hardware balancing: user wish list and prices

Moderators: mdrjr, odroid

What hardware or software item would You expect in a pro version?

extended DRAM (there's probably no alternative idea for more than 4GB, so we just keep in mind)
15
36%
upgradable, passive cooling
11
26%
bluetooth for serial debugging (at least rx 3.3-5V tolerance)
1
2%
board main current sensor
3
7%
native arm emulation software with N1 alike hardware specs ( like qemu2020 -m 1024M -smp 2 -machine N1 -cpu rk3399 -kernel 4.x -hda /dev/pcie_m2.img -hdb sata.img -cdrom ./N3-debugparty.iso -boot d )
2
5%
wide voltage power input (suggested was up to ~19V)
10
24%
 
Total votes : 42

hardware balancing: user wish list and prices

Unread postby back2future » Fri Mar 02, 2018 5:49 pm

odroid wrote:Let me share the price breakdown roughly to minimize misunderstanding with our important community members.

DRAM: ~$40 (1GB/$10: https://www.dramexchange.com )
SoC: ~$20
PCIe-SATA related circuitry: ~$8
Other components: ~$15
SMT/Manufacturing: ~$8

As you can guess, the profit is really limited if we consider the volume discount for our distributors and many B-to-B customers.
I don't think there is much room to reduce the cost even though we can build 10,000 boards every month.

One great news!
We are selling the XU4 board like a hot cake after announcing the N1 board today. :twisted:


Thanks for this interesting insight.
On user side it is easy to extend capable usb 3.x systems (this 3.0 is 5Gbps/per lane?/up to ~350MB/s swapon), s_ata or pci ports, but not very common doing that with d_ram sizes.
Developer boards would either support extendable ram choices or reduce this limitation with offering several price options.
Furthermore especially 4GB shows a wrong structural limitation, like dvd's are at 4.38GiB. Everything above 4.8GiB would be more advanced and pioneering.
This 10$ or 20$ option would be a suitable offer we feel.

[s]extended DRAM to 5 or 6GB: +1[/s] (or 2times 4GB, price for 1.5 boards and we are talking about confident software support with that?)
upgradable, passive cooling: +1
serial/bt debugging adapter 3.3V tolerant: +1

openly: Orange Pi RK3399 Development Board is at around 109$, with only 2GB_DDR3, but with strong graphics emphasis, cam and sensor support. For those seeking such a tool.

odroid-N1: current sensor: +1
Last edited by back2future on Sun Mar 04, 2018 1:20 am, edited 16 times in total.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby odroid » Fri Mar 02, 2018 5:56 pm

Please read the SoC datasheet.
viewtopic.php?f=153&t=30080
User avatar
odroid
Site Admin
 
Posts: 28292
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: hardware balancing: user wish list and prices

Unread postby back2future » Fri Mar 02, 2018 7:13 pm

[btw: multiboot for developing boards would require a graphic card solution supporting grub2 or syslinux or coreboot libreboot or
http://opensource.rock-chips.com/wiki_Rkdeveloptool or a functional similar beYOnd PoE developer server solution.
Thanks for reading.]
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby ASword » Sat Mar 03, 2018 2:37 am

So 4GB being the physical external memory limit of the SoC is your point about reading the datasheet?

Given the cost breakdown you provided (thanks!) it really makes no sense to offer any variants except one that significantly reduces memory. And even then why not just go to an XU4? Better to focus on a single board in order to amortize overheads across as many unit sales as possible.
ASword
 
Posts: 68
Joined: Fri Aug 04, 2017 12:48 pm
languages_spoken: english
ODROIDs: XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby rooted » Sat Mar 03, 2018 2:45 am

Using the video decoding functionality of the N1 VPU you don't need more than 2GB of RAM, so it's like having an XU4 that runs cooler with the media capability of the C2 with much better gaming.

This is why go with the N1 'Lite' vs XU4.
User avatar
rooted
 
Posts: 5181
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1, C1+, C2
XU3 Lite, XU4
N1
VU7+
HiFi Shield 2
Smart Power (original)

Re: hardware balancing: user wish list and prices

Unread postby back2future » Sat Mar 03, 2018 5:16 pm

odroid wrote:Please read the SoC datasheet.
viewtopic.php?f=153&t=30080


Needed some time for understanding this memory details.
Sorry, seems there are 2 ranks for 2 channels (16 or 32 bit data width), each capable of 4GB SDRAM, but address space is only for a total of 4GB available on rk3399 64bit.
There were creative people looking into dram multiplexing (with self-refreshing ram, so no 32 or 64ms interrupt for that at standardized room temps) some years ago and some ideas are even patented, but we would not guarantee for this approach in mass production.
:?:
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby back2future » Sat Mar 03, 2018 6:58 pm

DarkBahamut wrote: Likewise here. This seems like an interesting board and I can see how it could be useful for many people, but it's no real replacement for an XU3/4 device which is a shame (and what I hoped for)

The CPU block is particularly disappointing. It's likely the N1 will actually be no faster or even slower than an XU4 in heavy multithreaded workloads. Plus 8 watts in a single thread browser benchmark? That just goes to show how hard Rockchip are pushing those A72s on 28nm - a process the A72 wasn't targeted for. Personally I would rather have seen an extra $10-20 spent on the BOM for the SoC to get something that isn't quite so 'mid range'. I won't call the N1 'overpriced', but rather I wish it was just a tiny bit more expensive for a much better SoC on a newer low power process.

I do note the datasheet calls for a 1.8GHz max clock on those A72's. I assume the perf/w curves are being seriously abused to achieve the 2GHz clock in the interests of performance - I assume this is why it has to throttle on a large passive cooler even though it only has 2 'big' cores on a modern efficient architecture.

I remain hopeful of a true 'XU5' board someday which returns back to the high-end roots with cutting edge compute on CPU/GPU in the old small form factor. That will be something to get excited about :)


Xu4 was a game changer, mostly because of usb 3. on an economical sbc and wide community os support and compatibility to other arm sbc's.
There would be a chance for high-end sbc like a xu5 with a double rk3399 64bit soc on one board with usb 3. (displayport), eth, hdmi, 6GB DRAM, and additional extension board for pci_e, sata, backup battery and more? I'm wondering how much a dual xu4 (8GB DRAM) now would be?

RoHS/EWRA/J-MOSS 2019: Get a new sbc, give back your used sbc?

[btw: Green Computing Initiative - GCI offers the Certified Green Computing User Specialist (CGCUS), Certified Green Computing Architect (CGCA) and Certified Green Computing Professional (CGCP) certifications.]
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby elatllat » Sat Mar 03, 2018 11:42 pm

I'm interested in what the LPDDR4 performance difference would be but apparently no one knows.
elatllat
 
Posts: 1003
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4, N1

Re: hardware balancing: user wish list and prices

Unread postby odroid » Sun Mar 04, 2018 12:09 am

I also want to know the difference between DDR3-1600 vs LPDDR-1866 in the real world.
User avatar
odroid
Site Admin
 
Posts: 28292
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: hardware balancing: user wish list and prices

Unread postby elatllat » Sun Mar 04, 2018 1:38 am

odroid wrote:I also want to know the difference between DDR3-1600 vs LPDDR-1866 in the real world.

So are you going to have a N1 populated with some so we can see if it doubles the speed and halfs the power consumption as advertised?
elatllat
 
Posts: 1003
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4, N1

Re: SATA - The only numbers that matter

Unread postby back2future » Sun Mar 04, 2018 9:16 pm

tkaiser wrote:
crashoverride wrote:PCIe SATA card arrived. Tested with same SSD:
Code: Select all
02:00.0 SATA controller: ASMedia Technology Inc. ASM1062 Serial ATA Controller (rev 02)

$ sudo hdparm -tT /dev/sdc

/dev/sdc:
 Timing cached reads:   33852 MB in  1.99 seconds = 16993.52 MB/sec
 Timing buffered disk reads: 1130 MB in  3.00 seconds = 376.13 MB/sec


The two things of interest are that its reported as ASM1062 (not ASM1061) SATA (not IDE) controller and the performance is slightly better (376.13 vs 324.42).


I know it won't help you since you're actively rejecting reality but for everyone else: Testing wrongly leads obviously to wrong conclusions.

hdparm was a useful benchmark tool several decades ago, now with faster busses and devices it's just inappropriate. The block size hdparm operates with is 128KB which is too low as already explained several times. The PC you're testing with has a magnitudes higher memory bandwidth, that's why you see 376 MB/s vs. 324 MB/s on the N1 (switch on N1 to little cores that operate with a way lower memory bandwidth and you'll see most probably below 310 MB/s or even less). As soon as you would start to benchmark correctly you would be able to check for the real limitation (even @odroid told you already that your 324 MB/s number is just the result of 'benchmarking gone wrong').

I showed already in this thread that pinning the useless hdparm task to little or big cores makes a difference in reported throughput numbers and that's due to the well known fact that memory bandwidth (almost twice as high when running on big cores) matters with IO as long as the real bottlenecks are not the limit (385 MB/s).

I have an ASM1061 I thought would be an ASM1062 instead in the beginning. Why? Since showing up in Linux as ASM1062. Then I did a firmware update and now it's showing up as ASM1061. Same goes with the device class (IDE vs. SATA). It's all just what the firmware reports to the OS. Unfortunately for whatever reasons on N1 we can not update the firmware (@odroid tried it, I tried it, it doesn't work. @odroid said they might use an SPI protocol analyzer but I think it's not worth the hassles since sequential performance can not change anyway and the 385 MB/s we get everywhere -- yeah on N1 and on your PC! -- will remain. Maybe random IO performance could change but that's something you're not interested in for whatever reasons).

BTW: everything already documented including issues later N1 users might run into: https://forum.armbian.com/topic/6496-od ... ment=50088

On a related note: it's a pretty common 'feature' that SATA controllers show up as different models when queried through software. I have an ASM1053 that will be listed as ASM1153 on some installations while as ASM1053 or ASM1053E on others (depends on some detection logic that is somewhat different in some library somewhere), we deal with JMS578 that are silkscreened as JMS567 on the chip and so on. It's neither a good idea to trust in numbers without meaning nor in some random strings a software spits out.

We need active benchmarking to draw any useful conclusions.


We learned that a top 50% increase in bandwidth from a ASM1061 to ASM1062 could double cost from 4$ to 8$ (with 10k samples).
We did not learn, about advantages of an ai "deeper learning" algorithm (kernel) that sorts out 99,9% of distracting information (that's rk3399pro, Rockpro64-AI hardware NPU, ~99$, ~August 2018)?
It would be interesting what sale numbers are expected for high-end sbc's compared to low-cost x86-64, for being prepared more specific for open source (software) development, that is more influent on these device classes ( and www ).
Almost no emphasis on the impact of kernel features for bandwidth development with alternative settings. High-end would go for that.

Myself i'm struggling with numbers of parallel lines between cpu and memory. 2 times 32 but only 4GB address space. No advance in that compared to x86.
Last edited by back2future on Mon Mar 05, 2018 1:32 am, edited 3 times in total.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: Memory Bandwidth Tests

Unread postby back2future » Sun Mar 04, 2018 9:41 pm

DarkBahamut wrote:Just a point of interest. The N1 doesn't have DDR3-1866 memory. The memory chips on board are Samsung K4B8G1646D-MYK0 which are 8Gbit 16bit chips rated at DDR3L-1600 11-11-11 @ 1.35v (datasheet)

They might do 1866 but we'd effectively be RAM overclocking to do so. I guess this would explain why the uboot is loading the 1600 config ;)


thx
Is there a Xu4 2Gbyte LPDDR3 RAM ePoP stacked datasheet available?

03/11/2018: No public available datasheet so far. viewtopic.php?f=97&t=11940
Last edited by back2future on Sun Mar 11, 2018 8:21 pm, edited 2 times in total.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: Power consumption?

Unread postby back2future » Sun Mar 04, 2018 9:52 pm

neal wrote:
tkaiser wrote:
neal wrote:Booting, Idle and power off with same ondemand scaling_governor and no HDD plugging, but only the difference is the IC present or not on the board.


Can you also test whether disabling the PCIe DT nodes results in further savings? It's done in minutes: https://forum.armbian.com/topic/6496-od ... ment=49439


N1_ASM1061_onNoff_removeDTS.png

download/file.php?id=7078

Code: Select all
root@odroid:~# cat /proc/device-tree/model; echo
Hardkernel ODROID-N1 low power
root@odroid:~# lspci
root@odroid:~#


I added the green & purple line by tkaiser requested.


Thanks ODROID board for solving general questions :)
How could one assemble the influence of idling ram compared to full workload on power consumption (estimation from specs 1:6)?
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: KVM testing

Unread postby back2future » Sun Mar 11, 2018 6:16 pm

(KVM testing: viewtopic.php?f=150&t=30388&p=218146#p218146
If one wants to follow this on a xu3/xu4&h/mci for being prepared for a N1:
Getting into HYP mode, will require bootloader firmware u-boot at a (for example) 2017.05 level.

( Image starting 2048 sectors of https://wiki.odroid.com/odroid-xu4/os_i ... .9/minimal can provide that, in addition with its 4.9.44-56 precompiled kernel.
Exclude bytes 446(0x01BE)-512(0x0200), if an existing 2nd or additional partition for rootfs shall remain like used before.
dd_cautious if=./ubuntu-16.04.3-4.9-minimal-odroid-xu4-20170824.img of=/dev/sdcard bs=446 count=1
dd_cautious if=./ubuntu-16.04.3-4.9-minimal-odroid-xu4-20170824.img of=/dev/sdcard bs=512 count=2044 seek=1 skip=1
Copy zImage, uInitrd, exynos5422-odroidxu?.dtb (or search for exynos5422-odroidxu4-kvm.dtb¹) into boot partition, 4.9.44-56 kernel modules into wanted rootfs and adjust boot.ini.)

Additional instruction that can help (although it's for c2 boards):
https://wiki.odroid.com/odroid-c2/appli ... ualization

( Backup data. Experienced u-boot users could adjust this "boot.ini for SD" example for that purpose, non-sophisticated users would start with an additional sd-card,
https://wiki.odroid.com/odroid-xu4/soft ... _lineageos
Android needs 3 partitions (system.img, userdata.img, cache.img) for a default system usability, so more difficult for hypervisoring virtualization, any the less useful: viewtopic.php?f=151&t=30084 )

¹ https://github.com/hardkernel/linux/blo ... u4-kvm.dts

and thanks for all Your input to this poll.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby back2future » Sun Mar 11, 2018 9:34 pm

elatllat wrote:
odroid wrote:I also want to know the difference between DDR3-1600 vs LPDDR-1866 in the real world.

So are you going to have a N1 populated with some so we can see if it doubles the speed and halfs the power consumption as advertised?


Seems that there could be a lpddr3/lpddr4 socket possible.
Edit 04/2018: COM (ff ESM (149×71) or ESMexpress (125×95), type2 (55x84 - 110x155) or XTX (114x95) vs. NUC (102x102) ) seems to be a more stable approach towards customized ram sizes.
Image
(https://cdn57.androidauthority.net/wp-c ... hannel.jpg)

For a lpddr3 (pin compatible with lpddr2) with 16 data queue pins, there are
4 data strobe or data clock pins,
2 control pins,
10 address pins,
2 command pins,
2 clock pins,
1 (dynamic) clock calibration pin

Maximum socket distance from soc would be only few (<2-4) inches.
Last edited by back2future on Fri Apr 06, 2018 7:37 pm, edited 1 time in total.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby back2future » Fri Mar 23, 2018 6:54 pm

Maybe a ram expert could add some practical insight, which combination of dram and cpu overclocking gives best performance for given energy input?
[ exynos5422, dram overclocking: viewtopic.php?f=93&t=27117 ]

User interest seems (limited survey) being around on ram specific development, but not that big expert discussion on that central item?
( Case modding for cooling and power supply and driver optimization for storage improvement and stability seems to attract many. Multimedia is attractive for many. Debugging os/kernel or emulation for os benchmarking or interface interaction seems being less attractive for users here. Innovation is good news, but nothing really aiming for? Desirable this kind of poll would invite more users for contribution, because it is useful? )

Some way or other, anyway, thanks for all your feedback.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby back2future » Fri Apr 06, 2018 8:37 pm

One thing is to wait for SoC addressing RAM above 4GB.

The more interesting part would be, if LRU (least recently updated pages) are stored into enhanced ram memory, controlled by multiplexed ram command and clock lines. Well, that's no workaround for today's memory prices, but would provide longer usability for today's sbc into future demands. ( Keeping in mind that one is aiming for environmentally friendly development progress, also. )
Then there is no need for the soc address space being capable of addressing the additional 2-4GB?
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby odroid » Fri Apr 06, 2018 9:39 pm

I believe you must be happier with this $100 item which can satisfy your requirements.
https://www.asrock.com/mb/Intel/J4105-ITX/
User avatar
odroid
Site Admin
 
Posts: 28292
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: hardware balancing: user wish list and prices

Unread postby back2future » Fri Apr 06, 2018 11:05 pm

odroid wrote:I believe you must be happier with this $100 item which can satisfy your requirements.
https://www.asrock.com/mb/Intel/J4105-ITX/


Thanks for Your support, but they don't offer open hardware development. (Zen) Fulltime oc is just some kind of workaround, if one considers default component life time.
What do You think about this poll that is probably heavily influenced by early N1 users? Would this show a roadmap You want to follow for progress?

[ ATX 2.4 lowered minimum starting current to 0.05A for +12V (11,4-12,6V, +-5%) output. Just interesting. ]
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby odroid » Fri Apr 06, 2018 11:17 pm

We are not doing the open source hardware. We just release the full schematics only.
This poll is a sort of dream. The feasibility is quite low.
There is no way to add a SO-DIMM to the SoC since there are some missing signals.
We have not enough technology to overcome the SoC design limitation. So we can't add more memory than the SoC datasheet specifcation.
Try to find other SBC suppliers who can implement your wish list.
User avatar
odroid
Site Admin
 
Posts: 28292
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: hardware balancing: user wish list and prices

Unread postby back2future » Fri Apr 06, 2018 11:23 pm

We should talk about that wish list in about 5 years.
We like Your board style and friendly interaction, that's not everywhere style in big companies. Thanks for that also.

feasibility: 2 items, ram and qemu adjustment for N1 are tough, the others quite doable. Users just showed their preferences, so that's where we want to be in years to come.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby back2future » Sat Apr 07, 2018 11:23 pm

Thought about that.
We do not know who is following in this poll the other, but the strong emphasis at start on dram, laminar heattransfer to air or case and wide voltage input still is visible. Only one voted for a current sensor, that might support wake-on-[special peripherals, not only lan].

SO-DIMM (edit: also) became a common connector for SOM combinations of soc and ram and is therefore reduced to its electrical connection ability, without standardized pin out.

Don't know if there is sense in copying other companies approach with offering Odroid SoM boards and complementary main- or base board for mass production. For development that's more common, like shown in announcements from mobile section. Mentioning LPDDR4X, then 0.6V is lowest ram voltage we heard of.

For unique sbc one could think about a second level ram area like Odroid ramdisks ("recycled" ddr2 would be fast enough for that, but is not accessable by rk3399 protocols), one could hotplug from single device to next single device generation.
Otherwise we stuck with soc manufacturers abilities or roadmaps, considering ram.
[ Considering Futurebus (IEEE 896) history one could regret this race against improving pcie and thunderbolt/usb/(hdmi) or newer SoCs with 4-8GB ram? Better wait for next arm generations? ]

Rk3399 supports electrical and by protocol low power DDR3 and DDR4.

And yes, it would be interesting, what people at Rockchip would add towards creative ram extension.

Sorry. Back to what is doable with 4GB. Enough luxury.
Last edited by back2future on Sat Apr 21, 2018 12:44 am, edited 1 time in total.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby back2future » Sat Apr 14, 2018 7:47 pm

Not so many seem to analyse hardware based data compression.
2014 AHA compares software and hardware compression.
AHA37x Compression Accelerator Demo https://www.youtube.com/watch?v=mdNDuPy3NnI

AHA37x:
>> 2.85-6.25:1 gzip compression ratios
>> gzip, zlib, lzs (open standard)
(116J_Xeon ~6.8J_AHA372 per Gb)
>> pcie 2.0, 8x
(pcie cards 12-14W @ 100%io
Image)

Q1/2016:
Xeon E5-2640, 8GT/s QPIx2, 4x DDR4-2133_1.54TB@68,3GB/s_46b, 40 lane PCIe
Q3/2017:
Xeon® Platinum 8180M
i9-7960X, 8GT/s DMI3, 4x DDR4-2666, 44 lane PCIe
Ryzen 1950x, (4x) 8GT/s , 4x DDR4-2167, 60 lane PCIe
https://en.wikichip.org/wiki/WikiChip

[ Intel® QuickAssist Technology - Performance Optimization Guide https://01.org/sites/default/files/down ... nguide.pdf ]
Last edited by back2future on Sun Apr 29, 2018 2:15 am, edited 5 times in total.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby elatllat » Sun Apr 15, 2018 8:01 am

I don't know of any application where data compression is the limiting factor.
elatllat
 
Posts: 1003
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4, N1

Re: hardware balancing: user wish list and prices

Unread postby back2future » Sun Apr 15, 2018 9:34 am

For more accurate summary we would want analyses for partly loaded (1/5/15 minutes) systems[*2] as well, but this[*1] references an interesting insight into comparing bandwidth optimization and necessary energy foresight for communication devices. Don't think this only for 10Gbit servers like done with pcie expansion cards, AI-IA also enables state-of-the-art (upgrades through hardware exchange) memory optimization at system bus level. Additionally their concept reduces energy demand.
Any numbers available, how much their AHA3641/AHA3642 are in $?

1) http://www.aha.com GZIP_Benefits_Whitepaper11.pdf http://www.aha.com/Uploads/GZIP_Benefits_Whitepaper11.pdf pg. 3, fig. 6
2) Data centers, 2012, IBM research https://domino.research.ibm.com/library/cyberdig.nsf/papers/0C306B31CF0D3861852579E40045F17F/$File/rz3820.pdf pg. 10
compare 24 month average server workloads at diff countries: cpu7-25%, mem72-84%, disk64-81%, filesys_utilization39-55%

http://www.datacenterknowledge.com, 2016, LBNL-DOE-DC-energy-use-efficiency-impact.jpg
Image
alternative: https://en.wikipedia.org/wiki/Data_center#Energy_use
Last edited by back2future on Sun Apr 15, 2018 9:24 pm, edited 1 time in total.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby elatllat » Sun Apr 15, 2018 8:06 pm

elatllat
 
Posts: 1003
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4, N1

Re: hardware balancing: user wish list and prices

Unread postby ASword » Wed Apr 18, 2018 2:07 am

Well there's more to it than just Moore's Law. Specialized compression/decompression hardware can yield significant improvement in compression heavy workloads that CPUs cannot easily overcome in a generation or two because we hit the power wall a decade ago and the rate of process improvement has slowed considerably because the cost increment of each successive generation continues to grow while giving diminishing returns and additional problems.

However... I don't follow what relevance this has to N-1. Such compression hardware is quite specialized and not relevant to almost all use cases the N-1 is likely to be used for. As such, it doesn't make sense for HK to expend the design effort to add expensive and power-hungry silicon (and given AHA's target market, I can be confident that relative to the N-1, it is those things) to an already tightly packed board when a diminishingly small fraction of the N-1's market is going to use it. And if they were to purpose build a board to hit the market that will use the compression hardware, such a board would look substantially different than the N-1 (IMO).

I think the leading problem in the N-1 is the relatively high price point, which is tied directly to DRAM pricing and little else. Adding more to the board will only drive the price up further and thus is counter to addressing its leading problem. The main distinguishing feature of the board is its 4GB though, so there's not much to be done about it unless a lower cost source of DRAM can be found without compromising the board's design.
ASword
 
Posts: 68
Joined: Fri Aug 04, 2017 12:48 pm
languages_spoken: english
ODROIDs: XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby rooted » Wed Apr 18, 2018 2:17 am

ASword wrote:The main distinguishing feature of the board is its 4GB though, so there's not much to be done about it unless a lower cost source of DRAM can be found without compromising the board's design.


There can be, it's just slower.

Then it wouldn't be much different than the RockPro64, just more expensive.
User avatar
rooted
 
Posts: 5181
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1, C1+, C2
XU3 Lite, XU4
N1
VU7+
HiFi Shield 2
Smart Power (original)

Re: hardware balancing: user wish list and prices

Unread postby back2future » Wed Apr 18, 2018 5:19 am

Simultaneous hardware comp/decomp by AHA is one approach towards ram optimization.
You are right, this is probably more interesting for Rockchip people, (re)designing next SoC's?
Odroid is no open source hardware company, and a AHA expansion card on pcie would be way off for (most todays) SBCs.

for the bigger picture: People want more ram.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby ASword » Wed Apr 18, 2018 1:49 pm

rooted wrote:
ASword wrote:The main distinguishing feature of the board is its 4GB though, so there's not much to be done about it unless a lower cost source of DRAM can be found without compromising the board's design.


There can be, it's just slower.

Then it wouldn't be much different than the RockPro64, just more expensive.


You ignored my final clause — “without compromising the board’s design”. Slower DRAM would do that, IMO (and yours, by the sound of it). So no wiggle room there, and more motivation to have fewer designs and greater component commonality so that larger memory orders can be placed. So get out there and evangelize. :)

As for in-memory compression, this isn’t feasible in such a design. Those compression ASICS tend to have rather long latencies with high throughput. That’s good for disk I/O, but not for CPU / DRAM communication.
ASword
 
Posts: 68
Joined: Fri Aug 04, 2017 12:48 pm
languages_spoken: english
ODROIDs: XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby rooted » Wed Apr 18, 2018 2:20 pm

ASword wrote:
You ignored my final clause — “without compromising the board’s design”. So get out there and evangelize. :)


I wasn't ignoring it just saying the only thing I could think of to lower price, it certainly would be an illogical compromise.

I will hit up my black market contact in Asia and get us a deep discount on LPDDR4 (barely used by Samsung) :)
User avatar
rooted
 
Posts: 5181
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1, C1+, C2
XU3 Lite, XU4
N1
VU7+
HiFi Shield 2
Smart Power (original)

Re: hardware balancing: user wish list and prices

Unread postby tkaiser » Wed Apr 18, 2018 6:38 pm

rooted wrote:
ASword wrote:Then it wouldn't be much different than the RockPro64, just more expensive.


Has something changed? AFAIK the N1 uses currently DDR3-1600 and Hardkernel evaluates DDR3-1866 for the production batches? At least with RockPro64 it's somewhat similar: they want to use DDR3-1866 but in case they can not source those chips reliably they go with DDR3-1600 instead.

In case various board revisions get different DRAM chips from a software point of view the slowest batch has to be taken into account for DRAM initialization. I asked the Pine64 guys a while ago and they said RK3399 does not support dynamic DRAM calibration. I also asked about prices and they were still talking about '$59 to $65 with 2GB RAM, and $79 with 4GB RAM' -- using the same type of DRAM as N1.

Talking about prices. The Rock960 EE board with RK3399Pro (the NPU enabled variant) has a target price of $99 which I find a bit hard to believe...
tkaiser
 
Posts: 361
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby ASword » Thu Apr 19, 2018 12:04 am

Your quoting mis-attributes that line to me, tkaiser. I have no idea as I've not bothered to look at the RockPro64 in any detail. As for pricing choices, until they actually let me buy a device at the price point they're claiming, I don't believe anything. Its easy to hand wave about a lower price, but its another matter to actually source the parts inventory and be willing to sell your product when your accountant holds your feet to the fire.

For myself, even a $50 difference is irrelevant as I'm only buying 1-2 of them. If I can be more assured of board maker support, on-going production, and an active community, then $50 is more than worth my own time saved. If HK's market is mostly individuals doing projects at very low volumes, then they can do well with a solid reputation and a $10-20 price delta. Even better if more customers realize that sinking 10+ hours of their own time into struggling with a poorly supported board completely negates any cost savings.
ASword
 
Posts: 68
Joined: Fri Aug 04, 2017 12:48 pm
languages_spoken: english
ODROIDs: XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby elatllat » Thu Apr 19, 2018 12:29 am

ASword wrote:...until they actually let me buy a device...

4GB DDR4 $75
(but RK3328 and be prepared to wait forever on any support questions you may have)
elatllat
 
Posts: 1003
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4, N1

Re: hardware balancing: user wish list and prices

Unread postby tkaiser » Thu Apr 19, 2018 12:33 am

ASword wrote:Your quoting mis-attributes that line to me, tkaiser.


Sorry for that.

ASword wrote:I have no idea as I've not bothered to look at the RockPro64 in any detail. As for pricing choices, until they actually let me buy a device at the price point they're claiming, I don't believe anything. Its easy to hand wave about a lower price


That's IMO not what they're doing since I asked whether the prices are already final or not. Only the 2GB variant was not decided but the 4GB thing goes for $79. But you're right, let's wait and see. But it should be obvious that the price difference between N1 and such a RockPro64 is not due to different DRAM (BTW: RockPro64 will use LPDDR3 which to my knowledge is a little bit more expensive than DDR3 used on the N1 -- but I might be wrong).

Wrt software support: it's not about boards but SoCs. It just looks differently when talking about the most successful board makers so far (RPi folks and Hardkernel) since they use or used some SoCs exclusively. With RK3399 it's different and this SoC receives great support by the SoC vendor, there are already a bunch of RK3399 boards out there and a lot more coming soon and everything relevant ist driver/kernel stuff where situation with both RK's 4.4 LTS kernel and mainline already looks pretty good.

Differentiation will become quite difficult based on this especially since I consider software support situation around RK3328/RK3399 great already. Rock64/RockPro64 community plans to move to device agnostic OS images soon (there's a reason why all Pine64/Rock64 boards in the meantime ship with a sufficiently large SPI NOR flash) using also unified kernel branches.
tkaiser
 
Posts: 361
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby tkaiser » Thu Apr 19, 2018 12:40 am

elatllat wrote:
ASword wrote:...until they actually let me buy a device...

4GB DDR4 $75
(but RK3328 and be prepared to wait forever on any support questions you may have)


Again: it's about SoC and not board. All that's needed to run any great OS image on this Firefly thing is a Rock64 image with exchanged DRAM initialization and a different DT file. That's the whole set of changes to bring any of the Armbian variants to this board: https://github.com/armbian/build/pull/921/files

But RK3328 and RK3399 shouldn't be compared directly since the latter SoC costs $20 to my knowledge.
tkaiser
 
Posts: 361
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby crashoverride » Thu Apr 19, 2018 2:34 am

tkaiser wrote:it's about SoC and not board.

After many years of experience, I can say that while anyone can make an ARM board, there are few companies that can actually support the hardware they made. So its actually about 1) SoC, 2) board design, and most importantly 3) vendor support.

1) SoC - For all current Rockchip parts, 4K support is "tacked on". I have elaborated in other posts what is lacking in this area. The C2 is far more efficient at handling 4K than any RK part. RK3399 achieves 4K through "brute force".

2) board - Although both Rock64 and the Firefly 3328 board have the exact same SoC, the Rock64 choice of single channel DRAM limits its performance. You can have a low cost board or a fast board but not both.

3) vendor support - Firefly support is a joke. Most recently, I attempted to reflash my Firefly RK3399 but could not even download the image. Its been well over a month of people posting about the issue before it was corrected. While a generic RK3399 Armbian may boot, it does nothing for the support issues posted such as dual MIPI camera use, USB-C kernel crashes, etc. Without vendor support, the board is a paper weight.

My personal experience with support directly from Rockchip goes like this: https://github.com/rockchip-linux/libmali/issues/25
That is why vendor support is mandatory especially when it comes to Rockchip based devices. See also Tinkerboard.
crashoverride
 
Posts: 3713
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: hardware balancing: user wish list and prices

Unread postby tkaiser » Thu Apr 19, 2018 4:04 am

crashoverride wrote:1) SoC - For all current Rockchip parts, 4K support is "tacked on". I have elaborated in other posts what is lacking in this area. The C2 is far more efficient at handling 4K than any RK part. RK3399 achieves 4K through "brute force".


Great. So let's switch to an Amlogic SoC for the N1 better sooner than later!

crashoverride wrote:2) board - Although both Rock64 and the Firefly 3328 board have the exact same SoC, the Rock64 choice of single channel DRAM limits its performance. You can have a low cost board or a fast board but not both.


Can you interpret these results trying to compare both boards wrt memory performance? https://forum.armbian.com/topic/6772-ti ... 328-cc-1gb

BTW: Da Xue is the guy behind "Libre Computer" who partnered with Firefly for this RK3328 board. He said only reason for slightly more expensive DDR4 was 4K video support since requiring high memory bandwidth. I've neither the hardware nor am I interested in this use case at all (if I would need a Kodi box according to LibreELEC folks some el cheapo Amlogic S905X box should do)

crashoverride wrote:3) vendor support - Firefly support is a joke. Most recently, I attempted to reflash my Firefly RK3399 but could not even download the image. Its been well over a month of people posting about the issue before it was corrected. While a generic RK3399 Armbian may boot, it does nothing for the support issues posted such as dual MIPI camera use, USB-C kernel crashes, etc. Without vendor support, the board is a paper weight.


Then good luck hoping for Hardkernel doing a better job supporting Rockchip hardware than Rockchip.

BTW: I've not the slightest idea how the 'quality' of Firefly support is related to questions about DRAM costs or how N1 and other RK3399 devices like RockPro64 compare and why they are priced differently. But fortunately that's not my but Hardkernel's problem...
tkaiser
 
Posts: 361
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby back2future » Thu Apr 19, 2018 4:24 am

Lowest price now for a 4GB DRAM-controller is probably a Rock64/rk3328 at 12$/GB.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby ASword » Thu Apr 19, 2018 5:32 am

elatllat wrote:
ASword wrote:...until they actually let me buy a device...

4GB DDR4 $75
(but RK3328 and be prepared to wait forever on any support questions you may have)


Likely that nobody cares, but that device isn't interesting to me. I want the 2 big + 4 little cores plus stronger GPU of the RK3399, and the dual port SATA controller. So its not purely the 4GB, even though that's the primary price determinant. Oh, and strangely enough I don't care about the video output at all.


As for why the quality of support is related to pricing, I just repeat my assertion that for low volume purchases saving a few bucks is a false economy. If you're going to be tens of boards it might be worth saving the hardware cost, if you're buying hundreds then its more likely, and very likely if you're buying thousands. If you value your own time in a reasonable way, it doesn't take much to burn through a lot of money.


As for "better support for Rockchip than Rockchip", on some matters you're right -- they can't do better than the CPU maker. For providing a community, responsiveness to questions, accessible documentation, and improved packaging of software though, then there are opportunities. Can HK capitalize on this? Maybe. They're actually doing a pretty good job thus far, IMO.
ASword
 
Posts: 68
Joined: Fri Aug 04, 2017 12:48 pm
languages_spoken: english
ODROIDs: XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby elatllat » Thu Apr 19, 2018 10:27 am

tkaiser wrote:...All that's needed to run any great OS image...

If it were easy there would be armbian downloads for a lot more SBCs.

tkaiser wrote:...doing a better job supporting Rockchip hardware than Rockchip...

Like how Samsung is using an old unsupported kernel and Odroid is using a supported LTS kernel on Samsungs SOC?
or how Amlogic did not have a video player for Linux but one appeard on the Odroid forms?
Wayland, gl4es, qt5, etc.
Sure it's higher level stuff but if it's so easy please help us fix the video display on the N1 and get a kernel close to the latest LTS.

ASword wrote:...primary price determinant...

Yah just pointing out someone got a good deal on RAM.
elatllat
 
Posts: 1003
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4, N1

Re: hardware balancing: user wish list and prices

Unread postby rooted » Thu Apr 19, 2018 11:06 am

∆∆∆∆∆
Thumbs
User avatar
rooted
 
Posts: 5181
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1, C1+, C2
XU3 Lite, XU4
N1
VU7+
HiFi Shield 2
Smart Power (original)

Re: hardware balancing: user wish list and prices

Unread postby tkaiser » Thu Apr 19, 2018 1:40 pm

elatllat wrote:
tkaiser wrote:...All that's needed to run any great OS image...

If it were easy there would be armbian downloads for a lot more SBCs.


Armbian does not even support a single RK3328 device yet. For the simple reason we're too few developers and the amount of officially supported boards is still too high (in my opinion but 'project lead' has a different opinion). I was talking more general and it still should be obvious that it's only about SoCs and the board maker is pretty much irrelevant: If there's a LibreELEC build for one RK3328 device like Rock64 all that's needed to get this working on this Firefly thing is still just exchanging DRAM initialization and DT.

elatllat wrote:
tkaiser wrote:...doing a better job supporting Rockchip hardware than Rockchip...

Like how Samsung is using an old unsupported kernel and Odroid is using a supported LTS kernel on Samsungs SOC?
or how Amlogic did not have a video player for Linux but one appeard on the Odroid forms?


Obviously it's a waste of time discussing the differences between support per SoC vendor and also the differences between past ODROID boards and future ones here.
tkaiser
 
Posts: 361
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby crashoverride » Thu Apr 19, 2018 2:02 pm

tkaiser wrote:Obviously it's a waste of time discussing the differences between support per SoC vendor and also the differences between past ODROID boards and future ones here.

The reason we are here is precisely because there is no support there. :lol:
crashoverride
 
Posts: 3713
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: hardware balancing: user wish list and prices

Unread postby back2future » Fri Apr 20, 2018 6:17 am

Farthest one could look is probably Samsung's roadmap for AArch64/ARM8(.4)-A platform or other mobile soc developers in general *¹
(as of 2017: a total of ~100x10^9 Acorn RISC Machine devices world wide have been produced).
DDR2 would provide sufficient transfer rate (3-6GB/s) per module for memory prefetch (above 4GB) on rk3399, but these days average stock prices are too close at DDR3s for that approach.
So not much urgency for discussing older DDR2 (usb 3.0 otg) devices (hardware LTS? since 2003) nowadays.

Thxs @all for contribution.

[ Some numbers: 4k streaming is done from lowest ~15 Mbit/s to mostly 50 Mbit/s if caching needs are relatively high. ]
[ *1 qemu 2.12.0 @ 04242018 supports up to arm64v8.3 extension platforms/boards, https://wiki.qemu.org/ChangeLog/2.12#ARM, new feature: qemu-nbd server ]
Last edited by back2future on Sat Apr 28, 2018 11:18 pm, edited 2 times in total.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby back2future » Fri Apr 20, 2018 6:30 pm

tkaiser wrote:
rooted wrote:
ASword wrote:Then it wouldn't be much different than the RockPro64, just more expensive.


Has something changed? AFAIK the N1 uses currently DDR3-1600 and Hardkernel evaluates DDR3-1866 for the production batches? At least with RockPro64 it's somewhat similar: they want to use DDR3-1866 but in case they can not source those chips reliably they go with DDR3-1600 instead.

In case various board revisions get different DRAM chips from a software point of view the slowest batch has to be taken into account for DRAM initialization. I asked the Pine64 guys a while ago and they said RK3399 does not support dynamic DRAM calibration. I also asked about prices and they were still talking about '$59 to $65 with 2GB RAM, and $79 with 4GB RAM' -- using the same type of DRAM as N1.

Talking about prices. The Rock960 EE board with RK3399Pro (the NPU enabled variant) has a target price of $99 which I find a bit hard to believe...


SoC developers seem to externalize special functions and evolve work load into decentralized coprocessors. Lower bandwidth growth towards ram memory, but higher energy efficiency improvements.
Like dynamic dram calibration (cpu unbound) yields into lower consumption, while being prepared to keep used latencies.
( Maybe we have to take care to avoid marketing slogans responding with 'comparable to 4GB ram experience' ? )
Furthermore, it seems, we are at a point in dev where long hours of sleep (or low consumption standby) mode compete against boot/system-off states.
( Wake on LAN is now below 0.5W, was 2-8W at ~2011. )
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: hardware balancing: user wish list and prices

Unread postby tkaiser » Sat Apr 28, 2018 4:16 pm

ASword wrote:As for pricing choices, until they actually let me buy a device at the price point they're claiming, I don't believe anything.


https://www.cnx-software.com/2018/04/28 ... evelopers/
tkaiser
 
Posts: 361
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby rooted » Sat Apr 28, 2018 11:48 pm

tkaiser wrote:
ASword wrote:As for pricing choices, until they actually let me buy a device at the price point they're claiming, I don't believe anything.


https://www.cnx-software.com/2018/04/28 ... evelopers/
That's going to be difficult to beat price wise, if they support the software properly it looks good.
User avatar
rooted
 
Posts: 5181
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1, C1+, C2
XU3 Lite, XU4
N1
VU7+
HiFi Shield 2
Smart Power (original)

Re: hardware balancing: user wish list and prices

Unread postby ASword » Thu May 03, 2018 12:29 am

The hardware looks promising, but reading the forums about the OS distros it sounds like a bit of a nightmare. Fine if you want to spend your time hacking on the OS to get it to work and stable... but not if you actually want to do real work / usage. Will be interesting to see how things are when their board gets to a real consumer level rather than its current developers-only state.

Have to say though, that for an extra $20 adding the NPU to their RockPro64-AI version is pretty darn compelling. And its still less than the N-1's projected price. Now that the N-1 is delayed until at least August, they'll be up against this model too. I also like that I can choose between 2 SATA ports and an M.2 NVMe port.

Edit: it appears that the NPU is built into a pro version of the SoC. Unless the SoC's interface changes, that would be an easy upgrade for the N-1.
ASword
 
Posts: 68
Joined: Fri Aug 04, 2017 12:48 pm
languages_spoken: english
ODROIDs: XU4, HC1

Re: hardware balancing: user wish list and prices

Unread postby back2future » Fri May 04, 2018 7:27 pm

Developers already answer next questions about tech. future, like they don't only read and support what expresses their current point of view.
https://www.cnx-software.com/2018/05/03 ... -ethernet/
( For daily pipeline work one could choose x64 platforms that are well supported by established application support, why waiting for aarch64 therefore?
How long will DDR3/DDR4 shortage, compared to consumer demands, last into future?
[ a Samsung 960pro will last ~120-180h if its v-nand flash is rewritten constantly at full possible bus speed @ 2.1GB/s,
for low cost, unbranded flash, @ 150MB/s usb 3.0, one could assume todays lower than 0.3$/GB for swap capacity and below ~1/2 of that for TB_throughput.
( Samsung 970pro, 1TB, 1GB lpddr4 Cache, supports guaranted 5 years usage or up to 1200TB written. ~06/2018_~0.50$/TB_written )
Memory swap above available ram?
Page cache: https://en.wikipedia.org/wiki/Page_cache ]
Not sure, if this point is clear enough, people don't want more devices with each costly ram, people want usable ram, if they need that expansion.
That's a talk about resources, that we should care [meta dimensional: r2p] for. )

If there is fewer openly hardware development on N1, this sub board could evolve to a more general discussion about next possiblities. (Interest for AI visual detection and filtering is just growing, pushed by autonomous driving and mobile assistant systems. Like it or not.)
[ Neural processors: https://en.wikichip.org/wiki/neural_processor ]
Talking about Pine64 Rockpro64, it's an additional discussion on peripherals, that are developing parts of a sbc 'mainboard' ( pointing towards desktop replacement, but at far lower cost bought through peak power, bandwidth and flexibility limitations). Talking about Rockpro64 (vendor independently being a successor for xu4/c2), there's no need pushing Hardkernel into a lane they don't see purpose for.

Who will ask our next supportive questions? Desktops again?

[ Amlogic: S922
NXP: LX2xx0A ]
Last edited by back2future on Sun Jun 24, 2018 12:02 am, edited 4 times in total.
naturally beYOnd spectrum
back2future
 
Posts: 69
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Next

Return to Hardware and peripherals

Who is online

Users browsing this forum: No registered users and 1 guest