Hardkernel ODROID-XU minicomputer on Exynos5 Octa (big.LITTLE Cortex-A15). Personal impressions
As many who are interested in this topic, they know that the company Hardkernel, specializing in the production of ODROID brand mini-computers on the Samsung ARM SoC, recently announced the new ODROID-XU board.
The board is built around the Exynos5 Octa Cortex-A15 with big.LITTLE technology. This is the fastest SoC at the moment available in the form of ready-made solutions, although not the fastest of the announced.
This is practically the only board on the market that can be called a mini-computer, not a development board. Because it is made in a miniature form factor, not a shovel, and is enclosed in some kind of body. Quite convenient not only for developers, but quite suitable for enthusiasts who crave compact home computer systems.
By the will of fate, the fee itself fell into my hands before the official shipment began.
So in this topic I’ll write quite briefly exactly what you don’t read in the official announcements with which the Internet is filled in, which are listed as technical specifications.
')
Well, so that was, from what to make a start, it is necessary to start all the same with those very technical characteristics.
The board is made on the Samsung Exynos 5 Octa chipset, processor 5410. Not the newest, but goodness, how many can already be. You and the Playstation 4 probably have been waiting for almost a year since the day of the announcement, without buying the Playstation 3, because "it is outdated." I'm right? 5410 is a processor that is used in the Galaxy S4 and the newest of the available on the market in the form of ready-made solutions, although not the newest of the announced. I assure you, foolishness is enough for everyone.
The main feature of the processor is big.LITTLE technology: when the four Cortex-A15 cores are assisted by four Cortex-A7 cores. When there is nothing really to consider, the slower A7s work, when there is something to count - A15. This is possible by two methods depending on the software implementation: or 8 processors - 4 slower, 4 faster; or 4 processors - each of which is a combination of two A15-A7, which can work either slowly or quickly.
Videochip - PowerVR SGX544MP3, a three-processor monster from Imagination Technologies.
2 gigabytes of memory.
USB 3.0 as Micro-A connector, 4 regular USB 2.0
10/100 Ethernet
there is no built-in Wi-Fi
eMMC connector
SD / MMC slot
In general, the set is standard. Read more precisely what can be there on the site .
First of all, the board is very small. Photos do not convey this. I was ready for the fact that it had already become larger compared to the tiny ODROID-U2, and was rather surprised by its real parameters. Most likely, this is because of the fan. Looking at the photos, we imagine it quite massive and scale the entire board around it. But in fact, it is just tiny, and the board around it is quite small. Just look at the Ethernet connector and compare it with the fan. It is really only 4 times the size of a regular RJ-45 connector and is very short. So a shawl, though more than ODROID-U2, but not much. Literally, less than a centimeter wider than the Raspberry Pi. All the elements on the board are so tiny that there is no chance of poking around with your fingers, except to ask a friend. I had to disconnect the eMMC and switch the jumpers with a toothpick.
Regarding the fan - it certainly makes a noise, but not loudly. Plus, the core has speed control both under Linux and under android. And without load, and this is quite often, it stops completely. The control is quite sharp and accurate, not like on my MacBook, which can buzz a few minutes after the temperature has long fallen. The fan instantly starts at load and instantly stops when it falls. It can turn on literally for a second if you run grep on a large file, and instantly turn off after executing the command.
The company Hardkernel officially focused mainly on Android. My card came with android on the memory card. As far as I looked at him, everything was immediately ready there and quite harmoniously: video drivers, codecs - everything works, everything is present. Do not worry that the release of something will be missed, everything is already there. I'm not a big fan of android, so I looked at it all with one eye.
As for Linux, it is a priority for me, and everything is more complicated. The image of xubuntu appeared about yesterday. The kernel is actually a re-assembly of an android with a bunch of options specific to android: all sorts of memory controls and sheduler, all this still requires a lot of work. But the work is just underway. At least now Linux is already loading and running. I'm not at all sure how things will be with GLES, since PowerVR is known for its zhlobskoy position regarding drivers for Linux. But reverse engineering and writing free drivers is a priority for the Open Source community because of the widespread adoption of these chips, so there is hope. About, in fact, the very output on the screen - it works, X11 on the TV is present. Most likely, while without 3D acceleration, I have not tested it yet. At a minimum, I can say that there is enough foolishness in this board and draw as it should without drivers.
Now to the most popular topics of the forum wars. The so-called bug CCI-400, about which someone must rush to tell everyone the truth.
The bug lies in the fact that the Samsung Nakosyachil with the implementation of the interaction of the A7 and A15 cores working in pairs. And just locked this function. Problem Solved.
This leads to the fact that with a small load, the A7 core operates at a slow speed. With an average, but reasonable, works A7 at full 1.2 gigahertz. But with a large load it should have looked like the joint work of A7 and A15, but since the connection between them is locked, the entire task is transferred exclusively to A15.
What this leads to:
For a start, forget about the power point, we are not on the phone, we have a device connected to the network. Suboptimal power consumption? Hello ???
The second point is a bit more complicated, cutting off the A7 core, the processor loses its performance, but the A7 refers to A15 approximately as 10-15%, so you lose so much. This is not much of the fool that is sealed in this monster. You should not be a bore.
Plus, there is a delay in the transfer of the task, you have to reset the registers. This is nanoseconds, honestly, it is not noticeable, even if the 1080p video goes to full screen.
Actually, this is all that occurred to me that I would be interested if I wanted to learn more about this board.
I apologize for the screen taken from offsite: I am writing from work, pay at home, I wanted to post an article as soon as possible.
I am pleased to answer your questions in the comments, perhaps I will add an article, for any reason, except the android. I already took it from the flash drive, and the official image has not yet been posted anywhere, so it is irrevocable until mid-September.
UPDATES:
At the request of enclis :
On the reverse side of the board there is a FPC connector to which the MIPI DSI is connected to connect an external display.
The board is built around the Exynos5 Octa Cortex-A15 with big.LITTLE technology. This is the fastest SoC at the moment available in the form of ready-made solutions, although not the fastest of the announced.
This is practically the only board on the market that can be called a mini-computer, not a development board. Because it is made in a miniature form factor, not a shovel, and is enclosed in some kind of body. Quite convenient not only for developers, but quite suitable for enthusiasts who crave compact home computer systems.
By the will of fate, the fee itself fell into my hands before the official shipment began.
So in this topic I’ll write quite briefly exactly what you don’t read in the official announcements with which the Internet is filled in, which are listed as technical specifications.
')
Well, so that was, from what to make a start, it is necessary to start all the same with those very technical characteristics.
The board is made on the Samsung Exynos 5 Octa chipset, processor 5410. Not the newest, but goodness, how many can already be. You and the Playstation 4 probably have been waiting for almost a year since the day of the announcement, without buying the Playstation 3, because "it is outdated." I'm right? 5410 is a processor that is used in the Galaxy S4 and the newest of the available on the market in the form of ready-made solutions, although not the newest of the announced. I assure you, foolishness is enough for everyone.
The main feature of the processor is big.LITTLE technology: when the four Cortex-A15 cores are assisted by four Cortex-A7 cores. When there is nothing really to consider, the slower A7s work, when there is something to count - A15. This is possible by two methods depending on the software implementation: or 8 processors - 4 slower, 4 faster; or 4 processors - each of which is a combination of two A15-A7, which can work either slowly or quickly.
Videochip - PowerVR SGX544MP3, a three-processor monster from Imagination Technologies.
2 gigabytes of memory.
USB 3.0 as Micro-A connector, 4 regular USB 2.0
10/100 Ethernet
there is no built-in Wi-Fi
eMMC connector
SD / MMC slot
In general, the set is standard. Read more precisely what can be there on the site .
Now to the main part: what you do not read in the announcements.
First of all, the board is very small. Photos do not convey this. I was ready for the fact that it had already become larger compared to the tiny ODROID-U2, and was rather surprised by its real parameters. Most likely, this is because of the fan. Looking at the photos, we imagine it quite massive and scale the entire board around it. But in fact, it is just tiny, and the board around it is quite small. Just look at the Ethernet connector and compare it with the fan. It is really only 4 times the size of a regular RJ-45 connector and is very short. So a shawl, though more than ODROID-U2, but not much. Literally, less than a centimeter wider than the Raspberry Pi. All the elements on the board are so tiny that there is no chance of poking around with your fingers, except to ask a friend. I had to disconnect the eMMC and switch the jumpers with a toothpick.
Regarding the fan - it certainly makes a noise, but not loudly. Plus, the core has speed control both under Linux and under android. And without load, and this is quite often, it stops completely. The control is quite sharp and accurate, not like on my MacBook, which can buzz a few minutes after the temperature has long fallen. The fan instantly starts at load and instantly stops when it falls. It can turn on literally for a second if you run grep on a large file, and instantly turn off after executing the command.
About software.
The company Hardkernel officially focused mainly on Android. My card came with android on the memory card. As far as I looked at him, everything was immediately ready there and quite harmoniously: video drivers, codecs - everything works, everything is present. Do not worry that the release of something will be missed, everything is already there. I'm not a big fan of android, so I looked at it all with one eye.
As for Linux, it is a priority for me, and everything is more complicated. The image of xubuntu appeared about yesterday. The kernel is actually a re-assembly of an android with a bunch of options specific to android: all sorts of memory controls and sheduler, all this still requires a lot of work. But the work is just underway. At least now Linux is already loading and running. I'm not at all sure how things will be with GLES, since PowerVR is known for its zhlobskoy position regarding drivers for Linux. But reverse engineering and writing free drivers is a priority for the Open Source community because of the widespread adoption of these chips, so there is hope. About, in fact, the very output on the screen - it works, X11 on the TV is present. Most likely, while without 3D acceleration, I have not tested it yet. At a minimum, I can say that there is enough foolishness in this board and draw as it should without drivers.
About hard.
Now to the most popular topics of the forum wars. The so-called bug CCI-400, about which someone must rush to tell everyone the truth.
The bug lies in the fact that the Samsung Nakosyachil with the implementation of the interaction of the A7 and A15 cores working in pairs. And just locked this function. Problem Solved.
This leads to the fact that with a small load, the A7 core operates at a slow speed. With an average, but reasonable, works A7 at full 1.2 gigahertz. But with a large load it should have looked like the joint work of A7 and A15, but since the connection between them is locked, the entire task is transferred exclusively to A15.
What this leads to:
For a start, forget about the power point, we are not on the phone, we have a device connected to the network. Suboptimal power consumption? Hello ???
The second point is a bit more complicated, cutting off the A7 core, the processor loses its performance, but the A7 refers to A15 approximately as 10-15%, so you lose so much. This is not much of the fool that is sealed in this monster. You should not be a bore.
Plus, there is a delay in the transfer of the task, you have to reset the registers. This is nanoseconds, honestly, it is not noticeable, even if the 1080p video goes to full screen.
Finally.
Actually, this is all that occurred to me that I would be interested if I wanted to learn more about this board.
I apologize for the screen taken from offsite: I am writing from work, pay at home, I wanted to post an article as soon as possible.
I am pleased to answer your questions in the comments, perhaps I will add an article, for any reason, except the android. I already took it from the flash drive, and the official image has not yet been posted anywhere, so it is irrevocable until mid-September.
UPDATES:
At the request of enclis :
On the reverse side of the board there is a FPC connector to which the MIPI DSI is connected to connect an external display.
UnixBench result:
================================================= ====================== BYTE UNIX Benchmarks (Version 5.1.3) System: odroid: GNU / Linux OS: GNU / Linux - 3.4.5-odroidxu - # 3 SMP PREEMPT Mon Aug 19 07:32:01 CEST 2013 Machine: armv7l (armv7l) Language: en_US.utf8 (charmap = "UTF-8", collate = "UTF-8") 17:38:33 up 25 min, 3 users, load average: 0.66, 1.01, 1.16; runlevel 2 ------------------------------------------------ ------------------------ Benchmark Run: Mon Aug 19 2013 17:38:33 - 18:07:05 0 CPUs in system; running 1 parallel copy of tests Dhrystone 2 using register variables 8022186.3 lps (10.0 s, 7 samples) Double-Precision Whetstone 745.3 MWIPS (10.0 s, 7 samples) Execl Throughput 691.4 lps (29.7 s, 2 samples) File Copy 1024 bufsize 2000 maxblocks 161064.0 KBps (30.0 s, 2 samples) File Copy 256 bufsize 500 maxblocks 47219.8 KBps (30.0 s, 2 samples) File Copy 4096 bufsize 8000 maxblocks 468665.9 KBps (30.0 s, 2 samples) Pipe Throughput 248652.0 lps (10.0 s, 7 samples) Pipe-based Context Switching 51520.2 lps (10.0 s, 7 samples) Process Creation 2104.0 lps (30.0 s, 2 samples) Shell Scripts (1 concurrent) 2437.1 lpm (60.0 s, 2 samples) Shell Scripts (8 concurrent) 860.0 lpm (60.1 s, 2 samples) System Call Overhead 579711.5 lps (10.0 s, 7 samples) System Benchmarks Index Using the System Result INDEX Dhrystone 2 using the system 116700.0 8022186.3 687.4 Double-Precision Whetstone 55.0 745.3 135.5 Execl Throughput 43.0 691.4 160.8 File Copy 1024 bufsize 2000 maxblots 3960.0 161064.0 406.7 File Copy 256 bufsize 500 maxblocks 1655.0 47219.8 285.3 File Copy 4096 bufsize 8000 maxblocks 5800.0 468665.9 808.0 Pipe Throughput 12440.0 248652.0 199.9 Pipe-based Context Switching 4000.0 51520.2 128.8 Process Creation 126.0 2104.0 167.0 Shell Scripts (1 concurrent) 42.4 2437.1 574.8 Shell Scripts (8 concurrent) 6.0 860.0 1433.3 System Call Overhead 15000.0 579711.5 386.5 ======== System Benchmarks Index Score 333.9
Geek Porn:
Source: https://habr.com/ru/post/190540/
All Articles