MIPS Creator CI20 Now Available, Costs $65 USD

[uid313]

Very shitty, no Gigabit Ethernet, no 64-bit.
No HDMI, no DisplayPort, no eSATA, no USB 3.

The ARM boards seem like a much better choice.
Better hardware and better supported.

This MIPS board is very unconvincing.

[chithanh]

uid313,
Can you do anything else besides trolling and crapping into every thread that you post in?
The CI20 board has HDMI output.
DisplayPort, eSATA, USB3 and 64 bit are uncommon among ARM boards still.

Lack of Gigabit Ethernet is a valid criticism. As would be the only 1 GB RAM, lack of open source drivers for GPU acceleration and the price which is in the same ballpark as the better equipped Radxa Rock Lite.

[dh04000]

PowerVR?

Looks like this is a no-go for linux users unless they allow distributing their closed-driver like the raspberrypi gpu driver can be.

[Sonadow]

I have one of these boards running Gentoo, and it uses a proprietary firmware blob for the WiFi (this was already mentioned by jhogan).
The graphics uses the framebuffer or modesetting driver depending on the kernel version.

What's the performance like with the generic framebuffer / kms driver? Is it at least capable of moving windows without those horribly laggy frame-by-frame redraws?

[chithanh]

Performance is ok for text editing and web browsing, same as you would expect with xf86-video-modesetting on a desktop computer. For full-screen video it is not enough, though.

[ssvb]

I have one of these boards running Gentoo

Could you please paste the /proc/cpuinfo information here? I wonder if it supports at least the MIPS DSP ASE instructions. Yes, this CPU is supposed to have some sort of the Ingenic-specific undocumented XBurst SIMD, which is probably a better option for optimized software rendering in the long run. But unlike the standard MIPS DSP ASE, it is unlikely to be usable out of the box in any open source software.

Additionally, it would be great to have memory performance benchmark results. Even having fast SIMD is not very useful unless also backed by sufficiently fast memory.

[ssvb]

What's the performance like with the generic framebuffer / kms driver? Is it at least capable of moving windows without those horribly laggy frame-by-frame redraws?

The generic framebuffer / kms drivers are not exactly great for moving windows on any hardware. The biggest performance problem is the need to take care of extremely slow read back from the framebuffer, which is solved by introducing an extra shadow framebuffer layer in these drivers. The shadow framebuffer layer introduces a significant performance overhead by itself (that's an extra intermediate buffer with extra memory copy operations) and also has a nasty habit of postponing and coalescing screen updates (which looks like dropped frames on fast animation or jerky window moves). Moreover, handling expose events is the last nail in the coffin.

All these problems are resolved in https://github.com/ssvb/xf86-video-fbturbo which is much faster than the fbdev ddx on the Raspberry Pi (and on other ARM hardware). This is possible on ARM, because ARM devices usually don't have a dedicated video memory connected via a slow PCI-E bus, but instead keep framebuffer in the system memory. The framebuffer memory is uncached, so some tricks are required to ensure that reading back from it is sufficiently fast. However a single NEON or VFP instruction can transfer a lot of data at once, and doing NEON/VFP reads with proper alignment can reduce the uncached read penalty to the acceptable level. This allows to get rid of the shadow framebuffer band aid and improve overall performance.

As for the MIPS hardware, some experiments can be done with SIMD instructions and framebuffer read back. It might also work reasonably good.

[chithanh]

Could you please paste the /proc/cpuinfo information here?

I can do it tonight when I get to my CI20. If you want it faster, you can ask in #ci20 on freenode IRC.

I wonder if it supports at least the MIPS DSP ASE instructions. Yes, this CPU is supposed to have some sort of the Ingenic-specific undocumented XBurst SIMD, which is probably a better option for optimized software rendering in the long run. But unlike the standard MIPS DSP ASE, it is unlikely to be usable out of the box in any open source software.

Ingenic's MXU application specific extensions are supported by specially patched binutils. Upstreaming them is planned but has not happened yet:

https://groups.google.com/forum/#!topic/mips-creator-ci20/NqB9cGgSoj0

[chithanh]

It lists onlx MXU ASE, but that could be due to the kernel not having detection for other ASEs implemented. Or maybe the SoC doesn't support them after all.

ci20-gentoo ~ # cat /proc/cpuinfo 
system type             : CI20
processor               : 0
cpu model               : Ingenic Xburst V4.15  FPU V0.0
BogoMIPS                : 1196.85
wait instruction        : yes
microsecond timers      : no
tlb_entries             : 32
extra interrupt vector  : yes
hardware watchpoint     : yes, count: 1, address/irw mask: [0x0fff]
microMIPS               : no
ASEs implemented        : mxu
shadow register sets    : 1
kscratch registers      : 0
core                    : 0
VCED exceptions         : not available
VCEI exceptions         : not available
Hardware                : CI20
Serial                  : 28691e17 12810803 ee260000 8a61aa4f

processor               : 1
cpu model               : Ingenic Xburst V4.15  FPU V0.0
BogoMIPS                : 1196.85
wait instruction        : yes
microsecond timers      : no
tlb_entries             : 32
extra interrupt vector  : yes
hardware watchpoint     : yes, count: 1, address/irw mask: [0x0fff]
microMIPS               : no
ASEs implemented        : mxu
shadow register sets    : 1
kscratch registers      : 0
core                    : 0
VCED exceptions         : not available
VCEI exceptions         : not available
Hardware                : CI20
Serial                  : 28691e17 12810803 ee260000 8a61aa4f

[speculatrix]

The open-source hardware boards from Olimex appeal to me the most out of all the cheap embedded SoC boards I've seen