock wiki page.

quicksilver · 29 Mar 2018, 12:48

@thelostsoul it's your pi, you are welcome to do what ever you want with it. The whole point of testing for stability Is to test using the worst case scenario, that way you can be assured that your computer is reliable. What I have tried to say several times to you is that you cant always know when you your pi might be worked hard by something. I think I have said enough on that subject.

So, all that said your overclock settings are pretty mild and most Pi 3s seem to able to hit those speeds and be stable. If it were me I would test for a couple hours just to give me peace of mind. But if 15 minutes is good enough for you then go for it.

thelostsoul · 29 Mar 2018, 14:02

@quicksilver I understand what you wanted tell me. Yesterday I got a new micro sd card with more space, so I could install Quake 3.

I did 15 minute test for first try because I had to go to work and will test as long as you experienced guys would recommend. 1 hour sounds safe.

thelostsoul · 31 Mar 2018, 14:10

I added thermal paste and it is 20 degrees lower now? is this real? Or does the settings I slightly changed does this big difference? In comparison to my first post. I will test this with Quake 3 too, but wanted to share my experience at this place with the previously used stresstest, as I saw a very big difference in temperature to my previous testing.

Quake 3 test did last around 25 minutes (not sure the exact time it freezed) with my old setup, so I decided to add some thermal paste and optimize settings and retest again. First, this short little stresstest.

Stresstest

while true; do vcgencmd measure_clock arm; vcgencmd measure_temp; sleep 10; done& stress -c 4 -t 900s

New setup
temp after 4 minutes: 45°C
temp after 15 minutes: 55°C
with thermal paste
arm_freq=1300
core_freq=500
sdram_freq=500
sdram_schmoo=0x02000020
sdram_over_voltage=2
over_voltage=2
Old setup (first posting)
temp after 4 minutes: 70°C
temp after 15 minutes: 75°C
without thermal paste
arm_freq=1300
core_freq=500
sdram_freq=500
over_voltage=2

thelostsoul · 31 Mar 2018, 15:48

@thelostsoul
Ok, the settings from my last posting aren't stable on Quake 3, so I had to dial them down. Currently I run Quake 3 test for 1h (still running here) and case got hot, but no sign of over heating. Can't read the temps, as I have a problem connecting with SSH, but thats another story. Here my settings, which I may leave as it is now. I am shocked how much difference themal paste do.

Newer setup (1h success on Quake 3 test)
arm_freq=1250
core_freq=450
sdram_freq=450
sdram_schmoo=0x02000020
sdram_over_voltage=2
over_voltage=2

I am not into high overclocking, just want take advantage of better cooling and play psx and only some of the n64 with better experience and this helps me. But I am not sure about the sdram. Quake 3 don't test this right? And is sdram worth it, if I just play games? Yes, loading times of cd games (cd music) from sdcard would be improved I guess. Other than that, does this have an impact somewhere else? How can I be sure its not corrupting my data? Would you suggest dialing it back?

Most newest setup (2h success on Quake 3 test)
arm_freq=1275
core_freq=475
sdram_freq=475
sdram_schmoo=0x02000020
sdram_over_voltage=2
over_voltage=2

5 seconds after closing Quake 3, I can read 52°C for the last 2h test. So this is a huge success here I guess. So, I have not to worry about playing anything for long period of time.

quicksilver · 31 Mar 2018, 20:46

@thelostsoul you could also try increasing your over voltage setting to 3 or 4. This would potentially allow you to increase your overclock a little more and be stable.

thelostsoul · 31 Mar 2018, 21:22

@quicksilver I am not sure what the trade off for voltage is. I know it makes more stable and increase heat. Currently, I am really happy to get this what I have now and I consider to do higher settings. But I don't want fry my board, so I am careful about voltage. It does not use any fan. Now after I am a little more in to this overclocking things, I will have a closer look into the other thread.

quicksilver · 31 Mar 2018, 21:43

@thelostsoul your pi isn't at risk of becoming fried, your pi will throttle itself if it becomes too hot. Even this seems unlikely as you are using a good case for cooling. You can safely increase overvolt to a value of 6 without voiding your warranty. As you stated the only drawback of overvolting (up to 6) is increased heat and potentially shortening your Pi's overall lifespan, though you are much more likely to replace your pi with a newer model before that happens. But if your pi is stable at settings that you are satisfied with then maybe it would be best to leave your settings as they are now.

thelostsoul · 31 Mar 2018, 21:54

@quicksilver I am currently not determined. Didn't know that this would go that good, so I may change my initial mind. In fact, its not about the warranty, but about data loss or about suddenly stop playing any game and wait for a new model to arrive. If there is still room, I will sure do more on this. I mean, why not, I payed for this and if it can handle higher clocks, then I will do it.

Thank you for your patience! It helped me a lot and about Quake 3 test, I understand it now. It is something like "if it works under Quake 3, then you don't have to worry" test. So I will use this in future, alongside Perfect Dark intro and first level test. Would this cover all those 2D platforms too? If these works, then I don't need to test say, snes games right? They have to work good in this case.