Qualcomm has just introduced an upgrade to its Snapdragon 820 processor, with Snapdragon 821 (MSM8996 Pro) with basically the same features, but the CPU frequency boosted from 2.2 GHz to 2.4 GHz for the performance cores, and from 1.6 to around 2.0 GHz for the low power cores, as well as a likely boost to the GPU clock, resulting to about 10% improvement in performance.
The new SoC with feature the same Snapdragon X12 LTE modem delivering up to 600 Mbps. It’s unclear at this point which smartphone models will feature Snapdragon 821 processor.
Allwinner, a leading tablet SoC vendor, and Qualcomm have decided to collaborate, and introduced three new LTE tablet reference designs based on Qualcomm Snapdragon 425, 430, and 435 available to Chinese OEMs, on top of Snapdragon 210, 212 and 410 designs released last year.
The full technical technical details about the reference designs are only available to OEMS who signed an NDA, but the key specifications are as follows:
The tablet will run Android, but Windows 10 is also being worked on. The agreement only covers tablets, so Allwinner will not be involved in LTE smartphones.
Since both Allwinner and Qualcomm are silicon vendors, and competitors, you may wonder why they’ve partnered. Allwinner only provides WiFi and Bluetooth tablets with their own processors, so partnering with Qualcomm allows then to offer LTE tablets. Many Chinese manufacturers don’t have a license agreement with Qualcomm, so those will be able to provide Qualcomm tablets solution through Allwinner.
The responsibilities of each stakeholders is well explained in the first chart: Qualcomm will provide the chips, global marketing, and technical support, Allwinner will design the reference designs and work on the SDKs, both to be released to design houses such as Emdoor, working for manufacturers.
The only Snapdragon tablets currently offered on Aliexpress are made by Huawei, and certainly not designed through this program. But the tablets based on the first generation of reference designs, such as the ones made by Cube, have been showcased at Mobile World Congress Shanghai 2016 last week, so we should be expecting low cost Snapdragon based LTE tablets to competing with Mediatek ones in the near future.
A few weeks ago, I was informed that some code about DB600c board powered by Qualcomm Snapdragon 600 processor (APQ8064T) was making it into mainline Linux, and more recently I found a website listing DragonBoard 600c with a low resolution picture of the board. While we don’t have the complete specifications yet, the form factor of the board is quite interesting, as we’ll find the typical 96Boards CE form factor on the right, and some extra interfaces on the left with Ethernet and SATA. It turns out, as we’ll see below, it’s perfectly compliant (hardware wise) with 96Boards CE “Extended Version” specifications.
DragonBoard 600c vs DragonBoard 410c
Preliminary specifications of DragonBoard 600c board:
Connectivity – Gigabit? Ethernet via PCIe . I can’t see WiFi and Bluetooth on the board, but since “Wi-Fi 802.11g/n and Bluetooth 4.0 LE” are required by 96Boards the specs, it could be on the back of the board.
USB – 2x USB 2.0 host ports, 1x micro USB port
1x 40 pin low speed expansion connector – UART, SPI, I2S, I2C x2, GPIO x12, DC power
Power Supply – 6.5 – 18V DC input (based on 96Boards specs)
Dimensions – 100 x 85 mm
The board should support the latest version of Android as well as Debian 8, based on the work done by Linaro on DragonBoard 410c.
Click to Enlarge
I’ve included the mechanical drawing for 96Boards Consumer Edition Extended Version as it should that designer can pretty much do whatever they want in the extended area, except for the position of mounting holes and power jack, and the maximum height of components limited to 6.5mm for Extended A, and 15 mm for Extended B.
I’m not sure when the board will be formally introduced and available, but considering there are working samples for developers, and most features have been found to work, it might not be too far away. There’s also a DragonBoard 820c with APQ8096 processor in the works, but I could not find pictures, nor code commits about DB820c, so the launch is likely many months away, or possibly early next year.
I purchased Qualcomm DragonBoard 410c development board last year, and first tested it and run some benchmark on the 96Boards compliant hardware with Android. I found that it was still work-in-progress, and decided to wait before trying Debian on the board. I’ve now done so, and will report by experience installing Debian Linux, playing with the board, and running Phoronix benchmarks to compare it to other ARM Linux boards.
Installing Debian on DragonBoard 410c
The first challenge is to navigate through the documentation that is not always clear or up-to-date. I eventually ended up on DragonBoard 410c Wiki on Github.
You then have to decided which image you want. While there are two official operating systems with Android and Debian, you can three “entities” releasiong their own images. For Debian specifically, you have the Linaro image, and Reference Platform Build (RPB) image. I could not find any changelog or known issues with the former, but the latter as its own Wiki with the latest release being RPB 16.03 (March 2016), and the next one scheduled to be RPB 16.06 in June.
That’s the current list of known issues
bug 285 USB host doesn’t detect any plugged devices
bug 121 [RPB] Cannot soft power off or shutdown db410c
bug 284 [RPB] Dragon board Display sleep not working
bug 289 [RPB] USB devices don’t work after reboot
bug 207 [RPB] Bluetooth does not work on Dragon board debian
bug 153 [RPB] Missing information about hwpack usage
USB host not working did not inspire confidence, so I first tested the Linaro image. The (other) Wiki points to the “latest version”, but the link would point to Linaro Debian 16.02 release, while I could find a more recent Linaro Debian 16.04 which I downloaded in a terminal:
I used a micro SD card to install it. If you use Windows, simply use Win32DiskImager, but in computer running Linux or in Windows via Windows subsystem for Linux, you may want to do it in the terminal. First check the SD card device with lsblk. Mine was /dev/sdb, but your may be different, and I use /dev/sdX in the command below tp flash the Debian installer to a micro SD card:
Now remove the micro SD card from your computer and insert it in to the board, set the jumper to boot from SD card on the DragonBoard 410c, and connect the power. I could see LED 1 blinking, but nothing on my HDMI TV. Last time, I did not manage to make the serial console (requiring a 1.8V USB to TTL board or cable) using Hardkernel ODROID board, so I went to the support forums, and after several minutes of reading, I found that the RPB image is recommended, as well as a clear explanation between the Linaro and RPB images:
Use the Reference Platform Build instead of the Linaro release. The Reference Platform is an integrated build with support for multiple boards, and that is where all engineering effort is going. The Linaro build is the old single-platform image that we’re not working on anymore.
The reference platform will run on all 96boards CE (Consumer Edition) and EE (Enterprise Edition), while the Linaro image is built specifically for a given board, and they are not really working on it. [Update: This answer was specific to Hikey board, and for DragonBoard 410c there are two images provided by Qualcomm Landing Team and the Reference Platform team]
So let’s start again from scratch using the RPB image, and download the bootloader, Linux kernel and rootfs to my Ubuntu computer:
That was a lot of commands to install the operating system… Now you can unplug the board, remove the micro USB cable, and connect the power again. After a few seconds, you should see the kernel log, and eventually LXDE desktop environment.
Click to Original Size
You’ll be asked to configure WiFi, and you’re basically done.
DragonBoard 410c Debian System Info
I’ve then run a few command to learn more about the image and system:
Linux linaro-alip4.4.0-104-arm64#1 SMP Debian 4.4.0.linaro.104-1.linarojessie.1 (2016-03-01) aarch64 GNU/Linux
FilesystemSizeUsed Avail Use%Mounted on
Features:fp asimd evtstrm crc32
Features:fp asimd evtstrm crc32
Features:fp asimd evtstrm crc32
Features:fp asimd evtstrm crc32
One of the main advantage of 96Boards should be recent Linux version,and that’s exactly what we have here with Linux 4.4 running on the board. Out of a total of 866MB reported RAM, 64MB is free, and the 6.9GB rootfs has 4.8 GB available to the user. Snapdragon 410 SoC is correctly reported as being a quad core Cortex A53 (0xd03) processor.
I used file utility to make sure a 64-bit rootfs is being used here:
The thing that often do not work on ARM Linux board are 3D graphics and hardware video decoding, so I’ve specifically tested these two, and also played with the pre-installed Chromium browser.
If I understand correctly the debian image comes with Freedreno open source graphics driver, and if that’s the case I have the first ever platform with working open source 3D graphics drivers:
vertex shader info:
fragment shader info:
vertex shader info:
fragment shader info:
So that means both framebuffer and X11 3D graphics acceleration are working. Nice !
I also tried to play Tuxracer as it was part of the board’s test results provided by Linaro.
sudo apt-getinstall extremetuxracer
It works, but it’s so slow that it’s barely playable (see video below).
I installed VLC to play 1080op h.264 videos, but based on the CPU usage the system is clearly using software decoding, and there’s no audio via HDMI. I’ve asked about those two issues on the forums about 24 hours ago, but I have yet to get a reply.
Chromium loads OK, but I did notice some freezes during use, and YouTube will struggle at full screen at 1080p, in similar way to many other low end ARM Linux platforms.
After over 3 hours the results are in. Bear in mind that the board does not have heatsink, just a metallic shield, and this may affects the performance. It’s also running an OS with a 64-bit ARM rootfs, while platforms like Raspberry Pi 3 features a 64-bit processor running 32-bit code.
Click to Enlarge
I like to check John the Ripper for multi-threaded performance.
While FLAC audio encoding is nice to single threaded performance.
In theory the CPU performance of Snapdragon 410 and Broadcom BCM2837 (as found in RPi 3) should be equal since both are quad core Cortex A53 processors @ 1.2 GHz, but for some reasons DragonBoard 410c is a little slower in the multi-threaded benchmark, and quite faster during FLAC audio encoding likely due to software differences (Aarch64 vs Aarch32).
So this is later on a Sunday than my usual schedule, because I just couldn’t make up my mind whether I should do another rc8 or not, and kept just waffling about it. In the end, I obviously decided not to,but it could have gone either way.
We did have one nasty regression that got fixed yesterday, and the networking pull early in the week was larger than I would have wished for. But the block layer should be all good now, and David went through all his networking commits an extra time just to make me feel comfy about it, so in the end I didn’t see any point to making the release cycle any longer than usual.
And on the whole, everything here is pretty small. The diffstat looks a bit larger for an xfs fix, because that fix has three cleanup refactoring patches that precedes it. And there’s a access type pattern fix in the sound layer that generated lots of noise, but is all very simple in the end.
In addition to the above, there’s random small fixes all over-shortlog appended for people who want to skim the details as usual.
Go test, and obviously with 4.5 released, I’ll start the merge window for 4.6.
Linux 4.4 added support for a faster and leaner loop device, 3D support in virtual GPU driver, TCP improvements, various file systems improvements for BTRFS, EXT-4, CIFS, XFS etc… Some notable changes made to Linux 4.5 include:
Copy offloading with new copy_file_range(2) system call – Performance improvements on local file systems are marginal, but for networked file systems such as NFS, you could copy a file internally on a server drive without transferring file data over the network.
Btrfs free space handling scalability improvements – New, experimental way of representing the free space cache that takes less work overall to update on each commit and fixes the scalability issues for large drives (30TB+). It can be enabled with -o space_cache=v2 mount option, and you can revert to the one method with -o clear_cache,space_cache=v1.
Support for GCC’s Undefined Behavior Sanitizer (-fsanitize=undefined) – UBSAN (Undefined Behaviour SANitizer) is a debugging tool available since GCC 4.9. It inserts instrumentation code during compilation that will perform checks at runtime before operations that could cause undefined behaviors. Linux 4.5 supports compiling the kernel with the Undefined Behavior Sanitizer enabled.
Next gen media controller whose “goal is to improve the media controller to allow proper support for other types of Video4Linux devices (radio and TV ones) and to extend the media controller functionality to allow it to be used by other subsystems like DVB, ALSA and IIO”. See lkml for details
Some new features and improvements specific to the ARM architecture:
Allwinner A80 support – IR receiver driver, NMI controller,PRCM driver, R_PIO support, and RSB driver
Allwinner H3 SoC support – H3 USB PHY clocks
A10/A20 Video Engine clocks
MIC1 capture for sun4i codec
Audio codec enabled on various boards
Added board – Orange Pi Plus
Crypto module and io-domain driver enabled in multi_v7_defconfig
Tweaks for RK3368 SoC and eval board
Added Rockchip RK3228 SoC and eval board
New RK3228 subdriver in pinctrl
SPI driver fix
Added support for RK3399 in thermal driver
RK3036: Added SMP support, emac support
Expose USB PHY PLLs
Device tree changes – Add watchdog node to meson8b, add status LED for ODROID-C1
Watchdog timer modifications
eMMC/SDIO minor fixes usage of bindings on Snow and Peach chromebooks.
Remove FIMD from Odroid XU3-family because on XU3 it cannot be used yet and on XU3-Lite and XU4 it is not supported.
Remove deprecated since June 2013 samsung,exynos5-hdmi.
Add support for Pseudo Random Generator on Exynos4 (Trats2 for now). This depends on new SSS clock.
Add rotator nodes for Exynos4 and Exynos5.
Switch DWC3_1 on Odroid XU3 and XU3-Lite to peripheral mode because now it cannot be used as OTG.
Cleanup the G2D usage on Exynos4 and add it to a proper domain in case of Exynos4210.
Put MDMA1 in proper domain on Exynos4210 as well.
New pinctrl subdrivers for Qualcomm MSM8996, PM8994, PM8994 MPP support
Added Qualcomm PCIe controller driver
Qualcomm ARM64: Add fixed rate oscillators to dts, fixup PMIC alias and properties, change 8916-MTP compatible to be compliant with new scheme, fix 8×16 UART pinctrl configuration, add SMEM, RPM/SMD, and PM8916 support on MSM8916
ARM SoC multiplatform code – “This branch is the culmination of 5 years of effort to bring the ARMv6 and ARMv7 platforms together such that they can all be enabled and boot the same kernel”
ARM64 – hugetlb: add support for PTE contiguous bit; perf: add support for Cortex-A72;
Other new hardware or SoCs – Sigma Designs ARM Cortex-A9 Tango4 “Secure Media Processor” platforms (SMP8756, SMP8758, and SMP8759), TI-based DM3730 from LogicPD (Torpedo), Cosmic+ M4 (nommu) initial support (Freescale Vybrid), Veyron-mickey (ASUS Chromebit), BCM2836 and Raspberry Pi 2 B.
Add support for PIC32MZDA platform
bcm963xx: Add Broadcom BCM963xx board nvram data structure
dts: Add initial DTS for the PIC32MZDA Starter Kit
math-emu: Add IEEE Std 754-2008 ABS.fmt and NEG.fmt emulation
math-emu: Add IEEE Std 754-2008 NaN encoding emulation
zboot: Add support for serial debug using the PROM
If you want to get the full details, I’ve generated Linux 4.5 Changelog with comments only (12.2MB) using git log v4.4..v4.5 --stat, but it’s probably a better idea to simply check out Linux 4.5 changelog on kernelnewbies.org.
802.11ad is the latest and fastest WiFi standard working in the 60 GHz band and delivering up to 7 Gbit per second data transmission rates. The 60 GHz frequency band offers both advantages and disadvantages because it does not penetrate through walls nor water, meaning it can only be used within a room limiting the range, but at the same time it’s more secure since it cannot be snooped from the outside, and for people who worry about health effects it does not penetrate the human body. 802.11ad routers will also be able to switch to 2.4 and 5.0 GHz frequency bands in order to go through walls.
The table above nicely summarize the key features of 802.11ad over 802.11ac and 802.11n, however the throughput row shows the theoretical maximum throughput, but in practice, using 802.11ac as example, clients are often limited to 433 or 866 Mbps, and distance and obstacles will even lower the performance further.
Wikipedia also list the following key features for WiGig MAC and PHY Specification version 1.1:
Supports data transmission rates up to 7 Gbit/s – more than ten times faster than the highest 802.11n rate
Supplements and extends the 802.11 Media Access Control (MAC) layer and is backward compatible with the IEEE 802.11 standard
Physical layer enables low power and high performance WiGig devices, guaranteeing interoperability and communication at Gigabit rates
Protocol adaptation layers are being developed to support specific system interfaces including data buses for PC peripherals and display interfaces for HDTVs, monitors and projectors
Support for beamforming, enabling robust communication at distances beyond 10 meters. The beams can move within the coverage area through modification of the transmission phase of individual antenna elements, which is called phase array antenna beamforming.
Widely used advanced security and power management for WiGig devices
Applications for the higher bandwidth include faster download speeds, 4K wireless video, in-room gaming, etc…
60 GHz Frequency Bands for 802.11ac per Regions/Countries
So when will 802.11ad become available? Very soon, as TPLink unveiled Talon AD7200 Multi-band 802.11ad Wi-Fi Router at CES 2016, supporting up 7200Mbps Wi-Fi speeds over 2.4GHz (800Mbps), 5GHz (1733Mbps), and 60GHZ (4600Mbps) bands, and scheduled to be available in “U.S. stores in early 2016”, while LeEcho, previously known as LeTV, has just launched Le Max Pro (X900) smartphone featuring 802.11ad WiFi in China (also found in Aliexpress), and showcased in ARMDevices.net video where Qualcomm demonstrates 802.11ad with the phone by streaming a 4K video at 50 Mbps to a 802.11ad dock connected an UltraHD TV, and downloading data up to 2.6 Gbps with the phone.
When Samsung announced Exynos 8890 processor, it promised 10% lower consumption and 30% high performance compared to Exynos 7 Octa. The company also said it make its own custom ARMv8 cores for the new, but at the time details were limited. Anandtech has now published more information, and Exynos 8890 octa-core processor will make use of four Exynos M1 custom cores combined with four low power ARM Cortex A53 cores, combined with a Mali-T880MP12 GPU.
Exynos 8890 key features:
Low power cores – 4x ARM Cortex A53 cores @ 1.586GHz
High performance cores – 2x Exynos M1 @ 2.60 GHz, 2x Exynos N1 @ 2.29 GHz; If 4 cores are running at the same time: 2.29 GHz maximum
Now that Samsung Galaxy 7 has been announced with Exynos 8890 processor (at least one of its versions), as well as LG G5 with Snapdragon 820 processor using Qualcomm custom ARMv8 Cryo cores, benchmarks have also started to show up.
The two new processors are usually performing much better than last year devices in benchmarks except for graphics where the Apple A9 processor found in iPhone 6s still performs better, and somehow Galaxy Note 5 (Exynos 7 Octa) achieved a better than LG G5 in Vellamo Browser. There’s however a good reason for iPhone 6s out-performance here as its screen resolution is 1920×1080, while the new models have higher resolution (2560 x 1440). Luckily Anandtech has the good idea of running offscreen tests instead of on-screen ones, and the results are quite different.
It should be noted that MDPs (Mobile Development Platforms) can somtimes deliver better performance than smartphone due to better cooling. AnandTech also noticed that both Snapdragon 820 based LG G5 and Exynos 8890 based Galaxy S7 smartphones got pretty warm after testing.