2 2
e97

SBC with native dual gigabit ethernet (dual+ core, ~$50)

Recommended Posts

Hey folks,

 

Wondering if anyone has come across a dual core or better single board computer that has dual native gigabit ethernet?

 

I'm also trying to find something around $50.. so far the best I've found is the:   NXP FRDM-LS1012A-PA - $50

which has good BSP and hardware packet accelerator but only single core @ 800 MHz

 

As a last resort, I'm considering the NanoPi Neo4 with a USB to Gigabit ethernet adapter, thats about $75 with heatsink and USB 3.0 to Gigabit adapter.

Share this post


Link to post
Share on other sites

board-db often has incorrect information..

 

the only other board I've come across at ~$50 is the ESPRESSObin but its sw support doesnt include hw acceleration like the NXP LS1012A so its in fact slower/higher latency and has less RAM.

 

I asked OrangePI, FriendlyElec, Firefly, etc.. if they have any plans to make something similar and its under consideration but I think its unlikely.

 

An OrangePI R1 like device with dual gigabit and a quad core a8 at ~$30 would make an excellent firewall/network appliance thats lower power.

Share this post


Link to post
Share on other sites
1 hour ago, e97 said:

An OrangePI R1 like device with dual gigabit and a quad core a8 at ~$30 would make an excellent firewall/network appliance thats lower power.

The BPi R1, R2 and W2. All above $50. But you've got to pay what it costs, manufacturers aren't going to change prices.
No idea about the software support for them. EspressoBin is a good choice.
https://www.amazon.com/ESPRESSObin-SBUD102-Single-Computer-Network/dp/B06Y3V2FBK
You can also buy a Rock64 and add USB3 network adapters. Enough choice, you only have to make it work...

Share this post


Link to post
Share on other sites

Only curious to know if anyone has knowledge of a board I missed.

 

Economic law of supply and demand - if demand decreases meaning people don't buy => then price must fall or supply must fall until equilibrium is reached.

 

How are mfgs able to continuously make new boards with hw issues and no sw support?  Only way is if they are making money off of selling a few boards or are heavily subsidized.

 

Having manufactured in China gives me insight into the costs to produce electronics. I'd bet at least 10x margin from researching all these kickstarter boards.

 

Gambling on these kickstarter boards is a fools errand. I'll stick to a board with a detailed and tested BSP or I build my own.. at least then I know exactly what I'm getting - bugs and all.

 

 

 

 

Share this post


Link to post
Share on other sites
8 hours ago, e97 said:

An OrangePI R1 like device with dual gigabit and a quad core a8 at ~$30 would make an excellent firewall/network appliance thats lower power.

Actually you can always use VLAN to implement such firewall or router, combined with a switch which supports 802.1q aka VLAN tagging. The drawback is that you can only have 1Gbps in a single direction.

BTW, internal GbE on devices like ROCK64 don't do well in these scenarios, because when running Ethernet in full duplex, its CPU reaches 100% (single core), and the full duplex throughput is only around ~400Mbps. An RTL8153 attached to the USB3, however, does this job quite well. The CPU consumption stays very low even when stress testing full duplex. However Ethernet over USB sometimes has stability issues. I think these issues are the ones that could not be avoided when evaluating cheap devices. It's a trade off.

Share this post


Link to post
Share on other sites
1 hour ago, hjc said:

Actually you can always use VLAN to implement such firewall or router, combined with a switch which supports 802.1q aka VLAN tagging. The drawback is that you can only have 1Gbps in a single direction.

BTW, internal GbE on devices like ROCK64 don't do well in these scenarios, because when running Ethernet in full duplex, its CPU reaches 100% (single core), and the full duplex throughput is only around ~400Mbps. An RTL8153 attached to the USB3, however, does this job quite well. The CPU consumption stays very low even when stress testing full duplex. However Ethernet over USB sometimes has stability issues. I think these issues are the ones that could not be avoided when evaluating cheap devices. It's a trade off.

 

Excellent points.

 

I stumbled upon a post here (which I can't seem to find now) that discusses issues with internal USB hubs/power and Ethernet stability on various SBCs.

 

I'm trying to avoid VLAN tagging because I want to achieve full duplex line rate (1 Gbps) and avoid security concerns. I have a few RK3288 boards but I'd like two NICs for security.

 

NXP support tells me the LS1012A can do 1Gbps at <1W but I'm skeptical.

 

Also on my TODO is testing MQmaker WiTi running openwrt or debian to see if it can achieve this, MT7621A (MIPS 1004Kc @880MHz 2C/4T) and if all else fails will go back to an x86 server but the power usage will be high :(

 

 

 

Share this post


Link to post
Share on other sites
4 hours ago, e97 said:

I'm trying to avoid VLAN tagging because I want to achieve full duplex line rate (1 Gbps) and avoid security concerns. I have a few RK3288 boards but I'd like two NICs for security.

shouldn't vlan exactly solve such issues/concerns? I think it needs a vlan capable switch as well.. No idea, never spent much attention to VLANs. 

 

4 hours ago, e97 said:

Also on my TODO is testing MQmaker WiTi running openwrt or debian to see if it can achieve this, MT7621A (MIPS 1004Kc @880MHz 2C/4T) and if all else fails will go back to an x86 server but the power usage will be high :(

do those MT7621 boards get regular kernel updates? Otherwise this would IMO be a bigger security issue. I've no experience in MIPS world.. Only an old ASUS router and it was just the wrong device.. :D 

13 hours ago, e97 said:

Gambling on these kickstarter boards is a fools errand.

Isn't it a kickstarter board? :P

 

13 hours ago, e97 said:

Economic law of supply and demand - if demand decreases meaning people don't buy => then price must fall or supply must fall until equilibrium is reached.

or the company goes bankrupt and a new one tries it.. Or the cut support for a board which doesn't sell well.. :D

 

13 hours ago, e97 said:

How are mfgs able to continuously make new boards with hw issues and no sw support?  Only way is if they are making money off of selling a few boards or are heavily subsidized.

Try and error? until you find a board which sells good. Or it's not needed to make profit in the first years/months.

 

13 hours ago, NicoD said:

The BPi R1, R2 and W2. All above $50.

I think the R1 had issues with the switch (never owned one, you may ask other or use the search engine)? R2, never saw someone who reached GbE speed with mainline.. at least I didn't. W2 no idea, it sits there since a few weeks but I didn't have enough time to dive into the adventure (there are still a bunch of open questions for the board/SoC - mainline is more or less non existing, let's give it some time to see where it ends). :P 

 

5 hours ago, e97 said:

back to an x86 server but the power usage will be high

you've a x86 2GbE server vor <50$? or just reuse an old one? Power-consumption for a 24/7 system should also be considered.  Thought about a ARMADA A388 based one? e.g. ClearFog Base?

13 hours ago, e97 said:

Having manufactured in China gives me insight blockquote widgetht into the costs to produce electronics. I'd bet at least 10x margin from researching all these kickstarter boards.

Funnily complaining about software/hardware support but then setting a price-tag which is IMO a way to low.. Support is only for free when people waste their spare-time.. :lol: Otherwise it's part of a support contract or gets added at initial board-price. 

Share this post


Link to post
Share on other sites

Just sharing... 

 

Old project - looked at Armada 3720 and TI's Sitara (AM3352) - both had fairly flexible interfaces for Gigabit Ethernet... Sitara was interesting as the MAC's could be implemented as switched interfaces.... Armada had faster/better interfaces... so the choice was simple....

 

Project did two runs of HW - first 20 P0 boards were "stretch" boards, second run (P1) was in the Pi format - mostly to reduce overall costs - think Pi board without HDMI/GPIO, microUSB as console, power on barrel connector where Pi's do audio in/out - swap the USB/Network connectors for 2 USB and 2  1GBit connectors, and you get the picture... With no WiFi, we were able to crowd everything on to a single sided board to reduce manf costs - including 2GB RAM and an 8GB eMMC - it was a clever design - and I'm still happy to have done it.

 

What killed the project wasn't the Bill of Material - it was all the ancillary costs, because of agreements/constraints - basically everything else besides the board itself - building/running a business as a HW startup... We couldn't do manf in Shenzen, we had to do in US (contractual restraints), and then all the logistics... Regulatory - FCC Part 15B can be a killer in and of itself - to be legit, you gotta pass that one, and it's $80K just for that testing per run - 15B is unintentional RF radiators... luck was that the P1 boards did pass...

 

GlobalScale's EspressoBIN offered more functionality at a lower price to the customer... we just couldn't touch that price and with a very narrow market, just didn't make sense to continue - we would never sell enough units to recoup expenses as an ongoing business...

 

Our exit on the project - we sold the design to a white-box shop out of Japan that had Marvell in house already - older stuff based on Kirkwood and Orion - they were more interested in the BSP we developed rather than the HW itself - SW guys did do a back port over to Armada 38x for the SW side - they've since moved to Marvell ARMv8 - I'm not involved anymore on that one...

 

Anyways - getting back on topic - the EspressoBIN is likely the best choice for a multiple port Gbe board at the $50USD price point - it's a good board, yes, it has quirks, but still is better than pretty much anything else at the moment at that price point...

Share this post


Link to post
Share on other sites
15 hours ago, chwe said:

shouldn't vlan exactly solve such issues/concerns? I think it needs a vlan capable switch as well.. No idea, never spent much attention to VLANs. 

there have been VLAN CVE, rather encounter a security issue down the road - eliminate the problem with separate NICs. simple is always best

 

Quote

do those MT7621 boards get regular kernel updates? Otherwise this would IMO be a bigger security issue. I've no experience in MIPS world.. Only an old ASUS router and it was just the wrong device.. :D 

Isn't it a kickstarter board? :P

 

Mainline kernel has support. You may lose some HW features but patching is not too hard, fortunately I only need firewall and QoS so that shouldnt be an issue. yup it was the the last board I ever helped kickstart. Now I wait to see if a board survives to mass production - like the Pi. I did back Parallella and was pleasantly surprised but that speaks to the caliber of the people working on the project. The founder is now a DARPA AI/compute project lead. I am pretty excited about the Ryzen V1000 boards coming out.

 

Quote

 

or the company goes bankrupt and a new one tries it.. Or the cut support for a board which doesn't sell well.. :D

 

if you look at the "companies" releasing the boards, you'll see its the same people under a different "brand" name.

 

Quote

 

Try and error? until you find a board which sells good. Or it's not needed to make profit in the first years/months.

 

If you have lots of money to burn and want to encourage more bad boards, sure buy them all. I'll wait and see what survives. The smaller companies have to make a profit to survive unless they have tons of investment or are part of a much larger company. I mean look at Asus Tinkerboard, it still has issues and it used a few year old SoC and was made by a Tier 1 x86 motherboard manufacturer.

 

Quote

I think the R1 had issues with the switch (never owned one, you may ask other or use the search engine)? R2, never saw someone who reached GbE speed with mainline.. at least I didn't. W2 no idea, it sits there since a few weeks but I didn't have enough time to dive into the adventure (there are still a bunch of open questions for the board/SoC - mainline is more or less non existing, let's give it some time to see where it ends). :P 

 

Yes, have given up on any type of router boards - looking at the block diagrams and datasheets will reveal numerous issues..

 

Quote

you've a x86 2GbE server vor <50$? or just reuse an old one? Power-consumption for a 24/7 system should also be considered.  Thought about a ARMADA A388 based one? e.g. ClearFog Base?

 

Old x86 server I have around, it E3 Xeon uses ~25W since its server hardware. You can get older Xeons on ebay for very cheap.. But thats still 5x more vs a typically router ~5W. I forgot I also got a Linksys AC1200 refurb for $USD 40 which has an A385 :). Was testing LEDE and ZFS on it. Will be putting debian on it now and will likely meet my needs.

 

Quote

Funnily complaining about software/hardware support but then setting a price-tag which is IMO a way to low.. Support is only for free when people waste their spare-time.. :lol: Otherwise it's part of a support contract or gets added at initial board-price. 

 

Yes, that's why industrial boards are expensive in addition to better spec-ed components. But I'm not looking for an industrial board - just one that has proper hardware at least, the software I can be fix. However, the NXP board I mentioned has an EXCELLENT BSP and hardware and support and its $50.. perhaps its subsidized as NXP to sell more chips since NXP is usually pretty expensive SoC-wise. I'd like to see any other manufacturer come out with something of comparable quality.

 

2 hours ago, sfx2000 said:

Just sharing... 

 

Old project - looked at Armada 3720 and TI's Sitara (AM3352) - both had fairly flexible interfaces for Gigabit Ethernet... Sitara was interesting as the MAC's could be implemented as switched interfaces.... Armada had faster/better interfaces... so the choice was simple....

 

Project did two runs of HW - first 20 P0 boards were "stretch" boards, second run (P1) was in the Pi format - mostly to reduce overall costs - think Pi board without HDMI/GPIO, microUSB as console, power on barrel connector where Pi's do audio in/out - swap the USB/Network connectors for 2 USB and 2  1GBit connectors, and you get the picture... With no WiFi, we were able to crowd everything on to a single sided board to reduce manf costs - including 2GB RAM and an 8GB eMMC - it was a clever design - and I'm still happy to have done it.

 

What killed the project wasn't the Bill of Material - it was all the ancillary costs, because of agreements/constraints - basically everything else besides the board itself - building/running a business as a HW startup... We couldn't do manf in Shenzen, we had to do in US (contractual restraints), and then all the logistics... Regulatory - FCC Part 15B can be a killer in and of itself - to be legit, you gotta pass that one, and it's $80K just for that testing per run - 15B is unintentional RF radiators... luck was that the P1 boards did pass...

 

Thank you for sharing. Most people don't know the difficult to bring a physical product, much less consumer electronics to market. Fortunately, one of my advisors helps manage supply chain for a certain fruit company. Their advice and experience was invaluable.

 

After being burdened by one investor, I have learned to CAREFULLY read the legal and informal obligations of any contract/partnership and look for potential pitfalls, lest I be shackled again. One SoC vendor wanted free reign over our IP as they could be "independently" developing similar products and they had already released a similar product with a competitor. I declined. It sucked having to search for a new SoC vendor but it worked and we sold a good product. A year later we started talking again for a new project and they are more flexible having seen our success.

 

Its unfortunate you couldn't manufacture in China, because then you could also do the regulatory testing/certifications in the area (HK, Taiwain or Shenzhen). The same testing/certification facilities used by Tier 1 motherboards manufacturers will provide testing services including the one you mentioned for USD $5K - $15K depending on what all you need. RF products will cost more.

 

Quote

 

GlobalScale's EspressoBIN offered more functionality at a lower price to the customer... we just couldn't touch that price and with a very narrow market, just didn't make sense to continue - we would never sell enough units to recoup expenses as an ongoing business...

 

Our exit on the project - we sold the design to a white-box shop out of Japan that had Marvell in house already - older stuff based on Kirkwood and Orion - they were more interested in the BSP we developed rather than the HW itself - SW guys did do a back port over to Armada 38x for the SW side - they've since moved to Marvell ARMv8 - I'm not involved anymore on that one...

 

Glad to hear that you had an exit. Congrats :)

 

Quote

 

Anyways - getting back on topic - the EspressoBIN is likely the best choice for a multiple port Gbe board at the $50USD price point - it's a good board, yes, it has quirks, but still is better than pretty much anything else at the moment at that price point...

 

Yes, out of all the community boards the EspressoBIN seems OK. The Marvell chips are fairly robust and the capabilities/performance aligns with the datasheet specs. However, after discussing the boards with GlobalScale I got an errie feeling similar to working with other SoC manufactures so I backed out and wanted to wait until launch. It was delayed over half a year and the end result is what you have now. They did release some Google Compute related boards so perhaps they are more focused on that but I got the sense they didn't really care about the community version board.

 

I mean everyone wants Raspberry Pi level success and they seem to think the form factor is what does it, when its the fact they to long term support and committed part of their team to continue improving the BSP. "More wood behind fewer arrows". Rock64 says the same thing but shows different results. Releasing multiple boards with different SoCs with a small team will spread any small company too thin. Current SBC methodology is fire and forget and see what sticks. That generally works with software products but definitely not with hardware. Hence the bucket of EoL ARM devices that I and many others have sitting in the corner collecting dust. Hence why I refuse to support anymore campaigns without showing substantial thought into the product. Things do seem to be getting better as a couple mfgs are joining mainline kernel development and submitting patches and supporting projects like armbian. Just wish it didnt take companies 5 years to realize this.. The first "SBC" MK802 with Allwinner A10 was what changed the game back in 2012 and kicked off this SBC revolution and helped create a space which eventually gave us the Pi though the story with the Pi is it was built to get kids back into computers. Anyway I'm confident things will get better one way or another!

Share this post


Link to post
Share on other sites

Summary of potential boards:

 

NXP FRDM-LS1012A               1x 800 Mhz      512MB      2x NIC                      -                         = $50

GlobalScale ESPRESSObin      2x 1.3Ghz           1 GB         1x WAN, 2x LAN      -                         = $49

Linksys WRT1200AC (refurb)   2x 1.2 Ghz       512MB        1x WAN, 5X LAN     2x2 802.11ac     = $40

Share this post


Link to post
Share on other sites

 Marvell ESPRESSObin

Marvell Armada 3700LP (88F3720) dual core ARM Cortex A53 processor up to 1.2GHz,1xWan + 2x Lan GbE, 1x sata, 1x USB2.0, 1x USB3.0, 1x micro USB port,1x MiniPCIe

 

Share this post


Link to post
Share on other sites
19 hours ago, e97 said:

Thank you for sharing. Most people don't know the difficult to bring a physical product, much less consumer electronics to market. Fortunately, one of my advisors helps manage supply chain for a certain fruit company. Their advice and experience was invaluable.

 

After being burdened by one investor, I have learned to CAREFULLY read the legal and informal obligations of any contract/partnership and look for potential pitfalls, lest I be shackled again. One SoC vendor wanted free reign over our IP as they could be "independently" developing similar products and they had already released a similar product with a competitor. I declined. It sucked having to search for a new SoC vendor but it worked and we sold a good product. A year later we started talking again for a new project and they are more flexible having seen our success.

 

Its unfortunate you couldn't manufacture in China, because then you could also do the regulatory testing/certifications in the area (HK, Taiwain or Shenzhen). The same testing/certification facilities used by Tier 1 motherboards manufacturers will provide testing services including the one you mentioned for USD $5K - $15K depending on what all you need. RF products will cost more.

 

It's really hard to bring the first product to market - once that's done, the follow on products are probably easier, as much of the business infrastructure is already in place - and perhaps time to tune and tweak based on lessons learned from the first one.

 

With startups - yep, investors do have a say, so choose them wisely - hopefully one that has previous experience, and connections that can help down the road, getting into their "deal flow" when possible.

 

Going to SoC's - it's really hard to work with many of them - some for reasons like you mention, others for other reasons - for example, working with their existing partners in the OEM/ODM space, and yes, IP is always a concern - even to the point where one's own IP will be at risk - and then just the competitive risk, as soon as you ship a board at a certain pricepoint, someone else is going to clone it, and do it cheaper...

 

Going with US chips - well, there's Broadcom and Qualcomm - which are tough to deal with and expensive - Marvell, if it's in their interest, they'll work with you, TI and Freescale/NXP - TI is ok, Freescale used to be easy, but with NXP in the picture, things changed....

 

Production in China - that can cut both ways - I've done business previously in Shenzen, and the gongkai process/culture does have its merits... that and the supply chain benefits along with everything else can bring down the NRE significantly, and production costs are obviously lower, even then one has to be very smart about who to work with there - some are better than others. What I can say about working with US vendors vs. the PRD - if price/cost is not an issue, I'd rather work with US vendors since I'm based in the US, it's just easier - it's the time differences for one, logistics, culture even - but that's a business item - one can get good product anywhere, and it's hard to beat the Shenzen area for products like single board computers and the like.

 

Current SBC methodology is fire and forget and see what sticks.

 

FriendlyARM did a great run on their Allwinner series of board, IMHO, and like I mentioned above, once the business is sorted, and the first products launched, it becomes an iterative process - I'd add that Libre Computing (aka Shenzhen Libre Technology Co., Ltd) has made a good jump across different lowend Chinese SoC vendors (AllWinner/Rockchip/Amlogic) with a fair amount of success.

Share this post


Link to post
Share on other sites

2018-11-4 update

 

Looks like Amlogic S905x2 SoC boxes are being released.

quad Core A53 + 4GB + 32GB ROM + gigabit + usb3.0 ~ $57 + $10 USB 3.0 gigabit  RTL8153 adapter = $67

 

Amlogic has decent BSP support and arguably better color reproduction, good for TV boxes, not relevant for network appliances

 

Almost there, with armbian or arch support and $17 discount this COULD work.. don't get your hopes up

Share this post


Link to post
Share on other sites
37 minutes ago, e97 said:

Almost there, with armbian or arch support and $17 discount this COULD work.. don't get your hopes up

if you don't do it on your own, a TV box will likely never get 'official' armbian support.. maybe @balbes150 supports it. Even the Khadas Vim isn't supported and it's more likely a SBC..

 

Messing around with TV boxes is always a mess. They change things (e.g. change wifi chips, ram etc.), there are no schematics available and nobody has capacity to deal with such stuff. Better go for the cheapest RK3399 based TV box and buy the same USB GbE dongle.. Where you have proper mainline support and for multimedia use-cases chances are high that this stuff can be enabled with the BSP kernel.

Share this post


Link to post
Share on other sites
On 11/4/2018 at 4:41 PM, chwe said:

if you don't do it on your own, a TV box will likely never get 'official' armbian support.. maybe @balbes150 supports it. Even the Khadas Vim isn't supported and it's more likely a SBC..

 

Messing around with TV boxes is always a mess. They change things (e.g. change wifi chips, ram etc.), there are no schematics available and nobody has capacity to deal with such stuff. Better go for the cheapest RK3399 based TV box and buy the same USB GbE dongle.. Where you have proper mainline support and for multimedia use-cases chances are high that this stuff can be enabled with the BSP kernel.

 

I'm well aware. To be clear, I DO NOT expect ANYTHING from the community. If someone shares an interest and shares info/knowledge/code GREAT, happy to work together :)

 

Just documenting my thought process here as it may be helpful to others and shed some insight into SBCs. Its important that we improve the visibility of products and manufacturers that listen to feedback, continually improve their products and work with the community, not expect any support to be handled by the community.

 

HINT: If a board manufacturer lists forum/community under support - run! Run as fast as you can, as far away from them as you can.

Share this post


Link to post
Share on other sites

Success!

 

Using a minimal distro and an SBC with min specs:

 

+ dual core @ 1.8 Ghz

+ DDR4

 

I was able to get 850+ Mbps firewall with QoS!

 

Tested multiple methods:

 

- integrated NIC + USB 3.0 Gigabit adapter

- integrated NIC + PCI-e NIC

- dual PCI-e NICs.

 

For under $50 and under 15W

 

Used this guide https://blog.tjll.net/building-my-perfect-router/ as a starting point. Used unbound for local dns reslover and a few other changes to traffic shaping and metrics; using prometheus and grafana

 

There's still room for improvement, get power under 10W and maybe closer to 5W-8W.

Lower cost if there was a board with integrated dual gigabit NICs !!

Share this post


Link to post
Share on other sites

Haven't decided on a final config becuase theres more testing to be done.

 

Easiest would be Atomic Pi (x86), your fav minimal distro and a USB 3.0 to Gigabit Adapter, about $50 shipped in the US (not including power supply)

 

Cant recommend any MIPS / ARM boards because it requires backporting netflow to your kernel and/or SIMD optimization. If you can do that, you can find the boards from the specs listed above and take into account the state of firmware/software to make an informed purchase.

 

 

Share this post


Link to post
Share on other sites

No, I mean the SBC you used in your "success" message above, where you say you got 850Mbps throughput (with or without NAT btw?).

 

Share this post


Link to post
Share on other sites

Add Raspberry Pi 4 with USB 3.0 adapter!

 

RTL8153 works OOB on linux

 

 

 

On 6/7/2019 at 5:57 PM, olivluca said:

No, I mean the SBC you used in your "success" message above, where you say you got 850Mbps throughput (with or without NAT btw?).

 

Modem <-> SBC <-> Switch <-> Devices.

 

900+ with NAT on multiple SBCs.

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
2 2