Mediatek Labs Unveils Deca-core Helio X20 Development Board Compliant with 96Boards CE Specifications

Orange Pi Development Boards

More and more companies are joining the 96Boards bangwagon, and the latest move by Mediatek Labs will mean their Helio X20 development board, designed in collaboration with ArcherMind, will be the most powerful 96Boards CE edition released so far thanks to its deca-core processor with two Cortex A72 cores, and 8 Cortex A53 cores.

Mediatek_96Boards_Development_BoardMediatek Helio X20 board specifications:

  • SoC – Mediatek Helio X20 (MT6797) deca-core processor with two ARM Cortex A72 cores @ 2.3 GHz, four Cortex A53 @ 1.95 GHz, four Cortex A53 @ 1.4 GHz, and  ARM Mali-T880 GPU @ 800 MHz
  • System Memory – 2GB LPDDR3 @ 933 MHz
  • Storage – 8GB eMMC 5.1 flash + micro SD 3.0 (UHS-I) slot
  • Video Output – HDMI up to 1080p @ 30 fps
  • Connectivity – 802.11 b/g/n WiFi, Bluetooth 4.1, GPS with antenna connector
  • USB – 2x USB 2.0 host ports, 1x micro USB port (device only).
  • Expansion:
    • 1x 40 pin low speed expansion connector – UART, SPI, 2x I2C, 12x GPIO, DC power
    • 1x 60 pin high speed expansion connector – 4L-MIPI DSI, USB, 2x I2C, 2L+4LMIPI CSI
    • Analog expansion connector – Headset, Speaker, FM antenna, and more
  • Misc – Power, reset and volume buttons. 6 LEDS (4x user, 1x Wifi, 1x Bluetooth)
  • Power Supply – 8 to 18V DC input
  • Dimensions – 85 x 54 mm

The board currently supports Android 6.0, and for now documentation is available in Chinese on Alpha Star website, with an English translation coming soon. The pre-sales page also mentions a daughterboard, but I could not find further details about it, nor the price of either boards.

More details should eventually surface on Mediatek Labs Helio X20 development board page.

Via Phoronix

Support CNX Software - Donate via PayPal or become a Patron on Patreon

31
Leave a Reply

avatar
31 Comment threads
0 Thread replies
1 Followers
 
Most reacted comment
Hottest comment thread
13 Comment authors
adegorantkaiserlukasubowator Recent comment authors
  Subscribe  
newest oldest most voted
Notify of
Theguyuk
Guest
Theguyuk

Not sure I understand the benefits of different speed cores, yes it benefits phones and tablets on battery but would not scaling the GHz do just the same? Would it be better using the die space to up the graphics or extra feature rather than duplicate cores?

How much will the board retail for?

Fossxplorer
Guest
Fossxplorer

Amazing with 10C! Curious about it’s performance in Linux vs Core i3/i5.

davidlt
Guest
davidlt

Does this get official 96Boards support (i.e. upstream support)?

nobe
Guest
nobe

@cnx
this soc should support wifi 802.11ac according to http://mediatek-helio.com/x20/

Member

Well if it doesn’t cost an ARM and a leg ( pun intended ) I’ll want one of these 🙂

miniNodes
Guest

Yes, very interesting (and powerful) solution, but of course price will need to be a consideration.

blu
Guest
blu

And A72 for the masses – yay (or not, depending on price).

Also, MediaTek can stop powering on and off their cores ‘on demand’ – this is not a phone, this a goddamned dev board. It is annoying on the BQ when I need to run a ‘wake-up-all-cores’ app before running a normal app so it could make use of all cores (because otherwise the app sees one core at start and launches one thread).

Member

@blu
I think this is somewhere in the kernel. Hopefully this will have an open source kernel so that problem should be easily solved. Also Linux software isn’t really that smart ( read dumb ) to not look at the entire number of available cores ( I think that’s available via /proc ).

wator
Guest
wator

The vendor website is Down!!!

lukasubo
Guest
lukasubo

@Marius

It isn’t the software that’s dumb, it’s the power management. It isn’t a phone. And if the source is not available, this board I’d useless dead in the water trash.

tkaiser
Guest
tkaiser

Is Mediatek releasing kernel and/or Android sources in the meantime? All I found was a reference to 3.18.22 (so the usual Android ‘port and forget’ style) and regarding the three CPU clusters only contradictory statements (A72 cores will be shutdown when thermal tresholds are exceeded and then you have an octa-core A53 cluster running in HMP mode in opposite to clustered switching). Anyway since the PCB doesn’t allow a HUGE heatsink I wonder why people think about performance at all.

tkaiser
Guest
tkaiser

@cnxsoft
Thx, the last post on your second link is the most interesting IMO (explaining different types of heatsinks to be used with either convection or ‘forced air’) 🙂

In my opinion boards with such insanely overheating SoCs that are not factory equipped with a huge heatsink (compare with ODROIDs) are simply design flaws. Applies to all the cheap A53 boards, to every octa-core board I know of except of NanoPC-T3 (no buy since kernel sucks) and will apply for sure to this Helios X20 thingie here too.

And based on the currently available information this will remain an Android only dev board?

blu
Guest
blu

@tkaiser
You’re absolutely right about the neglect of heat issues. Luckily, some of those boards (e.g. the geekbox) come with a minimal heatsink, so here’s what I strapped together to ensure my RK3368 runs cool: http://forum.geekbox.tv/viewtopic.php?f=14&t=3094

nobe
Guest
nobe

imho, this pcb main cooling problem is the lack of heatsink mounting holes

tkaiser
Guest
tkaiser

@blu
Hmm… the only A53 board so far that does not suffer from throttling problems is ODROID-C2. And by looking at the heatsink anyone might get the idea why.

Regarding all these octa- or even deca-core smartphone/tablet SoCs (as used here in 96Boards form factor) in the meantime I came to the conclusion that they’re simply not designed to run on all cores with maximum clockspeed according to specs. It’s either/or at least with ‘factory defaults’ (no heatsink and insane throttling settings from SoC vendor — at least true for Banana Pi M3: octa-core @ 2GHz and in reality it’s 1.8GHz max and quad-core after short load bursts since CPU cores simply get killed due to overheating and not come back until next reboot)

Maybe the whole selling point of deca-core smartphone SoCs like the one we’re talking about is that the average flagship phone user ‘needs’ 2 more CPU cores if his neighbor has an octa-core phone regardless of application performance or anything else? Don’t know — no Android user.

Member

Yes the cooling will be a problem but maybe this is just a prototype and the final version will have a much better cooling solution.
Anyway it’s the first really powerful 64bit ARM board I’ve seen so if it’s reasonably priced ( < $200 ) then it's very good and a way too cool it can be found.

goran
Guest
goran

why not usb3.0? what is the throughput of bus?

tkaiser
Guest
tkaiser

goran :
why not usb3.0?

This is a smartphone SoC for flagship phones. They need high Antutu scores (made in a lab using liquid cooling) and many CPU cores since the average smartphone customer buys numbers. Everything else is not relevant 🙂

That this SoC found its way on an SBC doesn’t change the initial design requirements. If you want high IO bandwidth do not choose deca-core ARMv8 but dual-core ARMv7 instead (Marvell Armada 38x for example is dual-core Cortex-A9 but will outperform Helio X20 by magnitudes if its about bandwidth). At least the use cases the vendor is talking about — digital signage and stuff like that — require not much IO bandwidth.

Theguyuk
Guest
Theguyuk

The answer as mentioned is the soc was designed for tablets and phone and prolonging battery life, they are not designed for constant long running heavy load. Best thinking of them as a city car that can do short motorway runs at a higher speed but is designed for small city streets and shopping runs.

blu
Guest
blu

@Marius
The problem with off-line CPUs, to the best of my understanding, stems from the lack of a mechanism for an app to tell if and how a CPU which is currently off-line can become online. Otherwise yes, the kernel correctly reports which CPUs are online and which – off-line.

The situation with the BQ tablet, in particular (MT8163, kernel 3.10.93+ “Android custom”) is the following:

The kernel keeps a single CPU online most of the time, and the remaining 3 cores are deep-sleeping. When a multi-threaded app starts, it faces the following dilemma: the kernel says there are some cores off-line – shall I assume those will be waken up if I provide them with threads, or not? In reality, the kernel _does_ wake up those cores when there are sufficient number of threads ready to run. But the app cannot know that. So most well-behaved MT apps say ‘oh, there’s just one core online – I’ll restrict my threading as much as possible”. The way the user can counter that, is by launching an app that bullishly launches 4 threads and keeps those loaded for a while, and then quickly launch the ‘well-behaved’ app, so by that time the latter would see a system with 4 cores online.

I’ve seen systems with the opposite behavior as well, where an app sees N cores off-line, assumes those will be waken up upon available work, so it creates N threads, but the kernel never wakes up cores, to some abysmal effect for the app.

ade
Guest
ade

I think the Helio x20 is a smartphone SoC and therefore is supposed to embed an advanced 3G/4G modem => this board does not include 3G/4G features ? no SIM card slot ? no 4G/3G antenna port ?

Member

@cnxsoft
Any idea where to actually get it from ?