I have had enough RAM fail on me to want it to be swappable in some manner, even if you say we only sell/support DIMMs on our QVL (Qulafied Vendor List) and only have 2 SKUs on that list. Simular for the eMMC, I don't mind it being there as the default, but I would want a m.2 slot or similar that I can use if it dies or I have requirements for something larger or more proformant for something like a long packet capture or snort or whatever. I would also want to make sure that if the eMMC dies the mobo is not reliant on it to boot/use. Just some ways to reduce e-waste and make service easier. And again, you can have a QVL for the SSDs/nand/whatever and anything else is the users/communities problem to make work.
This is an amazing project to follow and very educational to see the process and decisions you make along the way, and I appreciate how well and clearly you explain things so that anyone can understand. Thank you for sharing your journey
Whilst I understand the economic drive to use ARM (and ARM is undoubtedly the future), this leaves difficulties with the choice of OS to run on it. You previously mentioned VyOS and OPNsense but both those are x86 only (at least in official builds). There are third party builds of OPNsense for ARM and experimental builds of VyOS for ARM, but neither are really ready for mainstream use on ARM from what I can see. Are you planning to fork/port either of them fully to ARM? As an aside, I am a little sad to see soldered main storage and RAM. It is limiting, not least in the case of repairability. And "8GB should be enough" has a good chance of being the next "640K should be enough".
Awesome to see the process behind this. However it seems too expensive due to the case costs, would also be nice to have 2.5gbe instead of 1gbe. Maybe the aluminum case should be an optional extra, with a standard plastic case?
Your journey inspired me to get the documentation from my first year in university and learn embedded systems again. I plan for january to get my first test PCB printed :)
Oh gosh no, just something I think is easy and doable. I want to create a digital pictureframe and a custom, (hopefully) small air quality probe, maybe with an option to integrate it to homeassistant. @@3nertia
As someone who has several ideas I'd love to see to market - including my own router design that I've since scrapped lol - I would just like to thank you. This series has already been insanely valuable in helping me to organize my thoughts - subscribed!
You should make the case optional for people willing to save. Self 3d printing a plastic case and just adding a heatsink+fan (as you don't have the aluminum case anymore) would make it considerably cheaper.
I find a small display very useful, especially if things don't work as expected. Small 0.91" OLED displays 128*32 Pixels cost about 2$ and would just do the job.
I recently found this channel and watched few videos, first one was ISP router. Loved it, keep it up. new sub. This is an amazing project, learned a lot more than I thought. Good luck
This is a great project. I'll be raising my budget when it becomes real. Why not have 2 SPF+ ports instead of 1 and a RJ45 port? It would save $10 per unit and board space. If the customer was bringing a RJ45 cable, they can get an adapter. I think using a USB-C for all connectors is fantastic. I use a 200W USB-C hub on my rack of utility devices
Thanks for the feedback! Fun fact, I'm releasing a new video next week addressing these issues specifically! We removed the 10Gb RJ-45 in favor of another SFP+ and also moved from 4x1Gb to 3x2.5Gb RJ-45 ports because processor supports that out of the box.
Are you not considering a direct supply purchase rather than through distributors? I do realise that you may be forced to do this due to the uncertainty around the potential sales figures and the bulk purchase obligations you might face with buying direct: however, is it not worth a discussion with a sales/engineering rep in parallel with this? You might be surprised. Via a distributor, you'll always be paying for their profits & overhead. Good analysis so far, with good, well-explained simplifications of the choices made. Excellent! This deserves to suceed.
I forgot to mention: might be worth pricing at 1 off, 10 off, 100, 1000, etc. production runs, and costing the production total/unit at each quantity. You can then graph it/build a pivot table to look at the best minimum unit sales figure to aim for. I know you're doing part of this with your market survey, but a graph might be interesting!
Thanks for the input, I'm well aware of the overhead but I found a distributor that's really helpful. Like if I give them a call and ask for their opinion, they'll pick up and talk to me, and help me find a solution. Despite the fact I haven't actually purchased anything yet. If you ask me, they're well worth that overhead. Plus the offers I got are super competitive. Stay tuned, a graph is coming in Tuesday. Not exactly the same one, but you might find it interesting nonetheless!
Came across this video as I was wanting to upgrade my network and intrigued on what you’re wanting to do and trying to figure out what I would like to do. In the rabbit hole of networking videos and various searches I came across the minisforum ms-01. It has two 10g sfp+ and two 2.5g nics. All of them appear to be intel based. This seems to fit a little bit of what you’re wanting to build and at a minimum could be used to compare your design to.
My experience tells me that you will have problem with bandwidth performance. This kind of CPUs really need their offload capabilities in order to achieve the bandwidth. You wanted 1Gibt/s IDS, that means that packet analysis probably have to be done in the main CPU, in that case the bandwidth the CPU can handle will be the upper limit. There might be solution but I don't think that it will work out of the box without special drivers.
As we're very early in the development stage, I have the luxury to change things. I'm waiting for the development board, and once I get it, I'll test it thoroughly, to make sure it can satisfy the requirements. Obviously, I will make a video about it (and not just one, most likely)
I just filled survey, one thing to add since it wasn't possible to add "custom note" there: Idea is cool, but I don't like formfactor. I would rather prefer 1U format or half-width U. And solid aluminium block is not requirement... I would happily go with simple metal sheets for enclousure if that could bring price down.
Thanks for the feedback - no worries, there's been a lot of changes, and once I get around to writing them down, I'm going to present them in a video. Will fit in 1U. :)
Please allow socketed ram with either 2 sockets/channel or support for 16gb modules optional. You saving support and reducing variance does not cater to your audience. Not to mention that with NIDS and nfv(containers) ram becomes useful. Same is true for endurance of your disk if you want to traffic log.
And this is strictly hardwire. I would still need to add a WAP at some point. This is fascinating but also quite the reality check for how much good hardware costs.
hey quick note since it's still in the bom (31 Dec 2023 02:10) I've noticed that at 15:46 and 15:55 the quantities are given in 2000 units instead of 1000 units and the price hasn't been cut in half in the BOM accordingly (or canceled the function) (Edit: it happend in many more placed throughout the video ^^' once i started to notice it i was seeing it all over the place. if it's corrected or not is not on my behalf and probably won't change anything, just my OCD is being triggered ^^') otherwise great video, even more so for the unexpected base level excel course
You need one? I would use a PC for that. I build DIY Firewalls with PC (Server only) parts for my customers all the time because i don't need all those additional fatures of the full feldged firewalls like Meraki etc.
Forgive my ignorance, but 599 seems REALLY expensive for a router. How is this better than slapping a bunch of older hardware together and sticking opnsense on it?
Size and consumption, primarily. Its not the same to have a computer consuming 200W that a small router consuming 15W. You could buy a fiber pcie card for about 70€ and a 2.5gb for 30€, and for a lot of people that would be enough. In my particular case, i could not afford 750€ ('cause taxes are paid, you know) for a router and i am not in the spectrum of "people willing to buy this kind of router", but its interesting :D
What makes it much more expensive than your regular household router is 10Gb capability. Well get to the reasons why in one of the following videos, but having just two 10Gb ports gives you like 5x the capability of your "normal" of the shelf 4x1Gb router.
For me personally arm, soldered ram and emmc is complete deal breaker. For me it would have absolutely no other use than just being nice looking router. For like 300-400$ i could buy some qnap or something like this and if I wanted my router to be something more for 600$ i can buy some used hw with x86 and much much much better cpu, ram etc missing just the looks and small power consumption
While the case is nice, I wouldn’t want to have to spend ~30% of the device’s price for it since this will be sitting in a closet. Maybe consider also selling a bare kit for someone that can 3D print their own case. Or see what a bent sheet metal case would run. Ubiquiti gear is a good example of simple but much more affordable aesthetic. BTW, I’d tone down the how to on gsheets - if someone doesn’t know how to setup a basic summing formula, they aren’t building a BOM for anything.
Just one comment about pricing, when I see a price like 599, I always think that the manufacturer wants to trick the buyers that it is a 500 something product which I find it a bit dishonest. Or may be the manufacturer thinks that the customers are stupid not to realize it is 1€ less than 600. Just ask for 600 please. I am sure your target customers are wise enough to evaluate the price correctly.
Psychologically, there are people out there who do evaluate $599 to $500. How? I have absolutely no clue But there are also considerations for shipping, VAT, and other costs which may be more expensive if the product has broken a barrier (for example, $600) Just something to be aware of when you see products that aren't a flat value
It's funny you mention an overpowered device.Device. I have 16 gig of RAM in my OPNsense firewall, thinking I would need the extra RAM. In the end I am just using over 3 gig of RAM with IDS/IPS, VPN, and Zen Armor running.
There are some smaller fabs thst make memory. But that’s less common with ddr4 and not happening yet for ddr5. I pretty sure oloy used those none big 3 memory chips on there low end ddr4 Modules. I believe they are called nanya
Really interesting project! Couple of questions though. First, why two different port types for the 10G? Wouldn't two SFP+ ports make more sense? Yes, I know, many use RJ45 in the home, but given the option, adapters are not that pricey! Second, why 4 1G ports? Here in the US, 2.5G is becoming more common as a baseline... Does NXP have a chip that would handle at least two of these ports, in addition to the aforementioned 10G? Just wondering... you have stimulated the tinkerer side of my brain!
First of all, Thank you for doing this. This is a project that needs to be done. Secondly Why not a Watch Timer? Or is that included in another BOM Part? Third, For the love of all that is holy, please consider a sheet steel case, and bring your BOM down!
Thanks for the kind words. This is not the final BOM, we will cut the costs down, but that'll come in a separate video as it's an important topic on its own! Stay tuned! :)
I am highly interested in this project, as I have been shopping to buy or build a 10 gig router for home use for several years. I've looked at more server level boards than I care to count, and the best value I've been able to even consider is Netgate's 6100 router, which weighs in at $800 USD. But it is still missing a feature I would highly like to see on a router, which is ECC RAM. I see from your BOM selections that you did not choose ECC RAM either, and I wanted to ask if you have considered it, and if so, what the cost difference is with ECC RAM. Thanks!
I myself would be at least interested in hearing/reading the advantages versus disadvantages of having versus not having ecc as well I'm aware that "ecc" is error correction but I don't really know how much value it actually has overall and if it would be worth the potential markup to the overall cost of the device
@@3nertia I read a study done years ago about random bit flips in computer memory and found that with memory densities as high as they are, they were seeing an average of 1 bit flip a month for each GB of memory. So for 4 GB you’d average one a week, and 32 GB would average about one a day. If the flip happens in unoccupied memory, nothing bad happens and life goes on. If it happens in a chunk of data, it’s possibly silent data corruption if there isn’t any validation. If it occurs in a critical piece of OS memory, it could result in a system crash. In desktop machines, which usually don’t have a long uptime, this problem is usually academic, but in longer running systems, those flips can potentially stack up and cause larger problems, so it’s good to have ECC memory to detect and correct single bit flips, or be able to alert the OS to the problem in case of double bit flips, which it can detect but not correct. Then the OS can decide how to handle the situation. Hope this was useful.
Hello, I know the 2013 Excel could pull prices from web pages, so I am sure Google Docs can do the same. that will keep your bill of Materials current. If I was building a dream router I would want it as future-proof as possible which means 2.5 Gbs ethernet ports for the LAN, as I am starting to see them come up more and more. For the WAN ports, I suspect you would want a 2.5 Gbs port that is compatible with a cable modem that has 2.5 Gbs ethernet. I don't know enough about the 10 Gbs ports, are they compatible with the 2.5 Gbs ethernet, then you could save a port. CPU needs to be powerful enough to speed through whatever software you are going to be installing especially at that 10Gbs pace. RAM you obviously need enough for all the necessary software to fit in it, that is your minimum size. But you know your power users are going to want expandability, so have a way to double the size. You did not talk about WIFI, if you decide to include it I suggest having it upgradable with the 2.M socket.
Thanks for the feedback! Building the BOM here was more of an exercise for the viewers and to drive a couple of other points across in the future videos, but you sure do have some valid points regarding 2.5Gb ports (10Gb already covers those needs) and wifi (likely an M.2 or mPCIe card).
With the current BOM price, I think this will sadly be DOA. Qotom recently released a product that offers more connectivity, has more performance, and is sold for a retail price only a few dollars more than what your BOM currently is in v2. Its also x86 so you can easily install any of the main router distros on it. I really want this project to succeed, but at this point I just don't see how it can when a more well-known competitor offers something better for cheaper.
All good. I'm not interested in competing with Mini PCs. If someone prefers that, that's okay, competition is always healthy. Currently working on a video in which we brought BOM down in half.
Why do you want to only use 1 RAM chip? you can use 2 in a T junction setup and that may drive the price down a lot overall but it is harder to design. Dedepending on that CPU you may also be able to use 4 chips that that will drive cost down a lot. 60$ for the PCB+ assembly seems cheap for a semi small amount but if you produce a lot it may well be even lower.
@@tomazzaman The workaround might be to just have 2 10SFP+ ports that can take one of the 1/2.5/5/10g SFPs and then 2 10/25G ports for LAN. We are seeing ISPs deploy gear with 2.5G ports and there are some places that are pushing out 3Gbps service, then you have super cool places like chatanooga that have 10Gbps for like $299/mo for residential.
Respect, what is you target and how do you try to reach it! :-) I think , the appereance is not so important, as you think. Sorry, this is only a router, which is normally not the part of a living room or a clean desk. It should be looks fine, but the contents/ingredients are much more important! This is the factor, where you could win a lot.
Do you really need the CR2032 for the RTC? Can't you just sync the time via NTP on bootup? What do SBCs like the Raspberry Pi do for it? I don't think they have a battery.
How does this make sense when the Qotom Q20311G9 exists now with 4x SFP+ for 10G and 5 2.5G RJ45s? Its around $450 even with a 10G SFP+ transceiver, RAM and SSD.
Another idea after I watched the dev board. The ram is it going to be soldered or a stick? If soldered, why? Just add a stick of 8GB or give the people the choice to go with or without memory. I just looked up and a stick of 16GB DDR4 3200MTs costs less then 10€ and with the change to ddr5 there's going to be a lot of ddr4 coming free :). But then again you need to test different dimm's to see if they work or not. Mmmmh a single slot for ddr4 and a replaceable stick of ddr4 or soldered mmmhhhhhh Or make it ddr4 sodimm, connectors costs less then 2€ memory itself is a little bit more expensive /GB (8GB 320MTs for les then 10€) but hey less vertical space needed WHAAAAT 30$ for 8GB and not replacable, don't like that. Please change that.
RAM is the one area I'm not 100% certain yet of which way to go. Luckily I do have a couple of meetings with people way smarter than me on this topic, and I'll see what the expert opinion is. SODIMM or SOLDER? 😂
Maybe some daughterboard options to cover different types of interfaces? For example relocating the RJ45/SFP ports to swappable modules would give some amazing flexibility, but maybe not economical in the first gen product. It would certainly be great to see on future versions though
I'm sorry but I have to make a remark: I DESPISE the idea of soldered ram. I have a certain number of machines running basically 24/7, depending of the years, 10 to 20, and in the last decade, it's 5 sticks of ram that failed me. With soldered ram, it wouldn't have been 30 minutes of downtime. It would have been hours to repurpose something else jank-ily to days if not weeks for a replacement. If we really are building the great home router for homelabers/tinkerers/ tech enthusiasts, the ram shouldn't be soldered. It isn't in tiny clients many use as nodes. And the ones who run enterprise servers in their closet know how to deal with reduced ram compatibilities. I'd rather have sticks limited to a peculiar frequency, CAS latency, or even timings, or even blow a few dozen bucks on some failed compatibility-checks than have to whip out the rework station or buy a whole new router because of a failed ram chip.
Thanks for the feedback! We're trying to strike a good balance between expandability, reliability/quality and price. The videos I'm making on it aren't anything final. And I'm making them precisely for this reason, so that you and the rest of the community can voice their concerns. So, thank you!
I had more problems with exchangable Cisco RAM in the last 20 years than with others. And with obsolete Hardware after 5-7 years the RAM is not the limiting factor anymore. What i want to say is: the hardware of every manufacturer, no matter if DIY, Sonicwall, Meraki must be exchanged after not more than 5-7 years anyway because it can not handle the next Software upgrade anymore.
@wolfgangpreier9160 I can see where you're coming from, but don't forget the router project isn't meant to cater solely to enterprise/professional use with a renewal cycle of half a dozen years. How many homelabers are running proxmox on 8-10 years old Xeons that are considered to be near e-waste at this point. From 2015 to 2019, my homelab's router was a thinkpad with a 3G cellular wan fail-over. I'm currently running an OPNsense firewall (and router for some sub networks) on an AMD Geode thinclient from ~2015 with a LP PCie NIC. Because I wanted my firewall to not be concerned by Intel's Management Engine nor Ryzen's AMD PSP. It's not fast by any means, but sufficient for the throughput it's used for... and that's what I had on hands. And being able to replace the mSata drive when it failed during covid, and I had a couple of them collecting dust on a shelf was also very useful. //// Back to the router at hand though... at 10gbps (if one of the 10G ports can be used as WAN, I have a 2.5G uplink atm already), I could see myself using@@tomazzaman's router for the next 15 years, eventually DDwrt'ing it when it get end-of-lifed? It just has to not have a chip die, then I'm ready to invest once into a piece of brand new hardware that would cost as much up front as 15years of jank solutions built out of e-waste that do the job all the same. I would even wanna pay a 10% upcharge for it to come barebone, without any storage or ram, and just a couple of SODDIMs and m.2/msata SSDs validated for it. But all in all, it's my personal feedback and feeling. Whatever path you go for, given your current direction, go for it, it will find an audience considering the clear demand for better network solutions. For instance, on top of just part-swapability, instead of ARM-based, I'm yearning for a RISC-V (or other opensource architecture) router, where there is no arm trustzone or blackbox 'security' by obscurity paradigm.
Excellent Post !! Soldered RAM is one of the biggest forms of e-waste EVER. Sorry for the caps. Got myself a CWWK i3-1315U where I can put 64GB DDR4 RAM on there (should of used credit card & got higher end one instead of using debit card). Also 2 NVMe's and 2 regular SSD's ( 2.5", but can't fit a 2nd one in the case :) However I've been looking at older... what are they called Micro, Mini PC's. The HP EliteDesk 800 G3's, G4 or G5's that I can run Proxmox and pfSense CE, OPNsense and numerous containers on. Well might have to run Proxmox and numerous Docker containers at some point in the future as my CWWK may not be able to handle what I have in mind after all. Then I'll do all those things and say, damn I don't need to run all those things and be fine with it Afterall :) But Please No Soldered RAM. Think how many millions of things get tossed world wide each and every year just because of it. Total Waste. @@Varengard
I was very interested in this project until you decided to solder the RAM on. I think you will lose a significant portion of your potential market for a boutique router, by prioritizing the reduction of support costs over the flexibility of RAM replacement. Thereby creating yet another nonrepairable, nonupgradeable device.
shift your tone away from the 'influencer' shit, that's not your target audience. Use the best parts, or what's the point. I'd rather buy 2 'solid' cheap chinese routers over a 'medium-high end' compromise. Make an enterprise grade switch for prosumer prices, no cutting corners, no RBG, no frilly cases etc. Be different enough to be interesting, don't just make a pretty mokerlink. Pay that dollar for the 10G rj45 port, or get rid of it even... Powerful, simple, elegant: in that order. That aluminum case is so dumb unless you've plans to consume enough power to need it as a heatsink, and you don't even have the power supply internal.. Steel is cheaper, stronger, heavier, and conducts heat just fine. Magnesium alloy casting might even be the way to go if you can justify tooling costs.
I have tried 2 cheap chinese routers and put them away after 2 hours of not working. I don't have the time and patience to play around with this things. They either work at once or never. I had to put in the effort learning mikrotik and since then i'm somewhat familiar. Much too convoluted and complicated but they work. 10GB SFP+ needs good cooling. Maybe thats the reason for Aluminium?
I would NEVER EVER pay even 300€ for a router. Not for my home or private use. Just because, who the hell needs that at his home? For a company thats another question. But there are very good brands that make good routers already. So I dont get the idea behind this at all. I am just here to tell you, who would NOT buy such a thing! Maybe helps with the calculations.
I have a pricing survey form out which currently has almost 800 submissions and currently the average that people are willing to pay for is 380€. The average. Which means half of the 800 are willing to pay more. And obviously, half are not. Thanks for the feedback!
Since you worry about compatibility for end users and the possibility that hundreds of different RAM sticks might be used by people who buy this, perhaps something you could do is to buy and test with just 6-8 different sticks of RAM that are easy to buy around the world. Maybe 3 sticks at 4GB each from 3 different main manufacturers, and 3 more 8GB sticks from those same manufacturers. That way you have a QVL of known compatible memory people can use as a reference but also give people options. This lets people have options by not being soldered onto the board but also lets users know of specifically tested sticks they should be trying to use.
@@LordSaliss he is not building a new ram controller, he is using an off the shelf one, their support, documentation and stability is robust. It seems the crap apple does rubs of on their customers, locking in and deciding on behalf of the user without giving the user any options
If this will be under 500 euro, consider me your first Romanian customer 🎉
Thank you! I highly doubt I'll be able to optimize it to that price point, but I'll certainly try to get it as close as possible! :)
I have had enough RAM fail on me to want it to be swappable in some manner, even if you say we only sell/support DIMMs on our QVL (Qulafied Vendor List) and only have 2 SKUs on that list. Simular for the eMMC, I don't mind it being there as the default, but I would want a m.2 slot or similar that I can use if it dies or I have requirements for something larger or more proformant for something like a long packet capture or snort or whatever. I would also want to make sure that if the eMMC dies the mobo is not reliant on it to boot/use. Just some ways to reduce e-waste and make service easier. And again, you can have a QVL for the SSDs/nand/whatever and anything else is the users/communities problem to make work.
This is an amazing project to follow and very educational to see the process and decisions you make along the way, and I appreciate how well and clearly you explain things so that anyone can understand. Thank you for sharing your journey
You're welcome! Happy to have you following along!
Holy shit what a project dude!
Yes! Would be very fun to see a tour of the factorys.
Gotcha, will arrange one!
Whilst I understand the economic drive to use ARM (and ARM is undoubtedly the future), this leaves difficulties with the choice of OS to run on it. You previously mentioned VyOS and OPNsense but both those are x86 only (at least in official builds).
There are third party builds of OPNsense for ARM and experimental builds of VyOS for ARM, but neither are really ready for mainstream use on ARM from what I can see. Are you planning to fork/port either of them fully to ARM?
As an aside, I am a little sad to see soldered main storage and RAM. It is limiting, not least in the case of repairability. And "8GB should be enough" has a good chance of being the next "640K should be enough".
Awesome to see the process behind this. However it seems too expensive due to the case costs, would also be nice to have 2.5gbe instead of 1gbe. Maybe the aluminum case should be an optional extra, with a standard plastic case?
Your journey inspired me to get the documentation from my first year in university and learn embedded systems again. I plan for january to get my first test PCB printed :)
Glad to hear that! And good luck - report back with results!
Whatcha makin'? Or is it hush-hush trade-secret squirrel type stuff?
Oh gosh no, just something I think is easy and doable. I want to create a digital pictureframe and a custom, (hopefully) small air quality probe, maybe with an option to integrate it to homeassistant. @@3nertia
WOW i thought this channel was a lot bigger. keep it up your doing great
Thank you!
As someone who has several ideas I'd love to see to market - including my own router design that I've since scrapped lol - I would just like to thank you. This series has already been insanely valuable in helping me to organize my thoughts - subscribed!
Thank you! And welcome on board!
You should make the case optional for people willing to save. Self 3d printing a plastic case and just adding a heatsink+fan (as you don't have the aluminum case anymore) would make it considerably cheaper.
Yup. I think that’s pretty much decided.
This was a refreshing video to watch. It's easy really informative!
Thank you! Glad you enjoyed it - more to come! :)
Thanks for the video even if I will never do anything like this in my life lol. All the best.
I find a small display very useful, especially if things don't work as expected. Small 0.91" OLED displays 128*32 Pixels cost about 2$ and would just do the job.
I had them in my first mikrotik routers. But only in the first ones. They are useless. IMHO of course.
I recently found this channel and watched few videos, first one was ISP router. Loved it, keep it up. new sub. This is an amazing project, learned a lot more than I thought. Good luck
This is a great project. I'll be raising my budget when it becomes real. Why not have 2 SPF+ ports instead of 1 and a RJ45 port? It would save $10 per unit and board space. If the customer was bringing a RJ45 cable, they can get an adapter. I think using a USB-C for all connectors is fantastic. I use a 200W USB-C hub on my rack of utility devices
Thanks for the feedback! Fun fact, I'm releasing a new video next week addressing these issues specifically! We removed the 10Gb RJ-45 in favor of another SFP+ and also moved from 4x1Gb to 3x2.5Gb RJ-45 ports because processor supports that out of the box.
Are you not considering a direct supply purchase rather than through distributors? I do realise that you may be forced to do this due to the uncertainty around the potential sales figures and the bulk purchase obligations you might face with buying direct: however, is it not worth a discussion with a sales/engineering rep in parallel with this? You might be surprised. Via a distributor, you'll always be paying for their profits & overhead.
Good analysis so far, with good, well-explained simplifications of the choices made. Excellent! This deserves to suceed.
I forgot to mention: might be worth pricing at 1 off, 10 off, 100, 1000, etc. production runs, and costing the production total/unit at each quantity. You can then graph it/build a pivot table to look at the best minimum unit sales figure to aim for. I know you're doing part of this with your market survey, but a graph might be interesting!
Thanks for the input, I'm well aware of the overhead but I found a distributor that's really helpful. Like if I give them a call and ask for their opinion, they'll pick up and talk to me, and help me find a solution. Despite the fact I haven't actually purchased anything yet. If you ask me, they're well worth that overhead. Plus the offers I got are super competitive.
Stay tuned, a graph is coming in Tuesday. Not exactly the same one, but you might find it interesting nonetheless!
Came across this video as I was wanting to upgrade my network and intrigued on what you’re wanting to do and trying to figure out what I would like to do. In the rabbit hole of networking videos and various searches I came across the minisforum ms-01. It has two 10g sfp+ and two 2.5g nics. All of them appear to be intel based. This seems to fit a little bit of what you’re wanting to build and at a minimum could be used to compare your design to.
Thanks for the suggestion! I've stumbled upon minisforum before and they sure do look interesting.
My experience tells me that you will have problem with bandwidth performance. This kind of CPUs really need their offload capabilities in order to achieve the bandwidth. You wanted 1Gibt/s IDS, that means that packet analysis probably have to be done in the main CPU, in that case the bandwidth the CPU can handle will be the upper limit. There might be solution but I don't think that it will work out of the box without special drivers.
As we're very early in the development stage, I have the luxury to change things. I'm waiting for the development board, and once I get it, I'll test it thoroughly, to make sure it can satisfy the requirements. Obviously, I will make a video about it (and not just one, most likely)
yes on the tour subscribed very interesting project
I just filled survey, one thing to add since it wasn't possible to add "custom note" there:
Idea is cool, but I don't like formfactor. I would rather prefer 1U format or half-width U. And solid aluminium block is not requirement... I would happily go with simple metal sheets for enclousure if that could bring price down.
Thanks for the feedback - no worries, there's been a lot of changes, and once I get around to writing them down, I'm going to present them in a video. Will fit in 1U. :)
Please allow socketed ram with either 2 sockets/channel or support for 16gb modules optional. You saving support and reducing variance does not cater to your audience. Not to mention that with NIDS and nfv(containers) ram becomes useful. Same is true for endurance of your disk if you want to traffic log.
And this is strictly hardwire. I would still need to add a WAP at some point. This is fascinating but also quite the reality check for how much good hardware costs.
Yep, will will bring it down in the future, but I want to guide my subscribers through the whole project.
hey quick note since it's still in the bom (31 Dec 2023 02:10)
I've noticed that at 15:46 and 15:55 the quantities are given in 2000 units instead of 1000 units and the price hasn't been cut in half in the BOM accordingly (or canceled the function)
(Edit: it happend in many more placed throughout the video ^^' once i started to notice it i was seeing it all over the place. if it's corrected or not is not on my behalf and probably won't change anything, just my OCD is being triggered ^^')
otherwise great video, even more so for the unexpected base level excel course
One additional idea: I have Unifi network. That limits what kind of gear would i buy.. what if.. i could put unifi controller on it?
You need one? I would use a PC for that. I build DIY Firewalls with PC (Server only) parts for my customers all the time because i don't need all those additional fatures of the full feldged firewalls like Meraki etc.
Forgive my ignorance, but 599 seems REALLY expensive for a router. How is this better than slapping a bunch of older hardware together and sticking opnsense on it?
Size and consumption, primarily. Its not the same to have a computer consuming 200W that a small router consuming 15W.
You could buy a fiber pcie card for about 70€ and a 2.5gb for 30€, and for a lot of people that would be enough.
In my particular case, i could not afford 750€ ('cause taxes are paid, you know) for a router and i am not in the spectrum of "people willing to buy this kind of router", but its interesting :D
What makes it much more expensive than your regular household router is 10Gb capability. Well get to the reasons why in one of the following videos, but having just two 10Gb ports gives you like 5x the capability of your "normal" of the shelf 4x1Gb router.
10 GB? The network cards alone that are useful for that bandwidth cost 200-300.
For me personally arm, soldered ram and emmc is complete deal breaker. For me it would have absolutely no other use than just being nice looking router. For like 300-400$ i could buy some qnap or something like this and if I wanted my router to be something more for 600$ i can buy some used hw with x86 and much much much better cpu, ram etc missing just the looks and small power consumption
I didn't really understand but I hope you'll be using the M.2 form factor for storage!
Not for the boot drive, but there will be an M.2 slot on it in case someone needs it.
10:26 - you can just write F instead of F50. It will sum all cells in column F.
Thanks!
While the case is nice, I wouldn’t want to have to spend ~30% of the device’s price for it since this will be sitting in a closet. Maybe consider also selling a bare kit for someone that can 3D print their own case. Or see what a bent sheet metal case would run. Ubiquiti gear is a good example of simple but much more affordable aesthetic.
BTW, I’d tone down the how to on gsheets - if someone doesn’t know how to setup a basic summing formula, they aren’t building a BOM for anything.
Just one comment about pricing, when I see a price like 599, I always think that the manufacturer wants to trick the buyers that it is a 500 something product which I find it a bit dishonest. Or may be the manufacturer thinks that the customers are stupid not to realize it is 1€ less than 600. Just ask for 600 please. I am sure your target customers are wise enough to evaluate the price correctly.
Psychologically, there are people out there who do evaluate $599 to $500. How? I have absolutely no clue
But there are also considerations for shipping, VAT, and other costs which may be more expensive if the product has broken a barrier (for example, $600)
Just something to be aware of when you see products that aren't a flat value
@@Spiker985Studios I could drive down there and get mine directly. Maybe cheaper than sending via postal service.
It's funny you mention an overpowered device.Device. I have 16 gig of RAM in my OPNsense firewall, thinking I would need the extra RAM. In the end I am just using over 3 gig of RAM with IDS/IPS, VPN, and Zen Armor running.
There are some smaller fabs thst make memory. But that’s less common with ddr4 and not happening yet for ddr5. I pretty sure oloy used those none big 3 memory chips on there low end ddr4 Modules. I believe they are called nanya
Never heard of them. But even if I did, I prefer to stick to the known ones.
Really interesting project! Couple of questions though. First, why two different port types for the 10G? Wouldn't two SFP+ ports make more sense? Yes, I know, many use RJ45 in the home, but given the option, adapters are not that pricey! Second, why 4 1G ports? Here in the US, 2.5G is becoming more common as a baseline... Does NXP have a chip that would handle at least two of these ports, in addition to the aforementioned 10G? Just wondering... you have stimulated the tinkerer side of my brain!
it would be awesome if you could power it through PoE
First of all, Thank you for doing this. This is a project that needs to be done. Secondly Why not a Watch Timer? Or is that included in another BOM Part? Third, For the love of all that is holy, please consider a sheet steel case, and bring your BOM down!
Thanks for the kind words. This is not the final BOM, we will cut the costs down, but that'll come in a separate video as it's an important topic on its own! Stay tuned! :)
Great Video
Thank you!
I am highly interested in this project, as I have been shopping to buy or build a 10 gig router for home use for several years. I've looked at more server level boards than I care to count, and the best value I've been able to even consider is Netgate's 6100 router, which weighs in at $800 USD. But it is still missing a feature I would highly like to see on a router, which is ECC RAM. I see from your BOM selections that you did not choose ECC RAM either, and I wanted to ask if you have considered it, and if so, what the cost difference is with ECC RAM. Thanks!
I myself would be at least interested in hearing/reading the advantages versus disadvantages of having versus not having ecc as well
I'm aware that "ecc" is error correction but I don't really know how much value it actually has overall and if it would be worth the potential markup to the overall cost of the device
@@3nertia I read a study done years ago about random bit flips in computer memory and found that with memory densities as high as they are, they were seeing an average of 1 bit flip a month for each GB of memory. So for 4 GB you’d average one a week, and 32 GB would average about one a day.
If the flip happens in unoccupied memory, nothing bad happens and life goes on. If it happens in a chunk of data, it’s possibly silent data corruption if there isn’t any validation. If it occurs in a critical piece of OS memory, it could result in a system crash. In desktop machines, which usually don’t have a long uptime, this problem is usually academic, but in longer running systems, those flips can potentially stack up and cause larger problems, so it’s good to have ECC memory to detect and correct single bit flips, or be able to alert the OS to the problem in case of double bit flips, which it can detect but not correct. Then the OS can decide how to handle the situation.
Hope this was useful.
Hello, I know the 2013 Excel could pull prices from web pages, so I am sure Google Docs can do the same. that will keep your bill of Materials current. If I was building a dream router I would want it as future-proof as possible which means 2.5 Gbs ethernet ports for the LAN, as I am starting to see them come up more and more. For the WAN ports, I suspect you would want a 2.5 Gbs port that is compatible with a cable modem that has 2.5 Gbs ethernet. I don't know enough about the 10 Gbs ports, are they compatible with the 2.5 Gbs ethernet, then you could save a port. CPU needs to be powerful enough to speed through whatever software you are going to be installing especially at that 10Gbs pace. RAM you obviously need enough for all the necessary software to fit in it, that is your minimum size. But you know your power users are going to want expandability, so have a way to double the size. You did not talk about WIFI, if you decide to include it I suggest having it upgradable with the 2.M socket.
Thanks for the feedback! Building the BOM here was more of an exercise for the viewers and to drive a couple of other points across in the future videos, but you sure do have some valid points regarding 2.5Gb ports (10Gb already covers those needs) and wifi (likely an M.2 or mPCIe card).
With the current BOM price, I think this will sadly be DOA. Qotom recently released a product that offers more connectivity, has more performance, and is sold for a retail price only a few dollars more than what your BOM currently is in v2. Its also x86 so you can easily install any of the main router distros on it. I really want this project to succeed, but at this point I just don't see how it can when a more well-known competitor offers something better for cheaper.
Look up the model number Q20331G9-S10 on google. Let me know if you want me to remove this comment with the model info and I will.
All good. I'm not interested in competing with Mini PCs. If someone prefers that, that's okay, competition is always healthy. Currently working on a video in which we brought BOM down in half.
Why do you want to only use 1 RAM chip? you can use 2 in a T junction setup and that may drive the price down a lot overall but it is harder to design.
Dedepending on that CPU you may also be able to use 4 chips that that will drive cost down a lot.
60$ for the PCB+ assembly seems cheap for a semi small amount but if you produce a lot it may well be even lower.
We must have at least one 2.5GB port. Or have a model that has 2.5GB ports.
Yes.
go big! bigger! 2.5g ports required
Yup. Done! 🙌
Maybe look at the Marvell Alaska M 2540 or Alaska M 413C? The Alaska M 3540 also looks neat.
All those are overkill, because they are quad port 5/10G.
@@tomazzaman The workaround might be to just have 2 10SFP+ ports that can take one of the 1/2.5/5/10g SFPs and then 2 10/25G ports for LAN. We are seeing ISPs deploy gear with 2.5G ports and there are some places that are pushing out 3Gbps service, then you have super cool places like chatanooga that have 10Gbps for like $299/mo for residential.
I also think I would want ECC RAM as well >_>
Can you avoid machining by using a standard profile and restrict customisation to a plate
Can you add a NE555 Timer?
I think they are neat
Respect, what is you target and how do you try to reach it! :-) I think , the appereance is not so important, as you think. Sorry, this is only a router, which is normally not the part of a living room or a clean desk. It should be looks fine, but the contents/ingredients are much more important! This is the factor, where you could win a lot.
Thanks, appreciate the input!
Do you really need the CR2032 for the RTC? Can't you just sync the time via NTP on bootup? What do SBCs like the Raspberry Pi do for it? I don't think they have a battery.
I guess we could skip the CR2032, but it's a relatively minor expense compared to other components, so I didn't see a reason to.
How does this make sense when the Qotom Q20311G9 exists now with 4x SFP+ for 10G and 5 2.5G RJ45s? Its around $450 even with a 10G SFP+ transceiver, RAM and SSD.
I mean, there’s plenty of devices by other manufacturers that are not the one you’re referring to. 🤓
Another idea after I watched the dev board. The ram is it going to be soldered or a stick? If soldered, why? Just add a stick of 8GB or give the people the choice to go with or without memory. I just looked up and a stick of 16GB DDR4 3200MTs costs less then 10€ and with the change to ddr5 there's going to be a lot of ddr4 coming free :). But then again you need to test different dimm's to see if they work or not. Mmmmh a single slot for ddr4 and a replaceable stick of ddr4 or soldered mmmhhhhhh
Or make it ddr4 sodimm, connectors costs less then 2€ memory itself is a little bit more expensive /GB (8GB 320MTs for les then 10€) but hey less vertical space needed
WHAAAAT 30$ for 8GB and not replacable, don't like that. Please change that.
RAM is the one area I'm not 100% certain yet of which way to go. Luckily I do have a couple of meetings with people way smarter than me on this topic, and I'll see what the expert opinion is. SODIMM or SOLDER? 😂
@@tomazzaman I just read on the ref. design that it only supports 8GB of ddr4 ECC 2100MTs, pfffff
@@Damicske so? show me a router (of a similar class) that has or needs more.
It would be great if you would sell just a board too. I don't think that I would pay +80€ for something that will be in my closet
I would like to see some different interchangeable board options as well to make the chassis slightly modular and upgradable
Maybe some daughterboard options to cover different types of interfaces? For example relocating the RJ45/SFP ports to swappable modules would give some amazing flexibility, but maybe not economical in the first gen product. It would certainly be great to see on future versions though
@@rfitzgerald2004 I would love if he could arrange that I get a sexy goth girl with a router.
I was looking at the Google Sheets. Does "Chicken Food" just mean waste/scraps?
Small resistors and capacitors mostly. They cost cents each. So in this stage I just added a ballpark.
how can you get a PCB for fine pitch bga parts for 30 dollars the board :O ?
What are you going to use for an AP? Didn't see any wireless components listed
Not in this version.
@@wolfgangpreier9160 then it's a managed switch
@@1HotLegendLS Router. Not switch.
Thoughts on the Droplet IOT encryption algorithm?
None, as I have no experience with it. Looks interesting though!
Am I the only one noticing he entered price for 100 units = $73.6 for 1000 units?
No, you're not. I got the price for 1000 unit as well, and it's the same.
I always try to select critical chips that has a dev kit
Same, waiting for a couple to arrive!
I'm sorry but I have to make a remark: I DESPISE the idea of soldered ram.
I have a certain number of machines running basically 24/7, depending of the years, 10 to 20, and in the last decade, it's 5 sticks of ram that failed me.
With soldered ram, it wouldn't have been 30 minutes of downtime. It would have been hours to repurpose something else jank-ily to days if not weeks for a replacement.
If we really are building the great home router for homelabers/tinkerers/ tech enthusiasts, the ram shouldn't be soldered. It isn't in tiny clients many use as nodes. And the ones who run enterprise servers in their closet know how to deal with reduced ram compatibilities.
I'd rather have sticks limited to a peculiar frequency, CAS latency, or even timings, or even blow a few dozen bucks on some failed compatibility-checks than have to whip out the rework station or buy a whole new router because of a failed ram chip.
Thanks for the feedback! We're trying to strike a good balance between expandability, reliability/quality and price. The videos I'm making on it aren't anything final. And I'm making them precisely for this reason, so that you and the rest of the community can voice their concerns. So, thank you!
I had more problems with exchangable Cisco RAM in the last 20 years than with others. And with obsolete Hardware after 5-7 years the RAM is not the limiting factor anymore.
What i want to say is: the hardware of every manufacturer, no matter if DIY, Sonicwall, Meraki must be exchanged after not more than 5-7 years anyway because it can not handle the next Software upgrade anymore.
@wolfgangpreier9160 I can see where you're coming from, but don't forget the router project isn't meant to cater solely to enterprise/professional use with a renewal cycle of half a dozen years.
How many homelabers are running proxmox on 8-10 years old Xeons that are considered to be near e-waste at this point.
From 2015 to 2019, my homelab's router was a thinkpad with a 3G cellular wan fail-over.
I'm currently running an OPNsense firewall (and router for some sub networks) on an AMD Geode thinclient from ~2015 with a LP PCie NIC. Because I wanted my firewall to not be concerned by Intel's Management Engine nor Ryzen's AMD PSP. It's not fast by any means, but sufficient for the throughput it's used for... and that's what I had on hands.
And being able to replace the mSata drive when it failed during covid, and I had a couple of them collecting dust on a shelf was also very useful.
////
Back to the router at hand though... at 10gbps (if one of the 10G ports can be used as WAN, I have a 2.5G uplink atm already), I could see myself using@@tomazzaman's router for the next 15 years, eventually DDwrt'ing it when it get end-of-lifed?
It just has to not have a chip die, then I'm ready to invest once into a piece of brand new hardware that would cost as much up front as 15years of jank solutions built out of e-waste that do the job all the same.
I would even wanna pay a 10% upcharge for it to come barebone, without any storage or ram, and just a couple of SODDIMs and m.2/msata SSDs validated for it.
But all in all, it's my personal feedback and feeling.
Whatever path you go for, given your current direction, go for it, it will find an audience considering the clear demand for better network solutions.
For instance, on top of just part-swapability, instead of ARM-based, I'm yearning for a RISC-V (or other opensource architecture) router, where there is no arm trustzone or blackbox 'security' by obscurity paradigm.
Excellent Post !! Soldered RAM is one of the biggest forms of e-waste EVER. Sorry for the caps. Got myself a CWWK i3-1315U where I can put 64GB DDR4 RAM on there (should of used credit card & got higher end one instead of using debit card). Also 2 NVMe's and 2 regular SSD's ( 2.5", but can't fit a 2nd one in the case :) However I've been looking at older... what are they called Micro, Mini PC's. The HP EliteDesk 800 G3's, G4 or G5's that I can run Proxmox and pfSense CE, OPNsense and numerous containers on. Well might have to run Proxmox and numerous Docker containers at some point in the future as my CWWK may not be able to handle what I have in mind after all. Then I'll do all those things and say, damn I don't need to run all those things and be fine with it Afterall :) But Please No Soldered RAM. Think how many millions of things get tossed world wide each and every year just because of it. Total Waste. @@Varengard
No sfp Slot?
Nope, just SFP+. You can put a SFP module in, and it'll work just fine.
@@tomazzaman okay i forget the plus 👌🏼
I was very interested in this project until you decided to solder the RAM on. I think you will lose a significant portion of your potential market for a boutique router, by prioritizing the reduction of support costs over the flexibility of RAM replacement. Thereby creating yet another nonrepairable, nonupgradeable device.
shift your tone away from the 'influencer' shit, that's not your target audience.
Use the best parts, or what's the point. I'd rather buy 2 'solid' cheap chinese routers over a 'medium-high end' compromise.
Make an enterprise grade switch for prosumer prices, no cutting corners, no RBG, no frilly cases etc.
Be different enough to be interesting, don't just make a pretty mokerlink.
Pay that dollar for the 10G rj45 port, or get rid of it even...
Powerful, simple, elegant: in that order.
That aluminum case is so dumb unless you've plans to consume enough power to need it as a heatsink, and you don't even have the power supply internal..
Steel is cheaper, stronger, heavier, and conducts heat just fine.
Magnesium alloy casting might even be the way to go if you can justify tooling costs.
I have tried 2 cheap chinese routers and put them away after 2 hours of not working. I don't have the time and patience to play around with this things. They either work at once or never. I had to put in the effort learning mikrotik and since then i'm somewhat familiar. Much too convoluted and complicated but they work.
10GB SFP+ needs good cooling. Maybe thats the reason for Aluminium?
You could ditch the RTC and just get the time from a public NTP server.
No you can not. You have to have the time before you get Internet connectivity. You may correct the RTC AFTER you got connection.
@@wolfgangpreier9160 yes if you use HTTPS but you dont for NTP
I would NEVER EVER pay even 300€ for a router. Not for my home or private use. Just because, who the hell needs that at his home? For a company thats another question. But there are very good brands that make good routers already. So I dont get the idea behind this at all.
I am just here to tell you, who would NOT buy such a thing! Maybe helps with the calculations.
I have a pricing survey form out which currently has almost 800 submissions and currently the average that people are willing to pay for is 380€. The average. Which means half of the 800 are willing to pay more. And obviously, half are not. Thanks for the feedback!
@@tomazzaman What would you concider a success? 10k units in 5 years?
why not just buy a banana pi?
The idea of making the RAM and storage embedded so you don't have to support different models of them down the line is extremely narrow minded
Thanks for the feedback!
Since you worry about compatibility for end users and the possibility that hundreds of different RAM sticks might be used by people who buy this, perhaps something you could do is to buy and test with just 6-8 different sticks of RAM that are easy to buy around the world. Maybe 3 sticks at 4GB each from 3 different main manufacturers, and 3 more 8GB sticks from those same manufacturers. That way you have a QVL of known compatible memory people can use as a reference but also give people options. This lets people have options by not being soldered onto the board but also lets users know of specifically tested sticks they should be trying to use.
@@LordSaliss he is not building a new ram controller, he is using an off the shelf one, their support, documentation and stability is robust.
It seems the crap apple does rubs of on their customers, locking in and deciding on behalf of the user without giving the user any options