Nano-pi Fire 3: The curse of Mali

Being able to buy SBCs (single board computers) for as little as $35 has revolutionized computing as we know it. Since the release of the Raspberry PI 1b back in 2012, single board computers have gone from being something electrical engineers work with, to something everyone can learn to use. Today you don’t need a background in electrical engineering to build a sophisticated embedded system; you just need a positive spirit, willingness to learn and a suitable SBC starter-kit.

Single board computers

If you are interested in SBC’s for whatever reason, I am sure you have picked up a few pointers about what works and doesn’t. In these times of open forums and 24/7 internet, the response-time from customers are close to instantaneous; and both positive and negative feedback should never be taken lightly. It’s certainly harder to hide a poor product in 2018, so engineers thinking they can make a quick buck selling sloppy-tech should think twice.

I have no idea how many boards have been released since 2016, but some 20+ new boards feels like a reasonable number. Which under normal circumstances would be awesome, because competition can be healthy. It can stimulate manufacturers to deliver better quality, higher performance and to use eco-friendly resources.

But it can also backfire and result in terrible quality and an unhealthy focus on profit.

The Mali SoC

The MALI graphics chipset is a name you see often in connection with SBC’s. If you read up on the Mali SoC it sounds fantastic. All that power, open architecture, partner support – surely this design should rock right? If only that were true. My experience with this chipset, which spans a variety of boards, is anything but fantastic. It’s actually a common factor in a growling list of boards that are unstable and unreliable.

I don’t have an axe to grind here, I have tried to remain optimistic and positive to every board that pass my desk. But Mali has become synonymous with awful performance and unreliable operation.

Out of the 14 odd boards I have tested since 2016, the 8 board that I count as useless all had the Mali chipset. This is quite remarkable considering Mali has an open driver architecture.

Open is not always best

If you have been into IOT for a few years you may recall the avalanche of critique that hit the Raspberry PI foundation for their choice of shipping with the Broadcom SoC? Broadcom has been a closed system with proprietary drivers written by the vendor exclusively, which made a lot of open-source advocates furious [at the time].

You know what? Going with the Broadcom chipset is the best bloody move the PI foundation ever did; I don’t think I have ever owned a SBC or embedded platform as stable as the PI, and the graphics performance you get for $35 is simply outstanding. Had they listened to their critics and used Mali on the Raspberry PI 2b, it would have been a disaster. The IOT revolution might never have occurred even.

The whole point of the Mali open driver architecture, is that developers should have easy access to documentation and examples – so they can quickly implement drivers and release their product. I don’t know what has gone wrong here, but either developers are so lazy that they just copy and paste code without properly testing it – or there are fundamental errors in the hardware itself.

To date the only board that works well with a Mali chipset, out of all the boards I have bought and tested, is the ODroid XU4. Which leads me to conclude that something has gone terribly wrong with the art of making drivers. This really should not be an issue in 2018, but the number of bankrupt mali boards tell another story.

Nano-PI Fire 3

When reading the specs on the Nano-pi fire 3 I was impressed with just how much firepower they managed to squeeze into such a tiny form-factor. Naturally I was sceptical due to the Mali, which so far only have ODroid going for it. But considering the $35 price it was worth the risk. Worst case I can recycle it as a headless server or something.

And the board is impressive! Let there be no doubt about the potential of this little thing, because from an engineering point of view its mind-blowing how much technology $35 buys you in 2018.

I don’t want to sound like a grumpy old coder, but when you have been around as many SBC’s as I have, you tend to hold back on the enthusiasm. I got all worked-up over the Asus Tinkerboard for example (read part 1 and part 2 here), and despite the absolutely knock-out specs, the mali drivers and shabby kernel work crippled an otherwise spectacular board. I still find it inconceivable how Asus, a well-respected global technology partner, could have allowed a product to ship with drivers not even worthy of public domain. And while they have updated and made improvements it’s still not anywhere near what the board could do with the right drivers.

The experience of the Nano-PI so far has been the same as many of the other boards; especially those made and sold straight from smaller, Asian manufacturers:

• Finding the right disk-image to download is unnecessarily cumbersome
• Display drivers lack hardware acceleration
• Poor help forums
• “Wiki” style documentation
• A large Linux distro that max out the system

More juice

The first thing you are going to notice with the Nano-pi is how important the power supply is. The nano ships with only one usb socket (one!) so a usb hub is the first thing you need. When you add a mouse and keyboard to that equation you have already maxed out a normal 5v 2a mobile power supply.

I noticed this after having problems booting properly when a mouse and keyboard was plugged in. I first thought it was the SD card, but no matter what card I tried – it still wouldn’t boot. It was only when I unplugged the mouse and keyboard that I could log in. Or should we say, cant log in because you don’t have a keyboard attached (sigh).

Now in most cases a Raspberry PI would run fine on 5v 2a, at least for ordinary desktop work; But the nano will be in serious trouble if you don’t give it more juice. So your first purchase should be a proper 5 volt 3 amp PSU. This is also recommended for the original Raspberry PI, but in my experience you can do quite a lot before you max out a PI.

Bluetooth

A redeeming factor for the lack of USB ports and power scaling, is that the board has Bluetooth built-in. So once you have paired and connected a BT keyboard things will be easier to work with. Personally I like keyboard and mouse to be wired. I hate having to change batteries or be disconnected at random (which always happens when you least need it). So the lack of USB ports and power delegation is negative for me, but objectively I can live with it as a trade-off for more CPU power.

Lack of accelerated graphics

It’s not hard to check if X uses the gpu or not. Select a large region of the desktop (holding the left mouse button down obviously) and watch in terror as it sluggishly tries to catch up with the cursor, repainting every cached co-ordinate. Had the GPU been used properly you wouldn’t even see the repaint of the rectangle, it would be smooth and instantaneous.

SD-card reader

I’m sorry but the sd-card reader on this puppy is the slowest I have ever used. I have never tested a device that boots so slow, and even something simple like starting chrome takes ages.

I tested with a cheap SD-card but also a more expensive class 10 card. I’m really trying to find something cool to write about, but it’s hard when boot times is worse than Raspberry PI 1b back in 2012.

1 gigabyte of ram

One thing that I absolutely hate in some of these cheap boards, is how they imagine Ubuntu to be a stamp of approval. The Raspberry PI foundation nailed it by creating a slim, optimized and blistering fast Debian distro. This is important because users don’t buy alternative boards just to throw that extra power away on Ubuntu, they buy these boards to get more cpu and gpu power (read: better value for money) for the same price.

Lubuntu is hopelessly obese for the hardware, as is the case with other cheap SBC’s as well. Something like Pixel is much more interesting. You have a slim, efficient and optimized foundation to build on (or strip down). Ubuntu is quite frankly overkill and eats up all the extra power the board supposedly delivers.

When it comes to ram, 1 gigabyte is a bit too small for desktop use. The reason I say this is because it ships with Ubuntu, why would you ship with Ubuntu unless the desktop was the focus? Which again begs the question: why create a desktop Linux device with 1 gigabyte of memory?

The nano-pi would rock with a slim distro, and hopefully someone will bake a more suitable disk-image for it.

Verdict so far

I still have a lot to test so giving you a final verdict right now would be unfair.

But I must be honest and say that I’m not that happy about this board. It’s not that the hardware is particularly awful (although the mali drivers renders it almost pointless), it’s just that it serves no point.

In order to turn this SBC into a reasonable device you have to buy parts that brings the price up to what you would pay for a ODroid XU4. And to be honest I would much rather have an ODroid XU4 than four nano-pi boards. You have plenty of USB ports, good power scaling (ODroid will start just fine on a normal charger), Bluetooth and pretty much everything you need.

For those rare projects where a single USB is enough, although I cannot for the life of me think of one right now, then sure, it may be cost-effective in quanta. But for homebrew servers, gaming rigs and/or your first SBC experience – I think you will be happier with an original Raspberry PI 3b+, a ODroid XU4 or even the Tinkerboard.

Modus operandi

Having said all that .. there is also something to say about modus-operandi. Different boards are designed for different systems. It may very well be that this system is designed to run Android as it’s primary system. So while they provide a Linux image, that may in fact only be a “bonus” distro. We shall soon see as I will test Android next.

Next up, how does it fare with the multi-threaded uae4arm? Stay tuned for more!