New Libre-Focused ARM Board Aims To Compete With Raspberry Pi 3, Offers 4K

[grok]

2D graphics with hardware video decoders working is an achievement, I think it doesn't even work on my linux desktop lol. Would have been good on VIA (VX900 chipset and such) but all you could get was raw unaccelerated 2D with better support for resolutions/refresh rates than VESA driver, if you downloaded or compiled some bit of code from somewhere.


I do have 3D acceleration but rarely even use it, for now. Google Earth was a good app while it lasted, but it's now deprecated.
I have a little surprise for you : I have no use for Gnome 3, Unity, Compiz

[LoveRPi]

We are one of the backers of the project and here's the priority list from a marketing perspective for industrial and consumer adoption:

1) Compatible form factor for re-use, less waste, and drop-in compatibility
2) More RAM and faster storage
3) Good power budget to prevent the hundreds of thousands of power supply issues caused by the MicroUSB connector
4) Android 7.1 and Upstream Linux support is supported by vendor
5) Good performance (the board archieve around 50% performance gain)
6) HDMI 2.0 with 4K
7) Modern codec (H.265 and VP9) support at 4K
8) Pin to pin compatibility with 40 pin GPIO of Raspberry Pi 3 (others advertise 40 pin GPIO but the pin alternate functions are completely different and thus most HATs will never work)
9) Gigabit Ethernet (this didn't make it since s905x doesn't have RGMII)
10) USB 3.0 (this didn't make it since s905x doesn't have USB 3.0)
11) SATA 2.0 or 3.0
12) PCI Express 2.0 or 3.0

We were lucky that we achieved 1-8 using a single board. That being said, the company will create a lot more boards using different SoCs for different markets. This specific board will be a long term support board with at least 5 years of production.

Here's a list of things we didn't want from the get go:
onboard wifi/bluetooth since this creates more problems than it solves and a cheap $5 USB dongle will perform way better
high performance OoO CPU since there's no power budget for it on such a small board
proprietary connectors like CSI and DSI on the Pi 3 (not to be confused with headers)
unproven or unreliable or risky technologies (this always bites you in the ass in engineering)

There was a lot of consideration of the strengths and weaknesses of every board on the market including upcoming boards like the Rock64 and the cheap Allwinner solutions. They all have critical weaknesses to presented considerable risk to all parties involved. Copious amounts of consideration and software and hardware engineering went into this board and we are pleasantly satisfied with the outcome.

Yes in magic fairy land we would have Gigabit Ethernet, USB 3.0, PCI-E, SATA 3.0 etc etc but they do not constitute a sufficiently large market for SoC vendors to design a chip that ticks every box. 95% of people will not use the Gigabit Ethernet, USB 3.0, PCI-E or SATA 3.0 features to justify the cost. The RK3399 would be the closest thing but that would add $50 to the bill of materials on a $50 board resulting in a $150 board. A $150 board will never generate enough sales to achieve critical mass.

A $10-20 board will not have enough margins to support a professional team of engineers. If you want a couple of hundred thousand of engineers, makers, students and such to waste a few days each to save $10 per board, you're wasting everybody's time and money.

[waxhead]

I agree, but I'm sorry to inform you that it is done like this because it is better for the manufacturer's usecase so it's not likely to happen.

Yeah I know... it is far cheaper to skip all the sockets and just get everything soldered at once, but not as cool

[caligula]

I made a list a while back of the things a Pi competitor would have to have in order for the community to embrace it at the same level. Thus far, I'm not aware of any attempt to hit the target.

I can't find it at the moment but I can summarize.. Any board that doesn't actually improve in an area the Pi is lacking is doomed. There are lots of places to improve on the Pi. Simple things, like a real power connector instead of micro usb. SATA (or even USB3) interface for connecting real storage instead of these god-awful slow emmc/sd nonsense. Multiple USB (buses is the B so Multiple USB is correct when you want more than one bus) so that your network interface and your storage interface aren't competing for a lane that either of them could saturate..

What we get instead is different brand SoC, maybe a gig-e port still running on usb2, and various video chipsets that manage to be even worse supported than videocore.

Your list is outdated. The competitors already do all that. The GPU is the only obstacle. If you don't care about a bigger power budget, there are already few x86 boards that even do (3D) graphics. There are multiple UP board variants, UDOO X86, ... If you want RPi3 perf on an IoT board, the new NanoPi Neo 2+ does all that. If you have a tight budget, the Orange Pi boards are dirt cheap. E.g. OPI PC2 is half the price of RPi3 but has like 10 times the I/O bandwidth.

[caligula]

That wasn't even counting the CPU/GPU/memory bandwidth requirements needed for loading and decoding, nor the computational requirements, just the framebuffer size.

There are limits to how much computational power and bandwidth one can cram into a US$30 embedded board's CPU.

Of course there are limits. Graphics processing is better off delegated to a dedicated chip IP anyways. It doesn't need to be a 3d accelerator.

I originally only replied to 'Nope. 3D isn't required for most embedded projects.' I'm not sure why the 4k graphics discussion started again. The truth is, quite many embedded boards DO NOT use 3D graphics, no OpenGL, no Direct3D, no OpenGL ES, no Vulkan. There. Go buy a $200 washing machine - does it run OpenGL 4.5 or Vulkan? I'm pretty sure it does not. Looking at smart watches, the best ones have a resolution of around 400x400x24b. Of course THERE ARE appliances that actually use 3D accelerated graphics, but then again some of those appliances don't need that much power. For instance, the touch screen you use to pay your groceries does not require smooth 60 fps 3D HDR VR graphics. It could even work without a double buffer @ 0.5 FPS and nobody would give a flying f*ck. In fact, I would guess that this is exactly how it works.

Having a machine capable of 4k display output does not imply that all graphics must be 4k, 3D, HiDPI, stereoscopic, VR, HDR, AND triple buffered. In fact if you bought a RPi form factor board, you would have hard time even buying a compatible, portable 4k display for a specific project. I only pointed out that you could indeed have playable, relatively high resolution graphics even on the early 1990s machines. What's wrong with that graphics now? If the machines are 100 times faster now, why isn't it enough to make the graphics 100 times better?

Btw, I checked the specs. So, it seems the memory bandwidth has grown 100-fold (single channel). Dual channel is common even on SBCs. CPU speed is enormously faster. Two DDR4 modules can transfer up to 50 GB/s. 1920x1080x24b @ 50 fps requires 600 MB worth of reading and 300 MB worth of writing per frame, assuming double buffering and a shared RAM for the frame buffer. So, that's less than 2% of the memory bandwidth. Even 4k @ 60 fps would use less than 10% of the bandwidth. You could certainly improve this without resorting to a 3d accelerator. It's already braindead to begin with normal DDR for these kinds of aggregate memory operations when super wide GDDR exists.

As such, it HAS to use the dedicated decoding hardware at even 720p, let alone 1080p. I doubt said hardware is capable of keeping up above that with anything but the lowest-power codecs (like MPEG1/2), and I further doubt it could handle even that at 4k resolutions.

Right, but video decoding is totally different. Video decoding uses fixed function pipelines / DSP / GPGPU code, whereas only the graphics requires Mali 3d acceleration. The Mali issue has nothing to do with video playback unless we talk about GUI layers on top of the video. Accelerated video playback is normal practice everywhere. All GPUs, APUs, most SBCs and desktop CPUs support it. It's also a lot easier to implement as you just push this stream through the vendor API and everything just works. Video engines of many boards have already been reverse engineered with little effort, Mali 3d graphics seems like a dead end.

[squash]

I made a list a while back of the things a Pi competitor would have to have in order for the community to embrace it at the same level. Thus far, I'm not aware of any attempt to hit the target.

I can't find it at the moment but I can summarize.. Any board that doesn't actually improve in an area the Pi is lacking is doomed. There are lots of places to improve on the Pi. Simple things, like a real power connector instead of micro usb. SATA (or even USB3) interface for connecting real storage instead of these god-awful slow emmc/sd nonsense. Multiple USB (buses is the B so Multiple USB is correct when you want more than one bus) so that your network interface and your storage interface aren't competing for a lane that either of them could saturate..

What we get instead is different brand SoC, maybe a gig-e port still running on usb2, and various video chipsets that manage to be even worse supported than videocore.

[mulenmar]

Yes, but 4 cores x 1200 MHz is with LPDDR3 is also a bit faster than 66 MHz 486 with EDO RAM. Many of the 320x200 games were often playable on 33 MHz 486SX. E.g. Doom. Settlers 2 used 640x480 or even higher res.

There's something like a couple orders of magnitude difference in how much memory and bandwidth is necessary to shuffle ((320 * 200 * 8bpp) / 32bits-per-byte) = 15.625 KiB, multiplied by 2 for double-buffering, out to a display at 20FPS... and how much is needed to shuffle ((3840 * 2160 * 24bpp) / 32bits-per-byte) = 6.075 MiB, multiplied by 2 for double-buffering or 3 for triple-buffering, out to a display at 60FPS or, worse, higher.

That wasn't even counting the CPU/GPU/memory bandwidth requirements needed for loading and decoding, nor the computational requirements, just the framebuffer size.

There are limits to how much computational power and bandwidth one can cram into a US$30 embedded board's CPU. As such, it HAS to use the dedicated decoding hardware at even 720p, let alone 1080p. I doubt said hardware is capable of keeping up above that with anything but the lowest-power codecs (like MPEG1/2), and I further doubt it could handle even that at 4k resolutions.

[mulenmar]

You mean this?

...*facepalm* Yes, that is it, thank you! I was looking in the wrong IRC, it seems.

[caligula]

4k is 130 times larger than 320x200, and this is without factoring in bit depth, which will make it more than .5k times larger

Yes, but 4 cores x 1200 MHz is with LPDDR3 is also a bit faster than 66 MHz 486 with EDO RAM. Many of the 320x200 games were often playable on 33 MHz 486SX. E.g. Doom. Settlers 2 used 640x480 or even higher res.

I kind of doubt anyone would use the 4k mode for desktop. It's mainly for the video playpack. Even some Rpi distros like OpenELEC might default to 720p for the GUI and 1080p for the video. It just can't handle such modes even with GPU acceleration.

[stefansaraev]

nope. if your embedded project requires 4k, it does require 3d

you serious?