The Next ODROID!

Moderators: odroid, mdrjr

Re: The Next ODROID!

Unread postby rooted » Fri Feb 09, 2018 9:35 am

To you lucky few who are receiving the N1, what is the first thing you will do after unboxing?

Personally I am going to compile various software to gauge general performance then I think I will try to get Android TV (Lineage) running on it for fun. I will also see how it fares as a desktop replacement for a solid week.
User avatar
rooted
 
Posts: 5902
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english

Re: The Next ODROID!

Unread postby phaseshifter » Fri Feb 09, 2018 9:39 am

EDIT: the first thing i will most likely do is make an image of the emmc
as i am not shure weather there will be anywhere to download a copy of it ..theoretically..there should be
but us the chosen few may be limited as to what resources are actually at hand...

i`d like to get "k.d.e" running on it but i dont think it will..i will also try out my vu7 and see how that goes..maybe even try my vu8c

try hooking up the weather board to it etc..what ever else` i find to hook to it
may apply 12 volts to the power cable of the sata..1 cable only see if i can hook in an optical drive...

hook up the spdif and see if/how it gets audio out if it may need software to enable it or not..??

try streaming..etc install some other browser`s see how they go...

time permitting...
Last edited by phaseshifter on Mon Feb 12, 2018 4:10 pm, edited 1 time in total.
Build It And They Will Come...Be Bold And Mighty Forces Will Come To Your Aid..!!!
phaseshifter
 
Posts: 3000
Joined: Fri May 08, 2015 9:12 am
languages_spoken: english
ODROIDs: N-1..c1+ ..c-2..xu3 lite,xu4...u-3..
u-2...other odroid acc`s as well

Re: The Next ODROID!

Unread postby Molorius » Fri Feb 09, 2018 11:39 am

Just curious, will the N1 have a boot select switch like some of the previous Odroids?
Molorius
 
Posts: 39
Joined: Tue Mar 01, 2016 4:40 am
languages_spoken: English
ODROIDs: C1+, C2, GO

Re: The Next ODROID!

Unread postby campbell » Fri Feb 09, 2018 12:06 pm

rooted wrote:To you lucky few who are receiving the N1, what is the first thing you will do after unboxing?

I very much hope that someone who gets one of the engineering samples can measure the low end of idle power draw (the figures posted by odroid earlier show about 3 watts, but I strongly suspect that was with the "performance" governor, which causes the idle power draw of the XU4 to jump from 1.5 to around 2.7 watts). Then start shutting off the USB bus, throttling down the RAM, setting the ethernet to 100 instead of Gbit, &c. like we've explored with the C2.
campbell
 
Posts: 309
Joined: Thu Sep 03, 2015 1:13 pm
languages_spoken: english
ODROIDs: XU3, XU4, C2, C1+, Cloudshell, Smart Power

Re: The Next ODROID!

Unread postby elatllat » Fri Feb 09, 2018 12:45 pm

rooted wrote:To you lucky few who are receiving the N1, what is the first thing you will do after unboxing?


Cackle with glee, I mean test as a NAS(maybe new kernel) then as a (fedora and/or arch) desktop (Debian eclipse was abandoned), answer questions from those who don't have one yet.
elatllat
 
Posts: 1097
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4, N1

Re: The Next ODROID!

Unread postby rooted » Fri Feb 09, 2018 12:57 pm

campbell wrote:
rooted wrote:To you lucky few who are receiving the N1, what is the first thing you will do after unboxing?

I very much hope that someone who gets one of the engineering samples can measure the low end of idle power draw (the figures posted by odroid earlier show about 3 watts, but I strongly suspect that was with the "performance" governor, which causes the idle power draw of the XU4 to jump from 1.5 to around 2.7 watts). Then start shutting off the USB bus, throttling down the RAM, setting the ethernet to 100 instead of Gbit, &c. like we've explored with the C2.


I'm guessing most of us will be willing to test features and statistics for others. I certainly am.
User avatar
rooted
 
Posts: 5902
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english

Re: The Next ODROID!

Unread postby ksubox » Fri Feb 09, 2018 2:12 pm

Exactly what I need for my home/work !!!
I needed mini servers with fault tolerant file system (at least 2 HD), 1G network, 4G memory for my application servers and a little bit faster than XU4.
Also OpenCL for processing is a big plus for me.
And now it's here!
Thank you, Hardkernel for this New Year gift :)
If you will not have delays due to hardware problems when we could expect devices available ?
ksubox
 
Posts: 33
Joined: Wed Jan 15, 2014 12:42 pm
languages_spoken: english
ODROIDs: ODROID-X2, X4, C2

Re: The Next ODROID!

Unread postby mad_ady » Fri Feb 09, 2018 3:57 pm

I too can take requests to test things or run programs on the N1.
I will try to run my scripts/programs and update them to support N1 if needed. Also, I plan on testing the USB ports - for instance to run the Odroid Webcam on the USB2/3 ports with a lowspeed/highspeed device in parallel. Also, under load. I want to see if I get corrupt frames, the way we get in odroid C2 (because of the USB port controller). I hope everything will be smooth.
I'm also interested in SPI boot flash, but I'll start a new thread to discuss it and its limitations.
Edit: here's the SPI thread: viewtopic.php?f=29&t=29976
User avatar
mad_ady
 
Posts: 4927
Joined: Wed Jul 15, 2015 5:00 pm
Location: Bucharest, Romania
languages_spoken: english
ODROIDs: XU4, C1+, C2, N1

Re: The Next ODROID!

Unread postby memeka » Fri Feb 09, 2018 5:21 pm

Finally the N1 is announced and I’m on holiday on the place with the worst internet in the world :)

First, the bad: so sorry to see the pcie issues. In theory, you could get 4 full speed SATA connections using all 4 lanes.
My opinion - should use the 2 lanes bridge if possible, it’s really a shame to use a 1 lane bridge on a 4 lane interface...
Also SPI flash would be great, to be able to boot off SATA...

Now, about the N1-lite: should be really cost effective. With 2GB ram, no SATA, no SPI flash - maybe even the layout of the XU4 if possible. Compared to XU4, you would still get 64bit, more efficienct network (less cpu usage), 2x native usb3 (instead of 1 with hub), better GPU, better VPU, 4K output.

Although everyone is split over wifi, it would be nice in the future - if N1 is successful - for HK to consider having a wifi+bt onboard (2x2 mimo N or AC) - not a cheap crappy halfN one, but there are people using wifi, and would appreciate having USB3 host available since the 3399 already has a fast interface for wifi that is not used in the N1.
Also they might consider a version with 3399Pro for AI, I think in applications like drones it would be very useful.
User avatar
memeka
 
Posts: 4063
Joined: Mon May 20, 2013 10:22 am
languages_spoken: english
ODROIDs: XU rev2 + eMMC + UART
U3 + eMMC + IO Shield + UART

Re: The Next ODROID!

Unread postby rooted » Fri Feb 09, 2018 5:53 pm

memeka wrote:Finally the N1 is announced and I’m on holiday on the place with the worst internet in the world :)


I was wondering where you were.

Interesting thought about the lite version taking on the shape of the XU4.
User avatar
rooted
 
Posts: 5902
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english

Re: The Next ODROID!

Unread postby crashoverride » Fri Feb 09, 2018 6:04 pm

memeka wrote:Although everyone is split over wifi, it would be nice in the future

If it were up to me, would add a "N1+Wifi" option to the product page which includes a WiFi 0 module stuck in a USB2 port for you! :lol:

rooted wrote:Interesting thought about the lite version taking on the shape of the XU4.

The RK3399 is a very large chip. I do not think it would be possible to offer it an XU4 form factor. All RK3399 boards are quite large. I think N1 is currently the smallest of them.
crashoverride
 
Posts: 4139
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: The Next ODROID!

Unread postby rooted » Fri Feb 09, 2018 6:20 pm

crashoverride wrote:
memeka wrote:Although everyone is split over wifi, it would be nice in the future

If it were up to me, would add a "N1+Wifi" option to the product page which includes a WiFi 0 module stuck in a USB2 port for you! [emoji38] .


I literally lol'd so there's that :)

What do you think of an SDIO option, unpopulated by default?
User avatar
rooted
 
Posts: 5902
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english

Re: The Next ODROID!

Unread postby crashoverride » Fri Feb 09, 2018 6:32 pm

rooted wrote:What do you think of an SDIO option, unpopulated by default?

I think a M.2 slot would be needed for fast wifi. SATA and USB2/3 would also need to be routed to it so you would still "lose" one of each connector. Otherwise, you end up with a "made up standard" connector where only certain things work.

Alternatively, to do a "module" interface, a custom connector and custom wifi boards will be needed and a USB3 port dedicated to it. You still "lose" as USB3 port, and face limited part selection since its all custom. While a SDIO+UART connection is commonly found, I do not think its performance is up to "best of class" wifi AC.

All the above considered, USB2/3 is still the best option. Parts are plentiful, low cost, and diverse enough to meet anyone's needs. It also does not add any cost to the board for those not using wifi.
crashoverride
 
Posts: 4139
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: The Next ODROID!

Unread postby tkaiser » Fri Feb 09, 2018 6:40 pm

crashoverride wrote:I think N1 is currently the smallest of them.


Nope: https://www.96rocks.com

But this is the 'makes no sense at all' 96boards form factor without Ethernet (needs another chip like RTL8211 and a lot of passive components so board size would have to increase a lot), you get 'nice' thermal problems with such a small board and to make use of PCIe there you need to attach something to the M.2 connector. They said they routed also RGMII data lines to the connector in a non standards compliant way so to get native Ethernet and SATA there both price and size will at least double :)

For a board that also allows for semi professional storage in addition to what the RK3399 has been made for (Android stuff, gaming and displaying video) the N1 is really not too bad when looking at size, consumption and overall costs (talking about a 2GB base variant and making the 4GB version a higher priced 'pro' version for those users who need maximum amount of DRAM or believe so)

@memeka: wrt PCIe and SATA let's wait and see. Wouldn't be the first time that an IP block that is inside a SoC functions well on its own but suffers from slow interconnects. First tests with NVMe on another ROCK3399 device didn't look that promising but it's way too early for any conclusions. Also I learned just recently that an ASM1062 almost doubles the SATA costs and requires a slightly different PCB compared to single lane ASM1061 so maybe we should already switch into 'use case first' mode with SATA and ODROID-N1.

2 HDD won't be bottlenecked by ASM1061 and having PCIe attached SATA will avoid so many hassles users now face with USB3 attached storage on the XU4. But there's a downside as usual the average user is not aware of: Internal SATA connectors are rated for 50 matings maximum while USB3 connectors or eg. eSATA are good for 5,000 mating cycles. I hope Hardkernel is able to address this issue by educating users and providing inexpensive but high quality cable and PSU sets (at least two: 12V/2A and 12V/4A)
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby memeka » Fri Feb 09, 2018 6:48 pm

@tkaiser even if 1061 is cheaper and better when using ssd, 1062 might be safer - if RK changed the data sheet to say Gen1, maybe Gen2 is not stable and on long term might be more hassle for support.
User avatar
memeka
 
Posts: 4063
Joined: Mon May 20, 2013 10:22 am
languages_spoken: english
ODROIDs: XU rev2 + eMMC + UART
U3 + eMMC + IO Shield + UART

Re: The Next ODROID!

Unread postby crashoverride » Fri Feb 09, 2018 6:50 pm

So I took a look a the popular AP6356 SDIO wifi chip and it states its interface speeds as:
"The module supports SDIO version 3.0 for
all 1.8V 4-bit UHSI speeds: SDR50(100
Mbps),SDR104(208MHz) and DDR50(50
MHz, dual rates) in additi
on to the 3.3V default
speed(25MHz) and high speed (50 MHz)."


So its pretty much the same as a "fast SD card" in performance.
crashoverride
 
Posts: 4139
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: The Next ODROID!

Unread postby tkaiser » Fri Feb 09, 2018 6:54 pm

crashoverride wrote:
rooted wrote:What do you think of an SDIO option, unpopulated by default?

I think a M.2 slot would be needed for fast wifi. SATA and USB2/3 would also need to be routed to it


How to route SATA to an M.2 slot when the SoC doesn't feature SATA? :)

SATA would've been either USB3 (not a great idea in this case) or PCIe attached so to be able to route SATA to an M.2 connector you need to put a PCIe SATA chip on the board anyway eating at least one lane but I've no idea whether/how's that possible with RK3399 without stuff like a PCIe multiplexer to split up the different lanes (me being a total PCIe noob wrt all the lower layers).

If Hardkernel would have chosen RTD1295 (has both SATA and PCIe) something like this would've been possible but with RK3399 without a lot of additional costs and PCB space it's simple not possible.

SDIO could work and there are modules out there that would fit (simply look at the other RK3399 devices and the Ampak modules used there) but again this would increase costs not only due to BOM but certifications (which would also increase prices a lot in certain areas of the world like Japan and Brazil who chose to establish their own wireless certification procedures. Or no ODROID-N1 in those countries).

The N1 has 4 USB ports. Why not just using them? 802.11n/ac RTL881xAU modules without MIMO, 2x2 or even 4x4 exist.
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby crashoverride » Fri Feb 09, 2018 6:56 pm

tkaiser wrote:How to route SATA to an M.2 slot when the SoC doesn't feature SATA?

N1 has 2 SATA ports. Once would need to be routed to the M.2 connector. This is what I meant by "losing one of each port".
crashoverride
 
Posts: 4139
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: The Next ODROID!

Unread postby rooted » Fri Feb 09, 2018 7:02 pm

crashoverride wrote:So I took a look a the popular AP6356 SDIO wifi chip and it states its interface speeds as:
"The module supports SDIO version 3.0 for
all 1.8V 4-bit UHSI speeds: SDR50(100
Mbps),SDR104(208MHz) and DDR50(50
MHz, dual rates) in additi
on to the 3.3V default
speed(25MHz) and high speed (50 MHz)."


So its pretty much the same as a "fast SD card" in performance.
I couldn't find a price on this but it seems like a nice little package.

Personally I'm fine with a USB3 AC dongle sticking out the side :)
User avatar
rooted
 
Posts: 5902
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english

Re: The Next ODROID!

Unread postby neph » Fri Feb 09, 2018 7:05 pm

Very nice product on paper, I can't wait to test this board :)
neph
 
Posts: 1
Joined: Fri Feb 09, 2018 7:00 pm
languages_spoken: english, french
ODROIDs: C2

Re: The Next ODROID!

Unread postby tkaiser » Fri Feb 09, 2018 7:19 pm

memeka wrote:@tkaiser even if 1061 is cheaper and better when using ssd


No, not 'better'. ASM1061 is limited by its single lane PCIe interface (I would say 390 MB/s maximum possible between disk and host). When you use the ASM1062 connected with 2 lanes to a capable PCIe host port then a single SSD behind now gets slightly faster (now the 6Gbps SATA between the ASMedia chip and the SSD become the bottleneck but only if the SSD itself is not already the bottleneck which is what most probably 99% of N1 users will face). Switching to ASM1062 would only mean that 'power users' are able to use 2 fast SSDs without being bottlenecked too much but only given that RK3399 itself is not the bottleneck (and I still miss a valid use case for this besides generating irrelevant benchmark numbers).

If we run into link training problems or if RK tells us to not take Gen2 granted then this needs to be revisited. But it's a bit too early and based on the experiences of the past Rockchip is very cooperative and providing in depth answers to technical questions. Dealing with some USB3 problems last year on ROCK64 I reported something, got an email later confirming the behaviour and explaining a lot of stuff I'm way too dumb to understand. See this attachment from an USB3 logic analyzer the engineer sent for my understanding ;)

Image
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby moon.linux » Fri Feb 09, 2018 7:48 pm

Looks like the USB3 logic analyzer seen to be handy too to understand the USB 3.0 internal. Might be use full for XU4.
moon.linux
 
Posts: 1161
Joined: Thu Oct 02, 2014 11:42 pm
languages_spoken: english

Re: The Next ODROID!

Unread postby tkaiser » Fri Feb 09, 2018 8:22 pm

crashoverride wrote:
tkaiser wrote:How to route SATA to an M.2 slot when the SoC doesn't feature SATA?

N1 has 2 SATA ports. Once would need to be routed to the M.2 connector.


Sure, but these SATA ports are behind a PCIe chipset so to be able to also route at least 2 PCIe lanes to the M.2 connector the RK3399 PCIe host controller would need to be configured as 2 x x2 but AFAIK it's a 1 x x4 only design? But I might be wrong, total PCIe noob wrt anything below the kernel layer. I've been sent recently to these slides https://pcisig.com/sites/default/files/ ... FROZEN.pdf and I came back assured I know close to nothing below the application layer (yet) :)
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby rooted » Fri Feb 09, 2018 8:27 pm

tkaiser wrote: I've been sent recently to these slides https://pcisig.com/sites/default/files/ ... FROZEN.pdf and I came back assured I know close to nothing below the application layer (yet) :)


By page 11 I can see I don't know anything, it's why I don't comment on these PCI discussions. It's like someone speaking a foreign language you don't understand but know a few of the words.
User avatar
rooted
 
Posts: 5902
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english

Re: The Next ODROID!

Unread postby tkaiser » Fri Feb 09, 2018 8:43 pm

rooted wrote:
tkaiser wrote: I've been sent recently to these slides https://pcisig.com/sites/default/files/ ... FROZEN.pdf and I came back assured I know close to nothing below the application layer (yet) :)


By page 11 I can see I don't know anything, it's why I don't comment on these PCI discussions.


Yeah, I think we should stop these discussions since it's all just speculation now. We're busy in the meantime testing with NVMe on RockPro64 where all 4 PCIe lanes are available so it should be the best device to test for PCIe limitations. Since I've seen yesterday that the 2 USB3 ports on ROCK3399 share the same IRQ in the meantime I also fear USB3 bandwidth limitations. But let's hold on now and wait for some tests first.

And while I was amongst the first who asked for ASM1062 instead of ASM1061 after a bit more thinking about (and knowing that it's a lot more expensive) I think given the role of ODROID-N1 as a general purpose device ASM1061 is not the worst choice since

* fast enough and not affecting real-world use cases anyway (people who want to attach more than 1 fast SSD should check other more suited architectures first -- with ARM those SoCs that are made for the use case for example, see Marvell ARMADA)
* keeps costs down
* consumption friendly. Those PHYs that work above at Gbps speeds all need quite some energy so the higher link rates and the more lanes are used the higher the consumption. There's power management and stuff but performance always comes at a 'cost' (this is also something to keep in mind when reading about '3W idle' published by Hardkernel, there sits an ASMedia chip on the board with 3 high-speed PHYs that might be able to be powered off when no SATA needs to be used)
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby rooted » Fri Feb 09, 2018 9:03 pm

3W at idle will be a problem for anyone wanting to run from battery, I'm sure this can be lowered with a bit of testing and kernel hackery.
User avatar
rooted
 
Posts: 5902
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english

Re: The Next ODROID!

Unread postby Raybuntu » Fri Feb 09, 2018 9:08 pm

campbell wrote:
rooted wrote:To you lucky few who are receiving the N1, what is the first thing you will do after unboxing?

I very much hope that someone who gets one of the engineering samples can measure the low end of idle power draw (the figures posted by odroid earlier show about 3 watts, but I strongly suspect that was with the "performance" governor, which causes the idle power draw of the XU4 to jump from 1.5 to around 2.7 watts). Then start shutting off the USB bus, throttling down the RAM, setting the ethernet to 100 instead of Gbit, &c. like we've explored with the C2.


I'll create a LibreELEC build. I will do some power consumption measurements. I have a very accurate power meter and I've done some measurements for the C2 before.
BTC: 12zLUYC7JzwM7a8cQKekAvZr9kWxjzTfxm
Raybuntu
 
Posts: 1245
Joined: Mon Nov 30, 2015 4:23 pm
languages_spoken: english, german
ODROIDs: C1+, C2

Re: The Next ODROID!

Unread postby tkaiser » Fri Feb 09, 2018 10:12 pm

rooted wrote:3W at idle will be a problem for anyone wanting to run from battery, I'm sure this can be lowered with a bit of testing and kernel hackery.


According to viewtopic.php?f=29&t=29932&start=150#p213822 they used ondemand governor. Without HDD 3.2W and with spinning but idle HDD attached 5.3W (lots of fluctuations -- I usually measure with averaged values).

Measuring with an ASM1062 mPCIe card connected to a single lane each used PHY (regardless whether PCIe or SATA) adds ~1W: http://forum.solid-run.com/hardware-f33 ... t3166.html

This is not just a bit 'kernel hackery' but understanding the trade-offs involved. If there's a SATA controller on the PCIe bus this thing needs some juice (switching from Gen2 to Gen1 might be able to reduce consumption -- needs testing). Attaching SATA disks that can be sent to sleep might still result in wasting 1W each since the SATA PHY remains active (and it totally depends on the disk in question how much power it draws even when asked to spin down -- I have several 3.5" disks here that need + 0.5W even if platters do not rotate).

And so on... but on the bright side something nobody mentioned so far: Hardkernel showed us an idle number of 5.3W with this 2.5" HDD while the 'full Samba access through network almost saturating GbE' number is at 8.2W. That's not even a 3W difference between 'doing nothing' and 'full NAS throughput'. With XU4, HC1 or HC2 the difference was a lot higher.
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby nobe » Fri Feb 09, 2018 10:27 pm

how much power would one save by disabling things like HDMI & GPU in DTS file ?
nobe
 
Posts: 121
Joined: Sun Feb 07, 2016 9:52 pm
languages_spoken: english, french
ODROIDs: Odroid-C2

Re: The Next ODROID!

Unread postby rooted » Fri Feb 09, 2018 10:43 pm

tkaiser wrote:Measuring with an ASM1062 mPCIe card connected to a single lane each used PHY (regardless whether PCIe or SATA) adds ~1W: http://forum.solid-run.com/hardware-f33 ... t3166.html
.


This seems like too much, is all that power dedicated to the card or is some wasted on the slot?
User avatar
rooted
 
Posts: 5902
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english

Re: The Next ODROID!

Unread postby tkaiser » Fri Feb 09, 2018 10:44 pm

nobe wrote:how much power would one save by disabling things like HDMI & GPU in DTS file ?


Will also depend on the drivers used. For example two years ago on those cheap Allwinner H3 thingies 10 minutes after a cold boot the SoC temperature decreased by 3°C and consumption was reduced by ~200mW. But only since I ran these devices headless all the times. It was just the HDMI driver timing out and disabling the respective IP block. Next smelly Allwinner kernel changed behaviour (no drop even if HDMI disconnected) and how it's done with mainline I don't know.

Since RK3399 is used in Chromebooks at least Rockchip's BSP kernels should do similar things to save as much energy as possible. But needs testing of course :)
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby tkaiser » Fri Feb 09, 2018 11:02 pm

rooted wrote:
tkaiser wrote:Measuring with an ASM1062 mPCIe card connected to a single lane each used PHY (regardless whether PCIe or SATA) adds ~1W: http://forum.solid-run.com/hardware-f33 ... t3166.html
.


This seems like too much, is all that power dedicated to the card or is some wasted on the slot?


These are PHYs (lowest physical layer, those things that establish links at the electrical layer). As written before PHYs that operate at Gbps speeds and especially above consume quite a lot, just look at the picture I posted here in the thread for the MacchiatoBin: viewtopic.php?f=29&t=29932&start=50#p213659

The 2 smaller heatsinks sit on the 10GBASE-T Ethernet PHYs since those not only consume a lot of energy but get also hot like hell. The MAC implementation (dealing with the stuff above the physical layer) sits inside the SoC just as with RK3399 but what really needs juice is maintaining the physical layer. Please remember we're talking here about 6 Gbps for SATA III and 5 GT/s with PCIe Gen2. And since the ASM1061 is connected through PCIe to the SoC there's of course also a PCIe PHY involved (would be different if RK3399 would be capable of 'native SATA').

But as already said: I'm a hardware noob, just citing some experiences and numbers. Anyway: wanting as low consumption as possible and having PCIe attached SATA on the same device is a little bit conflicting if there's no way to totally disable the ASM1061 (which probably is rather simple) and cutting power to connected disks (to get rid of SATA links that need PHYs active on both sides)
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby ard » Fri Feb 09, 2018 11:33 pm

moon.linux wrote:Looks like the USB3 logic analyzer seen to be handy too to understand the USB 3.0 internal. Might be use full for XU4.

But they are fsck expensive...
A 2.0 analyzer at least showed us bugs in the USB2 stack of the official hardkernel odroid c1 kernel release.
And looking at the code and comparing that to a recent upstream DWC2 driver made me cry.
(The most ARM SoC's that I know of have the DWC2 "IP". That's amlogic (c1,c2), exynos (xu*, U*), broadcom (raspberry pi), but not my intel nas, and except the exynos and maybe the PI at 4.14, they all have their own drivers, probably originated from an original very old exynos dwc2 driver).

The problem with the Xu4 however is that it is already using the upstream DWC3 driver, which is common on almost all platforms (pc's, arm...).
Otoh: that means that the problem must be somewhere in the glue code. Whatever the problem is.
ard
 
Posts: 70
Joined: Tue Jul 09, 2013 2:12 am
languages_spoken: english, dutch, german
ODROIDs: ODROID-U2

Re: The Next ODROID!

Unread postby elatllat » Fri Feb 09, 2018 11:41 pm

tkaiser wrote:...wanting as low consumption as possible and having PCIe attached SATA on the same device is a little bit conflicting...

Well said.
If you want a low power device it should have wifi instead of ethernet, and should implement android power management.
elatllat
 
Posts: 1097
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4, N1

Re: The Next ODROID!

Unread postby campbell » Fri Feb 09, 2018 11:54 pm

elatllat wrote:
tkaiser wrote:...wanting as low consumption as possible and having PCIe attached SATA on the same device is a little bit conflicting...

Well said.
If you want a low power device it should have wifi instead of ethernet, and should implement android power management.

Software selected (ethtool) 10baseT Ethernet draws barely any power (50 mW) and WiFi is a no go for certain applications/environments.

If this board really draws 3 W at idle with the ondemand governor and no peripherals, that’s a significant regression from the last generation of boards, and it’d be good to know why (and whether anything can be done about it).

Ideally, I should be able to, from in software, power up the SATA bus, read/write a bunch of data, and then power it back down.
campbell
 
Posts: 309
Joined: Thu Sep 03, 2015 1:13 pm
languages_spoken: english
ODROIDs: XU3, XU4, C2, C1+, Cloudshell, Smart Power

Re: The Next ODROID!

Unread postby crashoverride » Sat Feb 10, 2018 12:09 am

campbell wrote:If this board really draws 3 W at idle with the ondemand governor and no peripherals, that’s a significant regression from the last generation of boards, and it’d be good to know why (and whether anything can be done about it).

Watts = Volts * Amps;

3W = 5V * 0.6A = 600ma
3W = 12V * 0.25A = 250ma
crashoverride
 
Posts: 4139
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: The Next ODROID!

Unread postby tkaiser » Sat Feb 10, 2018 12:23 am

campbell wrote:Ideally, I should be able to, from in software, power up the SATA bus, read/write a bunch of data, and then power it back down.


Ok, I start to understand that users here want to believe they're talking about 'native' SATA where the host establishes a direct SATA connection to a disk. It's not like that but there's a controller in between:
Code: Select all
  Disk    6 Gbps         ASM1061       5 GT/s   RK3399
SATA PHY <------> SATA PHY | PCIe PHY <------> PCIe PHY


4 PHY in total, two links to maintain. Both SATA and PCIe specs define power management standards, this needs also driver support, sometimes this is missing, sometimes there are bugs preventing to use PM even if available -- see this recent Intel example: https://www.phoronix.com/scan.php?page= ... bata-power

So why not simply waiting 2 or 3 weeks and checking then again? If the ASM1061 on the board (old design, most probably in a very inefficient process made) establishes a PCIe link with the SoC then I wouldn't be surprised if that's responsible for more than 1W idle consumption anyway. So let's wait what we find out, maybe it's sufficient to toggle this from user space (telling the RK3399 to disable the PHY), maybe it needs a little board modification to cut power to the chip through a GPIO. Who knows (yet)?

Then: Hardkernel measured with 12V, there's a wide voltage range DC-DC converter on the board to generate stable 5V which has not a 100% efficiency but most probably only +90% (so there's some more consumption and a bit more heat generated). This all is different to XU4 or any other ODROID so far.

If they make an N1 light saving the PCIe chip, the SATA ports and wide voltage input (DC-DC stuff) allowing again to be powered with 5V then consumption will decrease for sure and all those crappy undervoltage problems clueless users run in when using USB storage are back again.
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby joerg » Sat Feb 10, 2018 12:33 am

rooted wrote:Hopefully just a software issue, if not the drive is still under warranty.


Sadly the motherboard broken and no warranty. It's still > 3 years old, a Asrock Q1900DC passive cooled. Damn.

@tkaiser
I assume you are a moderator of the OMV forum? I saw that there already exists ports to rk3399 sbc. So could it be a short way to port OMV also to the N1?
joerg
 
Posts: 808
Joined: Tue Apr 01, 2014 2:14 am
Location: Germany
languages_spoken: german, english, español
ODROIDs: C1, C1+, C2

Re: The Next ODROID!

Unread postby nobe » Sat Feb 10, 2018 1:09 am

@tkaiser
the way i understand it, when some (= not all) people say "native sata", you need to understand "not usb2 attached sata"
nobe
 
Posts: 121
Joined: Sun Feb 07, 2016 9:52 pm
languages_spoken: english, french
ODROIDs: Odroid-C2

Re: The Next ODROID!

Unread postby campbell » Sat Feb 10, 2018 2:03 am

crashoverride wrote:
campbell wrote:If this board really draws 3 W at idle with the ondemand governor and no peripherals, that’s a significant regression from the last generation of boards, and it’d be good to know why (and whether anything can be done about it).

Watts = Volts * Amps;

3W = 5V * 0.6A = 600ma
3W = 12V * 0.25A = 250ma

Not sure where you're going with this... there's no real reason that power consumption should go up anywhere close to linearly with supply voltage, unless the majority of the power is being dumped by the voltage regular itself, which I highly doubt is the case here. When the C2 is supplied with 4 volts or less (obviously without usb) instead of 5, the power consumption barely drops at all.
campbell
 
Posts: 309
Joined: Thu Sep 03, 2015 1:13 pm
languages_spoken: english
ODROIDs: XU3, XU4, C2, C1+, Cloudshell, Smart Power

Re: The Next ODROID!

Unread postby tkaiser » Sat Feb 10, 2018 2:07 am

joerg wrote:@tkaiser
I assume you are a moderator of the OMV forum? I saw that there already exists ports to rk3399 sbc. So could it be a short way to port OMV also to the N1?


I'm not aware of any OMV images for RK3399 devices yet, at least no official ones (those at https://sourceforge.net/projects/openmediavault/files/ and those from ayufan for ROCK64/RK3328). But there's nothing to be ported anyway since OMV sits just on top of a Debian. And while OMV on ARM needs a bunch of optimized settings to show decent performance the real magic happens one layer below.

With ODROID HC1 and old 3.x kernel plus insufficient settings you get a NAS throughput of ~35MB/s while with all the OMV/Armbian tweaks it's 110MB/s on same hardware. Settings matter, kernel matters, protocols (UAS) matter. And the most important step with RK3399 is checking for PCIe and USB3 performance/reliability with RK's 4.4 BSP vs. mainline kernel and then developing optimized settings (IRQ affinitiy and such stuff).

That said it's easy at least to install an optimized OMV on any Debian: https://forum.openmediavault.org/index. ... post167323 (please see the 2 lines in the install script that may need to be commented few posts below).

But for an official OMV image it would need also a repository where kernel+settings updates can be fetched (that's one of the reason almost all OMV images rely on Armbian) and I think I will provide only device agnostic OMV images in the future suited only for those SBC that implement SPI NOR flash booting.

nobe wrote:@tkaiser
the way i understand it, when some (= not all) people say "native sata", you need to understand "not usb2 attached sata"


Still problematic. If people want to talk about 'fast' SATA (for whatever definition of fast) they should do so. But native would mean it's a CPU/SoC capability and especially consumption is affected. I played around with a 4-port SATA mPCIe card on various boards (EspressoBin and Clearfogs) just to realize that I can add approx. 1W per disk in idle just due to the SATA controller having to maintain links. With native SATA on these boards (Marvell ARMADA) that's no problem since here drive and SATA controller implement power management correctly)
tkaiser
 
Posts: 593
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: The Next ODROID!

Unread postby joerg » Sat Feb 10, 2018 2:25 am

@tkaiser
Thank you for the explanation.
joerg
 
Posts: 808
Joined: Tue Apr 01, 2014 2:14 am
Location: Germany
languages_spoken: german, english, español
ODROIDs: C1, C1+, C2

Re: The Next ODROID!

Unread postby ASword » Sat Feb 10, 2018 3:30 am

How about we talk about these interconnects in more specific terms:

SoC-based SATA
PCIe-based SATA
USB-based SATA

"fast", "native", etc are too vague.

And yes, the more PHY serdes that need to be driven the more power is required. SerDes are very power hungry, Ethernet particularly so because of the 1000' requirement of the standard. Once you're off the SoC, power requirements grow rapidly. Power management on some interconnects allows them to turn off some SerDes when not in use, but this is problematic for serial buses because everything is communicated across 1-2 lines and those need to stay powered. And pretty much everything is serial nowadays. SoC designers have their own tradeoffs, however, so not every device can have all the I/O we want it to and as a result we get things like PCIe and USB bridges... which have SerDes on both sides plus the bridge device itself to power.

The SATA chip chosen by HK sounds like quite a good balance, and moving up to a more capable one not worth the ROI (i.e. the practical benefit of more performance not worth $ and W and other costs). Especially since we don't know the actual source of the bottleneck being measured -- a faster SATA chip might yield nothing extra in terms of performance if the limit is the SoC itself (and I'd really like to know more about the SoC's limits). And the incremental value of each additional MB/s once you're already up at ~400MB/s is dropping fast.
ASword
 
Posts: 124
Joined: Fri Aug 04, 2017 12:48 pm
languages_spoken: english
ODROIDs: XU4, HC1

Re: The Next ODROID!

Unread postby crashoverride » Sat Feb 10, 2018 4:23 am

campbell wrote:Not sure where you're going with this...

I am saying that 3W is "almost nothing" in comparison. (1A = 1000ma.)

USB2 power is 5V * 500ma = 2.5W. USB3 power is 5V * 900ma = 4.5W. The C2 power supply is 5V * 2A = 10W. The XU4 power supply is 5V * 4A = 20W.

I think the 3W number is too low.
crashoverride
 
Posts: 4139
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: The Next ODROID!

Unread postby campbell » Sat Feb 10, 2018 4:42 am

3 watts is not nothing. The XU4 draws barely 1.5 watts at idle, with the ondemand governor and no peripherals plugged in, and Gbit ethernet. Probably draws less if you dial the ethernet back. It draws 2.7 watts at idle with the performance governor, hence my original assumption that the 3 watt number for the N1 must have also been with the performance governor, since odroid says the XU4 and N1 are comparable in terms of power draw.

The C2 draws 1.1 watts at idle with Gbit ethernet using the latest kernel. It draws as little as 900 mW with the ondemand governor and the ethernet speed set to 100baseTX.

The N1 has the same four cortex-A53 cores as the C2, plus two more cores in a big.LITTLE configuration. The whole point of big.LITTLE is that the big cores can be more or less stood down when not needed. So there is NO good reason why the N1 should draw ~3x as much power as the C2 at idle.
campbell
 
Posts: 309
Joined: Thu Sep 03, 2015 1:13 pm
languages_spoken: english
ODROIDs: XU3, XU4, C2, C1+, Cloudshell, Smart Power

Re: The Next ODROID!

Unread postby lunat1k » Sat Feb 10, 2018 6:35 am

Love it.
It's a darn near perfect NAS (once the SATA 12V rail works).
One of the first boards ~ $100 w/ a 64bit memory path and this level of cpu/gpu.
PCI based SATA is fantastic.

I'm not sure how much it would add to the BOM but a second Gbe phy/port would expand it's uses to Router/gateway/firewall/VPN and near ideal Ceph node.
I'll be getting 4 once they're GA.

Also would be interested in tinkering with crypto-mining on the Mali if the Mali OpenCL SDK works on this board.

On another note,
I'm estimating it's a huge undertaking - but how big / expensive would you think it would be to make these Server Base System Architecture (SBSA) / (UEFI 2.4 and ACPI 5.1) compliant for RedHat support?
lunat1k
 
Posts: 1
Joined: Sat Feb 10, 2018 6:21 am
languages_spoken: english
ODROIDs: ODROID-HC2
ODROID-N1

Re: The Next ODROID!

Unread postby freegold » Sat Feb 10, 2018 7:44 am

I'm knida new to the discussion but awed none-the-less. We run a new community low power radio station in Eugene OR (KEPW 97.3fm) and are attempting to do on site live feeds for events broadcast to our downtown station using high speed Ethernet. We thought of doing a project with a Raspberry PI (I know boo hiss) and I then came upon the Next ODROID and was totally impressed. How do we get one to test our project??? Anyway, keep up the great work and hope that the ODROID ships soon!

Rick
freegold
 
Posts: 1
Joined: Sat Feb 10, 2018 7:21 am
languages_spoken: english
ODROIDs: interested in ODROIDs but only worked with pi b+ so far

Re: The Next ODROID!

Unread postby DarkBahamut » Sat Feb 10, 2018 9:26 am

campbell wrote:The N1 has the same four cortex-A53 cores as the C2, plus two more cores in a big.LITTLE configuration. The whole point of big.LITTLE is that the big cores can be more or less stood down when not needed. So there is NO good reason why the N1 should draw ~3x as much power as the C2 at idle.


The extra chips could be the reason as noted above. If things hanging off the PCIe bus aren't doing proper LPM then that will eat power. You can see this on PCs that use PCIe storage solutions. With LPM they idle at ~40mW, but if no LPM is in use then they idle at 1.2W. The same thing could be happening on the N1 depending when the drivers are up to.

Outside of that the CPU arrangement will depend. I don't believe the N1 will be using big.LITTLE, it will almost certainly be using hperf-hmp like the 4.14 kernel on the XU4 does. Since that sends all work to the big cores first, that will increase power consumption since the A53's will remain largely idle while the A72's won't be able to idle as much. It also depends if Rockchip support proper low power modes on the CPU cores like C2 and retention states. Rockchip don't target this at phones, so it possible the power management drivers are lacking as this all requires kernel support to work.

Hopefully the 3W figure can be lowered though. It is quite a bit of power for a ARM board sitting idle and lacking LPM would be a bit bad on the kernel side. It an area to find some improvements I'm sure :)
DarkBahamut
 
Posts: 321
Joined: Tue Jan 19, 2016 10:19 am
languages_spoken: english
ODROIDs: XU4, N1

Re: The Next ODROID!

Unread postby rooted » Sat Feb 10, 2018 9:53 am

crashoverride wrote:
campbell wrote:If this board really draws 3 W at idle with the ondemand governor and no peripherals, that’s a significant regression from the last generation of boards, and it’d be good to know why (and whether anything can be done about it).

Watts = Volts * Amps;

3W = 5V * 0.6A = 600ma
3W = 12V * 0.25A = 250ma


It isn't that bad if it is only 250mAh, the largest reason it matters so much is while running from battery.
User avatar
rooted
 
Posts: 5902
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english

Re: The Next ODROID!

Unread postby madbrain » Sat Feb 10, 2018 4:27 pm

Sounds like a nice board, and I will likely own one at some point.

One thing I would like to see is support for DSD bitstreaming over HDMI.
Even on a significantly more powerful machines, nVidia and AMD have so far not added any support to their drivers for doing so.
This has forced me to continue to use physical discs (SACD and SACD-R, and find a player that plays the later). I would very much like a media player box that can bitstream DSD without using physical discs.
I realize this is probably purely a software issue, but nevertheless, it's something I'd like to see.
madbrain
 
Posts: 33
Joined: Sat Aug 05, 2017 5:40 pm
languages_spoken: english
ODROIDs: XU4

PreviousNext

Return to General Chat

Who is online

Users browsing this forum: No registered users and 0 guests