Raspberry Pi 4, 8GB DRAM version, as desktop computer

[Cowboy T]

I like that setup, especially the PoE injector.

[Cowboy T]

Been working on cooling solutions for the Pi 4 over the last week. Just finished up today. We're doing four cases, with graphs.

The test is "stressberry-run", exercising all four cores of the CPU. Where indicated, I also exercised the GPU at the same time using the glxgears OpenGL demonstration program. For all cases, I ran tests at the stock clock speeds (CPU is 1.5 GHz, GPU is 500 MHz) and overclocked to 2.0 GHz and 750 MHz, respectively. The test ran for an hour each, to get a good idea of an extended load.

The four cases are as follows.

1.) Official Raspberry Pi 4 case, with fan and CPU heatsink
2.) Argon 1 Neo fanless case
3.) Argon One case with fan
4.) Micro Connectors aluminum case with fan (tested both with and without CPU heatsink)

The first will be the official Raspberry Pi 4 case with the fan+heatsink. The tests were run at 1.5 GHz (stock speed), 1.8 GHz, and 2.0 GHz. Here are those graphs (thumbnails are clickable).



Now, let's compare this to the Argon 1 Neo case, which is the same size, and uses no fan (it's basically a big heatsink). For the Neo, I did 1.5 GHz and 2.0 GHz speeds, so there are only two graphs.



As you can see, the Neo with no fan got very similar results as the official Pi case with the fan. Indeed, as we clock up, we see that the Neo does just slightly better, staying at just barely below the throttling temperature. The official Pi case just barely eeks above that and momentarily throttles down to 1.3 GHz. This is why I did the test at 1.8 GHz for the official Pi case afterwards.

So, I've gotta say, this Neo case is pretty good.

Now, let's compare this to the Neo's bigger brother, the original Argon One. Both graphs are at the 2.0 GHz overclock.



The original Argon One, which is larger, is Argon's first case model and how they got started. It's not only a heatsink, it's also got a (software-controllable) fan. As we can see, fans work; even at 2.0 GHz, we don't go beyond 63 deg. C. I did the test twice at 2.0 GHz just to make sure the results were repeatable, and yep, they were. I didn't graph it, but at the stock 1.5 GHz, we were right at 55 deg. C. Had I programmed the fan to run at 100% all the time, based on another test that I did with the Micro Connectors aluminum case (includes fan), we would've gotten about 34 Celsius at idle, 49 Celsius at 1.5 GHz, and 60 Celsius at 2.0 GHz.

Speaking of the Micro Connectors aluminum case, here are the graphs for that one as well, both at 1.5 GHz and the 2.0 GHz overclock. In this case, I also fired up the glxgears OpenGL demo to exercise the GPU. It appears to make a difference of about 2-3 degrees C. Despite that the tests were done at only two clock frequencies, there are three graphs (explanation below).



Note: the Micro Connectors case comes with little aluminum heatsinks for the chips that tend to get hot, including the CPU. I did not use those for the first two tests, as I wanted to see how just the fan would do. The first two graphs show that it does pretty well. At 1.5 GHz, the Pi stayed right at about 49-50 Celsius. At 2.0 GHz, though, we went up to about 71 Celsius, about 69 when not also running glxgears. That's on average about 10 degrees higher than I had expected, which goes to show the value of the "heatsink" nature that both of the Argon cases provide. Even without the CPU heatsink, the Micro Connectors case cooled considerably better than the official Raspberry Pi 4 case+fan. The official case's fan came with a CPU heatsink, which *was* used during the tests...and it still got beat by the MC case without its CPU heatsink. How, therefore, would the MC case perform *with* the CPU heatsink?

The third graph is the answer to that very question. With the CPU heatsink, it performs every bit as well as the Argon One case and far-and-away better than the official Raspberry Pi case+fan+heatsink. There is a dip in temperature there close to the end for a few minutes. That's when I momentarily stopped glxgears to see if there would be another 2-3 degree temperature drop, like before (yep, there was).

We can see from these results the following.

1.) Fanless with big heatsink is pretty good.
2.) Fan is pretty good.
3.) Fan with heatsink is very good.

So, which one?

Remember that most desktop users don't put full CPU loads on their computers for a full hour like I did here. However, watching a long video, such as a two-hour movie, could put a load on the CPU for an extended period. That's why tests like this are valuable. Also, the overclocking is applicable here, as we are talking about "daily driver" desktop computer duty here.

Fanless applications, the Neo does the job, better than the official case with its optional fan. I like this case enough that I went and bought a second one. At US $15, you get a high quality unit that has essentially no risk of parts breakage.

High loads that would require the fan? Original Argon One, because should the fan fail, it will act like the Neo and continue to cool passively. That level of cooling backup is a very nice thing to have. At US $25, this case not only performs well, but it has several other helpful features that I like for a desktop PC.

The Micro Connectors aluminum case also put in very good numbers at the stock clock frequency, and even with the Pi 4 overclocked, it did remarkably well when the CPU heatsink was added, matching the Argon One's performance at full load. Additionally, the Micro Connectors aluminum case comes with a good power supply (5V, 3A), which the Argon cases do not. At US $20, this makes the MC aluminum case a nice little bargain!

The official Raspberry Pi 4 case just cannot compare to these other three options, and that's with the optional fan+heatsink. Together, this is a US $12 package, and for that money, I'd step up to one of the other three options. They're just that much better.

[the86d]

What I did notice is that Linux, unlike Windows typically uses WAY less RAM to do the same tasks, even with a GUI running, unless using a *Buntu based distro with their windows manager.

Less about tracking you, and bloating stuff up, and more about lean and mean...

Slackware VM with KDE running:

root@ix:/home/****# free
              total        used        free      shared  buff/cache   available
Mem:        4018924      540416     2687168        4404      791340     3106628
Swap:       2097148           0     2097148
root@ix:/home/****#

[dizzyblonde]

For even better performance:

echo 1 > /proc/sys/vm/swappiness

That will keep it from swapping at all until it's essentially otherwise out of memory. Linux, like most operating systems, will start pumping stuff into swap even if it doesn't need to unless you take steps to make it not do that. Even USB 3.0 SSD's are slower than RAM.

[javaduke]

Wow, I was a bit surprised to find this RPi thread on Calguns, out of all places I started tinkering with Pi recently, trying to come up with an electronic airgun target, thought RPi would be perfect for this.
My idea was to have two cameras detecting pellet's X and Y position relative to the center of the target, but apparently the only two cameras solution I could find is Arducam, which is way too slow and can't reliably detect the pellet. Plus for some reason the exposure jumps up and down, triggering false object detection.
Wondering if you guys have any advice on how to improve the performance.

[pepsi2451]

Wow, I was a bit surprised to find this RPi thread on Calguns, out of all places I started tinkering with Pi recently, trying to come up with an electronic airgun target, thought RPi would be perfect for this.
My idea was to have two cameras detecting pellet's X and Y position relative to the center of the target, but apparently the only two cameras solution I could find is Arducam, which is way too slow and can't reliably detect the pellet. Plus for some reason the exposure jumps up and down, triggering false object detection.
Wondering if you guys have any advice on how to improve the performance.

I am not sure I understand what you are describing. Since you are using two cameras I assume you are not just looking at holes in a target. Are you trying to capture the actual bb/pellet in flight with the cameras?

[Rvg2151]

Looks nice, ive been playing with the OS on a VM.

[Cowboy T]

There are some new cameras available for the Pi. I haven't tried any of the camera functions at this point, so I'm just not knowledgeable in that area. But the Pi 4 definitely has two camera plugs on its board, so perhaps you could just use two separate cameras. Might be a good idea to head over to raspberrypi.org and see what they've got. Here's their high-quality model. I don't know what its reaction time is, but maybe it'll get you started.

https://www.raspberrypi.org/products/raspberry-pi-high-quality-camera/

[traviskeese00]

It looks like awesome


Sent from my iPhone using Tapatalk

[Cowboy T]

I have also recently tried out Ubuntu 20.10 "Groovy Gorilla" as a desktop OS on the Pi 4. This is the with Ubuntu's stock GNOME 3 desktop, which most people running Ubuntu use on their desktop or laptop PC's.

Short version: the Pi 4 can run it, but it's pretty sluggish with GNOME. Raspbian--now renamed "Raspberry Pi OS"--is a better choice for desktop usability.

Longer version: GNOME has always been a fairly "heavy" desktop, much like KDE. There's a lot of graphics stuff going on under the hood with GNOME, as befits a desktop environment designed for traditional (and very powerful) desktop computers. On a Raspberry Pi 4, GNOME's a bit too much, especially with Ubuntu's tweaks. With the standard Ubuntu GNOME desktop, the CPU definitely needs an overclock, as at the stock 1.5GHz speed, it's just...sluggish. Likewise, playing YouTube videos results in quite a few frames dropped. Overclocking to 2.0 GHz helps, but it's still pretty sluggish compared to Raspbian. Raspbian, by contrast, will play 1080-resolution YouTube videos without visible frame drops at 2.0 GHz. Further, the LXDE desktop environment included in Raspbian is light enough to run well, yet full-featured enough to let me just do my work without getting in my way.

Now, let's talk Ubuntu Server. For this purpose, since we're talking about a server, I chose Ubuntu 20.04 LTS "Focal Fossa", because servers are intended to be stable with high levels of uptime. Therefore, that's what I'd expect most server deployers to use.

Ubuntu Server 20.04 LTS installed just fine. The problem was when I went to use the "vcgencmd" utility. In a lot of cases, such as "vcgencmd bootloader_version" and "vcgencmd bootloader_config" simply do not work. The former returns "unknown", when it should return the bootloader firmware version. The latter returns...nothing...when it should return the boot configuration. That is emphatically *not* good when you need to tweak the box's boot parameters. I'm told that this problem is fixed in Ubuntu Server 20.10 "Groovy Gorilla", and if so, then that's great, but for that to be a problem on a Long-Term Support server release is inexcusable. If they couldn't get the proper version out at 20.04 release time, then that should've been an "apt update" fix. But it's still broken. The one advantage to Ubuntu Server 20.04 LTS at this time is that there's a 64-bit version. But unless your process uses more than 3GB DRAM total, you won't benefit from the 64 bit version (the 32-bit version can and does address > 4GB).

Therefore, Raspbian ("Raspberry Pi OS") version 10 "Buster", even though it's currently only 32-bit, really is the best choice. Yes, it can and does address all 8GB of DRAM in the latest Raspberry Pi 4, so long as each individual process doesn't use more than 3GB DRAM. It's considerably more responsive and more efficient with resources, especially graphics resources, than Ubuntu Desktop. And its "vcgencmd" utilities and such--the Pi-specific hardware utilities--actually WORK!

The Raspberry Pi Foundation folks were right. For general-purpose server or desktop computing, Raspbian really is the way to go on these little boxes.

[Cowboy T]

It looks like awesome

If used and configured properly, actually yeah, they are. They're neater than I expected.

[yellowsulphur]

Cowboy T I liked your review and analysis of the RPI4 cases. The Neo 1 looks nice and is stackable.

Currently working with VMware ESXi on the RPI4. Installed it on one board with the goal of using four Pi's for 16 threads and 32 GB of ram. My goal is the ability to deploy VM's like a cloud provider. Would be cool if it works.

[Cowboy T]

Hmm...didn't know there was VMware for the ARM processor. Could you let us know how this turns out?

[Cowboy T]

One more temperature test. It should be pretty evident by now that I like the Argon Neo case. As a straight passive cooling case, it's hard to beat.

However, it turns out there's a fan HAT, by Argon, specifically built to fit the Neo case, including with the removable magnetic top on, to protect from dust. However, given that the case is well built with tight tolerances, i. e. not very many ways for air to blow in and out...how would such a fan HAT perform with the Neo?

It cost me $10 at my local Micro Center to find out.

The Pi 4 is overclocked to 2GHz with ARM, and 750MHz for graphics. Without the fan HAT--just the normal passive cooling--and running stressberry-run for an hour, the Pi never throttled down, though after about 40 or so minutes, it came close at between 81 and 82 degrees Celsius. I had hoped that the fan HAT would help cooling with the magnetic top on. Sadly, no, it doesn't help any. The temperatures went up at the same curve as without the fan, and yes, the fan was working (I briefly lifted the top to check). We ended up just shy of 82 Celsius, exactly like without the HAT. So, I took the top off at the 50-minute mark, and the temperature went right down to about 54 Celsius on average (bounced between 52 and 57, with fan going on and off). The top stayed off during the cool-down period of 5 minutes, during which the temperature pretty quickly went down to about 39 Celsius, at which point the fan turned off and the temperature settled back to around 43 Celsius. Temperature stayed right there during the rest of the cool-down period.

I have a graph showing this, at this link. The difference is stark at around the 3000-second point, and you can see the sawtooth pattern from there. The dip close to the 4000-second point where the fan finally turned off is also evident.

Thumbnail is clickable.



For the HAT to help at all with the Neo case, you really do need to leave the magnetic top off. Perhaps the Neo case could be lightly machined in strategic places to allow for better cool air intake and hot air evacuation. And for passive cooling, it's hard to beat the Neo. But for active (that is, with a fan) cooling, especially with an overclocked Pi, the larger Argon One, with its pre-cut air holes, is the better option.

[Danodog]

I have a Pi 4B sitting in a drawer with a case, but I haven't decided what to do.....