ITX laptops & means to build one

[Essence of Flowers]

Intent of this thread is to provide a place for ideas and concepts for ITX laptops. Ill start by documenting every viable idea I've run by to make such projects feasible.

Spoiler: Motherboards

The choice of motherboard to use for projects like these seem to be between the Thin-ITX standard or Mini-ITX standard. Both options seem to have their pros and cons but are equally viable options in my opinion. A basic rundown of the two below goes as follows:

Thin-ITX is beat suited for projects like these out of the box. They use SODIMM memory slots, making them much more compact. They often feature eDP or LVDS connectors which allows for LCD panels to be used for a monitor. They also feature DC jacks which allows powering these things much easier, 4 pin power connectors are also available for any potential battery systems. The main downside of thin-ITX however is that they're primarily meant for industrial use so they rarely see commerical markets. They're also dominantly intel boards, very few AMD options exist, with most not being for commercial sale. All in all one's main struggle with Thin-ITX is going to be getting one in the first place, but once obtained it'll prove a good solution to most problems these builds suffer.

Mini-ITX on the other end is practically a polar opposite. Getting ahold of one is very easy, however the board itself will most likely need modding to fit the needs of this project. The average mini-ITX board tends to have a very bulky I/O, so unless we want to build cinderblock laptops we'll need to remove most of the I/O either via desoldering methods or as Socket Science did, yanking them off. Losing the entire I/O is a bit problematic however plenty of solutions exist to restore functionality, such as using the USB header connectors meant for a chassis to install USB hubs. Very few Mini-ITX boards feature eDP/LVDS however fortunately there are PCIe cards that offer thunderbolt USB-C, which could be used as a means to get a connection to a display, certain video cards most likely also feature such options. For power, the typical 20/24 pin connector is going to be our option here. Pico-PSUs exist as a simple solution, and UPSs by Mini-Box also seem to offer potential solutions.

Spoiler: CPUs and GPUs

Choosing between intel and AMD CPUs is dependant mostly on your goals and choice of motherboard. AMD offers a good choice of APUs which actually perform not bad in benchmarks, thus if you're aiming to avoid installing a dGPU they're your best bet. Intel is pretty much your only option for Thin-ITX, and although Arrow Lake is supposed to feature some serious iGPUs, its a completely new socket and therefore will require new Thin-ITX boards be released, best of luck to anyone planning to get their hands on one. The most realisitc option for Thin-ITX is most likely using a dGPU and removing its stock heatsink. Once more pointing to socket science this is a viable option, just requires handiwork on the builder's part.

Spoiler: Batteries

Probably the hardest part to work out for builds like this is how exactly will powering the system work? Mini-Box has a few solid options for battery systems such as OpenUPS, or on the flipside devotion laptop proved that custom building a battery system is also possible. Of course, there is also the option to omit the battery, reducing the portability of a device meant to be portable by nature, but certainly a way to simplify the build.

Spoiler: I/O

Regardless of which motherboard chosen utilizing the USB headers to add USB connectors for a keyboard and trackpad is an option, depending on the model of keyboard wirless is even an option. Some options even offer headphone jacks but its difficult to say how good they actually are. Beyond that many commerical solutions exist for other things, power buttons that are simply as-is and not connected to a chassis exist. Some of these expansion bays offer SD card slots, and depending on how much cablework you want going on inside internally a CD drive is even possible. For displays thin LCDs that use eDP or LVDS is easily the best option for thin-ITX. But plenty of slim displays exist for mini-ITX too in the form of thunderbolt 4 USB-C connectors.

 

[Lux]

Intent of this thread is to provide a place for ideas and concepts for ITX laptops. Ill start by documenting every viable idea I've run by to make such projects feasible.

Spoiler: Motherboards

The choice of motherboard to use for projects like these seem to be between the Thin-ITX standard or Mini-ITX standard. Both options seem to have their pros and cons but are equally viable options in my opinion. A basic rundown of the two below goes as follows:

Thin-ITX is beat suited for projects like these out of the box. They use SODIMM memory slots, making them much more compact. They often feature eDP or LVDS connectors which allows for LCD panels to be used for a monitor. They also feature DC jacks which allows powering these things much easier, 4 pin power connectors are also available for any potential battery systems. The main downside of thin-ITX however is that they're primarily meant for industrial use so they rarely see commerical markets. They're also dominantly intel boards, very few AMD options exist, with most not being for commercial sale. All in all one's main struggle with Thin-ITX is going to be getting one in the first place, but once obtained it'll prove a good solution to most problems these builds suffer.

Mini-ITX on the other end is practically a polar opposite. Getting ahold of one is very easy, however the board itself will most likely need modding to fit the needs of this project. The average mini-ITX board tends to have a very bulky I/O, so unless we want to build cinderblock laptops we'll need to remove most of the I/O either via desoldering methods or as Socket Science did, yanking them off. Losing the entire I/O is a bit problematic however plenty of solutions exist to restore functionality, such as using the USB header connectors meant for a chassis to install USB hubs. Very few Mini-ITX boards feature eDP/LVDS however fortunately there are PCIe cards that offer thunderbolt USB-C, which could be used as a means to get a connection to a display, certain video cards most likely also feature such options. For power, the typical 20/24 pin connector is going to be our option here. Pico-PSUs exist as a simple solution, and UPSs by Mini-Box also seem to offer potential solutions.

Spoiler: CPUs and GPUs

Choosing between intel and AMD CPUs is dependant mostly on your goals and choice of motherboard. AMD offers a good choice of APUs which actually perform not bad in benchmarks, thus if you're aiming to avoid installing a dGPU they're your best bet. Intel is pretty much your only option for Thin-ITX, and although Arrow Lake is supposed to feature some serious iGPUs, its a completely new socket and therefore will require new Thin-ITX boards be released, best of luck to anyone planning to get their hands on one. The most realisitc option for Thin-ITX is most likely using a dGPU and removing its stock heatsink. Once more pointing to socket science this is a viable option, just requires handiwork on the builder's part.

Spoiler: Batteries

Probably the hardest part to work out for builds like this is how exactly will powering the system work? Mini-Box has a few solid options for battery systems such as OpenUPS, or on the flipside devotion laptop proved that custom building a battery system is also possible. Of course, there is also the option to omit the battery, reducing the portability of a device meant to be portable by nature, but certainly a way to simplify the build.

Spoiler: I/O

Regardless of which motherboard chosen utilizing the USB headers to add USB connectors for a keyboard and trackpad is an option, depending on the model of keyboard wirless is even an option. Some options even offer headphone jacks but its difficult to say how good they actually are. Beyond that many commerical solutions exist for other things, power buttons that are simply as-is and not connected to a chassis exist. Some of these expansion bays offer SD card slots, and depending on how much cablework you want going on inside internally a CD drive is even possible. For displays thin LCDs that use eDP or LVDS is easily the best option for thin-ITX. But plenty of slim displays exist for mini-ITX too in the form of thunderbolt 4 USB-C connectors.

This is a pretty good summary of what to consider for an ITX-Laptop design. Seems this post is a fork of Devotion: Thin ITX Laptop and I assume you are considering undertaking your own project?

If you are trying to build a platform to share/market, then I believe Devotion is headed in the right direction with thin-ITX. As you noted AMD has some really good iGPU options and I would trust the AsRock X600TM-ITX will come to the public market soon. Hopefully board manufactures will follow with Alder lake. In the mean time, I would prototype around an inexpensive thin-ITX board.

Alternatively, if this project is just for yourself I would consider up-cycling something OEM. Do a unique build around an HP elitedesk mini, for example. The included 4650ge is already optimized a 35w. Biggest challenge would be attaching a battery but it can be done.

Having a dedicated GPU nearly rules out a battery powder unit. Expensive option, but an MXM GPU with MXM to PCIe adapter might be low power enough to still allow a battery. Lastly, some low/mid teir GPU's can run on 4 PCIe lanes without bottlenecking. The 1050ti for example I've read was unaffected on PCIe 3.0 x4. Thus in it's possibly to take PCIe lanes from an M.2 with another adapter. Lol

 

[Essence of Flowers]

This is a pretty good summary of what to consider for an ITX-Laptop design. Seems this post is a fork of Devotion: Thin ITX Laptop and I assume you are considering undertaking your own project?

If you are trying to build a platform to share/market, then I believe Devotion is headed in the right direction with thin-ITX. As you noted AMD has some really good iGPU options and I would trust the AsRock X600TM-ITX will come to the public market soon. Hopefully board manufactures will follow with Alder lake. In the mean time, I would prototype around an inexpensive thin-ITX board.

Alternatively, if this project is just for yourself I would consider up-cycling something OEM. Do a unique build around an HP elitedesk mini, for example. The included 4650ge is already optimized a 35w. Biggest challenge would be attaching a battery but it can be done.

Having a dedicated GPU nearly rules out a battery powder unit. Expensive option, but an MXM GPU with MXM to PCIe adapter might be low power enough to still allow a battery. Lastly, some low/mid teir GPU's can run on 4 PCIe lanes without bottlenecking. The 1050ti for example I've read was unaffected on PCIe 3.0 x4. Thus in it's possibly to take PCIe lanes from an M.2 with another adapter. Lol

Yeah this is essentially a fork of devotion laptop, since as this build comes more to fruition its definitely going to be an undertaking of a thread.

On the topic of batteries since its been my focus lately. It is definitely possible to get a dGPU rig with a battery but its going to be VERY bulky, power inefficient, and heavy. The concept i drafted up for one would require using two OpenUPS together and feeding them to an HDPlex PSU. Not only is this very expensive of a solution but it actually broke the space budget my current goals are set to. It would result in 480W of power which is great but chances are it wont have a long runtime and itll probably need a very good cooling system from all the heat going around.

For my personal build im leaning very strongly towards simply using an APU, and leaving room for a dGPU bay should the market shift and a low power GPU that outperforms iGPUs comes out. HDPlex offers a 200W pico PSU that accepts a wide input voltage which makes it actually a very good option for a more lightweight dGPU setup, however frankly OpenUPS here is our main bottleneck, having 200W PSUs doesn't matter all that much if the batteries can only push out 140-ish watts at their peak. That said it does come with another complication which is OpenUPS works with 4 pin connectors while HDP uses 6 pin, in theory it should be fine to just hook up as is but thats a bold assumption on my part.

Edit: interesting find on MXM. I had assumed it was an effectively deprecated standard but i see a few companies out there are still making them. I might consider looking into using one albeit i still need to work out a power budget.

 

[Lux]

That said it does come with another complication which is OpenUPS works with 4 pin connectors while HDP uses 6 pin, in theory it should be fine to just hook up as is but thats a bold assumption on my part.

You can rewire the 6pins from the HDplex to the 4pins on the OpenUPS without concern. Since your limitation is the 140w output on the OpenUPS. Wire gauge is the important part and HDP has already rated those for 200w. Also the connectors are called Mini-Fit Jr. and Molex official rates them max at 13amps per contact.

140w with what you are trying to do seems pretty reasonable. The key to making your battery last is going to be controlling your system power overall. Create profiles to under-clock components and enable features like link state power management. Can really advanced with some Linux utilties.

Edit: Oh and Nvidia Quadro tend to be well optimized for lower power usage without much performance loss. Something like this RTX A2000.

 

[Essence of Flowers]

You can rewire the 6pins from the HDplex to the 4pins on the OpenUPS without concern. Since your limitation is the 140w output on the OpenUPS. Wire gauge is the important part and HDP has already rated those for 200w. Also the connectors are called Mini-Fit Jr. and Molex official rates them max at 13amps per contact.

140w with what you are trying to do seems pretty reasonable. The key to making your battery last is going to be controlling your system power overall. Create profiles to under-clock components and enable features like link state power management. Can really advanced with some Linux utilties.

Edit: Oh and Nvidia Quadro tend to be well optimized for lower power usage without much performance loss. Something like this RTX A2000.

I've had half the mind almost to use two OpenUPS2 to power the system, putting us at the 200w needed to reach what HDP can achieve. Certain workstation GPUs like the radeon w7500 has been benchmarked to only take around 150w accounting for the full system load + a heavy workload. However this will once more put me back down the research rabbithole of will these low power GPUs actually hold up against their APU counterparts.

Even should it outperform the APU, 470-ish USD is a serious price tag for what may not be that large of a performance boost, but its yet to be foreseen.

Likewise using two OpenUPS2s is going to more or less double the cost of the battery system... Perhaps that'll be worth the investment if it means longer battery runtime even if dGPUs fail to see the light of day for this build?

 

[Lux]

I've had half the mind almost to use two OpenUPS2 to power the system, putting us at the 200w needed to reach what HDP can achieve. Certain workstation GPUs like the radeon w7500 has been benchmarked to only take around 150w accounting for the full system load + a heavy workload. However this will once more put me back down the research rabbithole of will these low power GPUs actually hold up against their APU counterparts.

Even should it outperform the APU, 470-ish USD is a serious price tag for what may not be that large of a performance boost, but its yet to be foreseen.

Likewise using two OpenUPS2s is going to more or less double the cost of the battery system... Perhaps that'll be worth the investment if it means longer battery runtime even if dGPUs fail to see the light of day for this build?

Are you familiar with any CAD software or desire to learn one? Doing a mock-up in 3d might be the bast way to move forward. Get an idea of how things need to be laid out and what is an acceptable total unit size.

 

[Essence of Flowers]

Are you familiar with any CAD software or desire to learn one? Doing a mock-up in 3d might be the bast way to move forward. Get an idea of how things need to be laid out and what is an acceptable total unit size.

Oh I'd love to learn CAD software but since this is gonna be an ambitious project im going to have my partner handle it, they're far more experienced with making CADs and wanted to help contribute to this project by 3D printing a prototype chassis.

In terms of rough drawings ive done im actually just about ready to share some really sloppy sketch work. I mostly just wanna figure out how all the guts would sit in this system and then from there ill have a better idea of what features are realistically going to fit.

 

[Essence of Flowers]

so I wanted to write a status update here regarding battery systems. As previously mentioned its probably one of the most complicated aspects of builds like these. I was previously investigating MiniBox's solutions of OpenUPS models vs building an entire system from scratch. I can happily say that I actually found a third alternative available but also go into why I've yet to commit to any option yet.

For starters, OpenUPS is insanely expensive and potentially not a safe pick. only the base OpenUPS model itself can provide the wattage needed to power this system and even then it can only achieve 144W at peak, 151W if we seriously push it by letting everything charge to 4.2v and using all 6 slots. However the 8700G APU has reported wattage peaks of 150W so even when we push a single OpenUPS to its absolute limit we're still coming up short. I had conceptualized the idea of rigging multiple OpenUPS units together which could achieve much better results but after consulting with a friend of mine who does batteries professionally, he pointed out quite a few hurdles that would make this actually not the safest setup.

For instance, there's nothing really stopping 2 openUPS units from feeding the other power into their output which would definitely damage parts. Even if that isnt a risk, we still run by the problem that OpenUPS as far as i can tell, is not designed to be used with multiple units together. The OS environment will most likely not be able to differentiate the two or more units and since said units will not be aware of the other's existence, it may result in one doing most of the workload while the other sits around doing nothing, and then one OpenUPS ceases power output to avoid overdischarge thus overwhelming the workload of the other... This is not only to mention that we're looking at this costing well over 200 USD and the space management for hosting multiple openUPS models is just a nightmare.

All this said I got curious and looked more into MNT, for reference MNT has a set of products known as Reform which are modular laptops that are also fully open source in documentation. MNT has already released a set of BMS boards that have a cell holder built into them, thus only active balancing and charging would need to be resolved. These cells are LiFePO4 which is a nice side benefit since they're far less likely to go boom if something breaks. But what excites me more for what MNT is cooking is this recent article they posted regarding their next unit.

In Reform Next, the battery packs have 4 user replacable 18650 cells in series each. ... They have their own monitoring and balancing circuits and abstract all measurements like cell voltages, current, and alert states digitally over I2C. ... The coordination of the packs, the charger, and the USB-C power delivery living on Port Board one are implemented centrally in free software running on the RP2040 microcontroller on the motherboard. The default chemistry of the cells is LiFePO4. The chemistry can be switched to Li-ion using DIP switches on the motherboard — this optional user choice trades longer runtime for less safety and environmental friendliness.

The newest laptop they have planned will feature a modular motherboard, with the "Port Board One" board mentioned in this article having main control over the battery system. My hopes are, if its anything like their previous Reform unit, that they design this around from the perspective of someone hijacking this part of the motherboard for their own hardware projects such as this one. If not then at the very least the pre-existing BMS boards they have up for sale right now will definitely simplify the process of homebrewing my own battery system, when I consulted with my friend regarding using it to make my own he even mentioned a MOSFET may not even be neccessary depending on how I set it up, which is great because I am not smart enough to make anything remotely close to a MOSFET.

 

[Lux]

Your friend is quite right, chaining multiple charge/discharge units isn't safe without additional protections. One reason Lithium batteries go spicy, is the cells not being evenly charged. That said, 18650's are safer than flat cell lithium. Simply due to the hard casing and fail-safe vents, if they do go critical.

That is a super cool platform MNT is building. I think you could adapt part of what they've built to your needs. Something I noticed though, is the lack of a 12v output rail. They use the LM62460 and MAX1837EUT33 converters to supply 5v, 3.3v & 3.3vSB, which is good. You do require those voltages but 12v as well.

Thus you'll need to add another DC/DC boost converter. Keep in mind all of your high draw components(CPU) will be on the 12v rail. My research says, it's typically not recommended to draw more that 3.5amps per 18650. If you've got eight in a system that translates to ~120w peak. So to be safe, you'd want to limit the CPU to ~80 watts.

Keep in mind, most laptops these days have CPU's between 10-45watts. Partly for cooling but also to reduce battery wear. High performance laptops will sometimes have charging bricks up to 180w but they won't pull that much unless it's plugged in and under high loads. On battery the hardware knows to set limits.

Here the DC/DC boost converter I'd probably start with. You can dial it at 12v up to 100w. https://www.jacobsparts.com/items/DCMOD-D

Hope that helps!

 

[devotion-laptop]

so I wanted to write a status update here regarding battery systems. As previously mentioned its probably one of the most complicated aspects of builds like these. I was previously investigating MiniBox's solutions of OpenUPS models vs building an entire system from scratch. I can happily say that I actually found a third alternative available but also go into why I've yet to commit to any option yet.

For starters, OpenUPS is insanely expensive and potentially not a safe pick. only the base OpenUPS model itself can provide the wattage needed to power this system and even then it can only achieve 144W at peak, 151W if we seriously push it by letting everything charge to 4.2v and using all 6 slots. However the 8700G APU has reported wattage peaks of 150W so even when we push a single OpenUPS to its absolute limit we're still coming up short. I had conceptualized the idea of rigging multiple OpenUPS units together which could achieve much better results but after consulting with a friend of mine who does batteries professionally, he pointed out quite a few hurdles that would make this actually not the safest setup.

For instance, there's nothing really stopping 2 openUPS units from feeding the other power into their output which would definitely damage parts. Even if that isnt a risk, we still run by the problem that OpenUPS as far as i can tell, is not designed to be used with multiple units together. The OS environment will most likely not be able to differentiate the two or more units and since said units will not be aware of the other's existence, it may result in one doing most of the workload while the other sits around doing nothing, and then one OpenUPS ceases power output to avoid overdischarge thus overwhelming the workload of the other... This is not only to mention that we're looking at this costing well over 200 USD and the space management for hosting multiple openUPS models is just a nightmare.

All this said I got curious and looked more into MNT, for reference MNT has a set of products known as Reform which are modular laptops that are also fully open source in documentation. MNT has already released a set of BMS boards that have a cell holder built into them, thus only active balancing and charging would need to be resolved. These cells are LiFePO4 which is a nice side benefit since they're far less likely to go boom if something breaks. But what excites me more for what MNT is cooking is this recent article they posted regarding their next unit.

The newest laptop they have planned will feature a modular motherboard, with the "Port Board One" board mentioned in this article having main control over the battery system. My hopes are, if its anything like their previous Reform unit, that they design this around from the perspective of someone hijacking this part of the motherboard for their own hardware projects such as this one. If not then at the very least the pre-existing BMS boards they have up for sale right now will definitely simplify the process of homebrewing my own battery system, when I consulted with my friend regarding using it to make my own he even mentioned a MOSFET may not even be neccessary depending on how I set it up, which is great because I am not smart enough to make anything remotely close to a MOSFET.

Just coming back on your requirements, assuming it's the 8700G - I feel sure you should be able to run a build like that with considerably less than 150W, because that sounds crazy high! The highest such figure I can see there is this TechSpot review where they report 148W ingame, and they also report 41W idle, which is also high; my 5600G idled ~20W, and peaked ~80W with PBO on (~60W with it off).

I'm a bit skeptical of those figures, and they seem so much higher than other reviews...maybe they are instantaneous peak values (which should be OK if they're only for a few milliseconds)...but regardless, assuming they're realistic...if you can accept the idea of capping the frequency (e.g. by disabling boost/PBO, or similar for the IGPU only), then 150W would be impossible, and I don't think you would lose much performance by doing this. It's what I'm doing for the Devotion prototypes anyway, testing of which is where I got the ballpark figures of "enable boost to increase power draw by ~35% and performance by ~15% (or cap somewhere inbetween)". That's kind of the whole idea of the APU, efficiency versus DGPUs; low power and modest performance, more for basic games than the latest triple-A.

One more thing - all that wattage from the CPU is going to be dissipated by one heatsink you'll put on it, in a small form factor (in other words not a giant tower like an NH-D15). 150W or anything close to that is a very large demand on a small heatsink and the ventilation generally; best case scenario is that you'll have a burning hot surface temperature with the fan at max. All that is to say that I'd personally consider a realistic max power budget closer to 100W, which is within one OpenUPS or OpenUPS2.

The point about price with the OpenUPS is fair, because we know that production laptop battery packs don't cost that much at all, but that's really economies of scale, and it might be difficult for you to spend much less if you have to experiment with a one-off build. I deeply admire MNT, but as Lux mentioned, I think you would need to make modifications to integrate one of their cell boards (which I also looked at). They're primarily meant to be used in an MNT laptop, which is designed to run at low power draw - so low in fact that they make a point of being fanless.

Though nobody asked, I will also offer my experiences with an OpenUPS2, because I have integrated one into my current Devotion prototype, and it may be of value to you. (I also bought an OpenUPS just to stare at, and haven't done anything with it yet.)

Spoiler: do care, did ask, plus I love you

Firstly, regarding compatible 18650 cells. Examining the OpenUPS2 documentation, I see no reference to CDR (current rating) requirements for the cells, whether LFP or Li-ion/NMC. Same for the MNT Reform by the way, which also relies on LFP cells, though their system is much lower power. I don't know if that's because LFP cells are assumed to be safe when exceeding their current rating, but my understanding is that they are very much still rated for a maximum current. Now, caveman logic tells me that at peak loads of 120W, where the output is 24V at 5A, the cells in their 3S1P config are going to need to provide that 120W boosted from around 10V, which is something like 12A, and that's before efficiency losses.

The "consumer" market for LFP cells is much smaller than for standard Li-ion/NMC, and I was just barely able to get some LFP cells with a CDR around 5A and capacity of 1800 mAh (NX ACL9073). There do exist 30A cells like this one, but I think they don't ship here, and I absolutely do not trust AliExpress with cells. So while 5A LFP cells are probably enough for my peak 60W usage - if I don't chug that consistently on battery - I wasn't very happy having to settle. It would've been very easy for me to meet those current requirements with Li-ion/NMC cells, but at least I can guess that LFP cells won't turn into incendiary bombs if I exceed their rating for too long.

About cables. The OpenUPS2 store page doesn't say (or I failed the perception check), but it does come with some cables. At least one of these 4-pin male-to-male. Maybe also a 4-pin extension (male-to-female), or that might be my own one, can't remember, but that's what I've got inside this thing now, so you might want one too. A JST PH 3-pin to 2-pin motherboard header cable, which lets it turn on the motherboard via its P6 connector. And finally a USB cable which lets it communicate with the OS and lets you configure it; this is a JST PH 5-pin to a 4-pin USB motherboard header thing, but I can't provide a great pic cause I can't find one online and I've already mauled mine to give it a USB-A connector, since I ran out of internal USB headers.

Separate to the included cables, I also bought one of their 4-pin to DC jack cables, and that standard jack happens to fit my existing power bricks. I also had a basic push button which I haphazardly wired to the P10 connector to act as a power button; when it's pressed, the UPS may be able to switch itself off, though I'll get to that.

So that's how wiring goes. As the manual says though, before you use it in anger, you should configure it over USB. I used "laptop" mode and an output voltage of 12V. You might find that unexpected, since I've been talking non-stop about it at 24V, but the boost converter will be most efficient when the input (~10V on battery) is closer to the output, as long as I actually keep it within the 5A budget at 12V (60W) while on battery - which I have to anyway, given the aforementioned cell CDR fiasco. So I expect to save a small amount of battery runtime there. Another "small" quirk is that if your input voltage (from power brick) is higher than the set output, then it'll bypass the boost converter, which I assume is more efficient (not that anyone would care particularly here). So the UPS feeds my motherboard straight 19V when my 19V power brick is plugged in, and 12V when it's not. That works for me because my thin-ITX board (IMB-1240-WV) accepts a wide voltage like that, but a PicoPSU might not like it.

Now with that, it actually just works fine. Charges the battery and runs the system while power brick is plugged in, and seamlessly continues to run off battery when unplugged. I can also query the status (like capacity and stuff) over USB. But there are two considerable issues I'm facing with it currently.

First, I mentioned the issue of vampire draw in the Devotion thread. My IMB-1240-WV also draws ~5.5W while off, so this must not happen while the system is on battery. The UPS can be told to shutdown, which will totally cut power to the system as I desire. However, the conditions to do this are not intuitive. It can only be shut down if the power brick is already unplugged. Having the power brick in will prevent it from being shut down, and will also wake it. This means that the simple laptop use case of turning off the system while it's plugged in, and then unplugging it and putting it in your bag, will keep the UPS on and draining battery power. I was able to run my cells flat in about three hours just by doing that - I wasn't joking about 5.5W being "really substantial". At least I can say there's a workaround for this though if you exercise diligence. I would attempt patching the firmware so that it doesn't behave this way, but for the life of me, I can't seem to find the firmware sources - presumably because they don't publish them.

Second is simply that the battery integration with the OS doesn't work out-of-box for me, at least on Kubuntu. It's kind of reported as a battery, but doesn't really say anything useful (like remaining capacity or AC presence). I briefly looked at it in Windows, but it seemed to kinda just show 100% and not respond when I unplugged my power brick. It is supposed to support Windows properly though so maybe I've done something wrong there. This is an area where abratchik's Arduino-based USB HID-compliant UPS literally just worked, on Windows or Ubuntu (albeit with a udev rule I had to add there), so I'm a bit confused about it. Again, if I could build my own firmware, I could possibly do something about it, but I can't. I'll see if the OpenUPS is any better here. Or maybe re-flash firmware...or it might be the crappy soldering I did on the USB cable, which I am going to replace shortly...so stay tuned for the next episode of Nobody Asked...

2024-10-28 edit: Re-flashing the LiFePO4 firmware changed nothing about this. My cable soldering was also not at fault, because it's the same deal with a new cable. However, it does actually work in Windows and Linux...but the status is very slow to update, or only does so under strange conditions, like re-establishing the USB connection (physically, or in software). I expect the OpenUPS to behave similarly, but haven't tried it yet. Also note that following the steps to use the OpenUPS2 with NUT on Linux caused my OS to completely stop recognising it as a battery until those steps were reversed, as it seems the interface can't be shared. Similarly, using the CLI tools to query status with `sudo openups -t openups2 -s` does the same thing, until the USB connection is re-established (again, physically, or for example by `sudo modprobe -r usbhid && sudo modprobe usbhid` - beware, do not run only the first part of that command alone, as it will disable all USB devices, including probably your keyboard, until you restart).

Anyway, I'm looking at it because it's probably closest to what I need to design for the production Devotion. The design being accomplished with some hired help, of course. Suffice to say that I don't think the OpenUPS2 is quite optimised for laptop use as-is.

 

[Essence of Flowers]

Just coming back on your requirements, assuming it's the 8700G - I feel sure you should be able to run a build like that with considerably less than 150W, because that sounds crazy high! The highest such figure I can see there is this TechSpot review where they report 148W ingame, and they also report 41W idle, which is also high; my 5600G idled ~20W, and peaked ~80W with PBO on (~60W with it off).

I'm a bit skeptical of those figures, and they seem so much higher than other reviews...maybe they are instantaneous peak values (which should be OK if they're only for a few milliseconds)...but regardless, assuming they're realistic...if you can accept the idea of capping the frequency (e.g. by disabling boost/PBO, or similar for the IGPU only), then 150W would be impossible, and I don't think you would lose much performance by doing this. It's what I'm doing for the Devotion prototypes anyway, testing of which is where I got the ballpark figures of "enable boost to increase power draw by ~35% and performance by ~15% (or cap somewhere inbetween)". That's kind of the whole idea of the APU, efficiency versus DGPUs; low power and modest performance, more for basic games than the latest triple-A.

One more thing - all that wattage from the CPU is going to be dissipated by one heatsink you'll put on it, in a small form factor (in other words not a giant tower like an NH-D15). 150W or anything close to that is a very large demand on a small heatsink and the ventilation generally; best case scenario is that you'll have a burning hot surface temperature with the fan at max. All that is to say that I'd personally consider a realistic max power budget closer to 100W, which is within one OpenUPS or OpenUPS2.

The point about price with the OpenUPS is fair, because we know that production laptop battery packs don't cost that much at all, but that's really economies of scale, and it might be difficult for you to spend much less if you have to experiment with a one-off build. I deeply admire MNT, but as Lux mentioned, I think you would need to make modifications to integrate one of their cell boards (which I also looked at). They're primarily meant to be used in an MNT laptop, which is designed to run at low power draw - so low in fact that they make a point of being fanless.

Though nobody asked, I will also offer my experiences with an OpenUPS2, because I have integrated one into my current Devotion prototype, and it may be of value to you. (I also bought an OpenUPS just to stare at, and haven't done anything with it yet.)

Spoiler: do care, did ask, plus I love you

Firstly, regarding compatible 18650 cells. Examining the OpenUPS2 documentation, I see no reference to CDR (current rating) requirements for the cells, whether LFP or Li-ion/NMC. Same for the MNT Reform by the way, which also relies on LFP cells, though their system is much lower power. I don't know if that's because LFP cells are assumed to be safe when exceeding their current rating, but my understanding is that they are very much still rated for a maximum current. Now, caveman logic tells me that at peak loads of 120W, where the output is 24V at 5A, the cells in their 3S1P config are going to need to provide that 120W boosted from around 10V, which is something like 12A, and that's before efficiency losses.

The "consumer" market for LFP cells is much smaller than for standard Li-ion/NMC, and I was just barely able to get some LFP cells with a CDR around 5A and capacity of 1800 mAh (NX ACL9073). There do exist 30A cells like this one, but I think they don't ship here, and I absolutely do not trust AliExpress with cells. So while 5A LFP cells are probably enough for my peak 60W usage - if I don't chug that consistently on battery - I wasn't very happy having to settle. It would've been very easy for me to meet those current requirements with Li-ion/NMC cells, but at least I can guess that LFP cells won't turn into incendiary bombs if I exceed their rating for too long.

About cables. The OpenUPS2 store page doesn't say (or I failed the perception check), but it does come with some cables. At least one of these 4-pin male-to-male. Maybe also a 4-pin extension (male-to-female), or that might be my own one, can't remember, but that's what I've got inside this thing now, so you might want one too. A JST PH 3-pin to 2-pin motherboard header cable, which lets it turn on the motherboard via its P6 connector. And finally a USB cable which lets it communicate with the OS and lets you configure it; this is a JST PH 5-pin to a 4-pin USB motherboard header thing, but I can't provide a great pic cause I can't find one online and I've already mauled mine to give it a USB-A connector, since I ran out of internal USB headers.

Separate to the included cables, I also bought one of their 4-pin to DC jack cables, and that standard jack happens to fit my existing power bricks. I also had a basic push button which I haphazardly wired to the P10 connector to act as a power button; when it's pressed, the UPS may be able to switch itself off, though I'll get to that.

So that's how wiring goes. As the manual says though, before you use it in anger, you should configure it over USB. I used "laptop" mode and an output voltage of 12V. You might find that unexpected, since I've been talking non-stop about it at 24V, but the boost converter will be most efficient when the input (~10V on battery) is closer to the output, as long as I actually keep it within the 5A budget at 12V (60W) while on battery - which I have to anyway, given the aforementioned cell CDR fiasco. So I expect to save a small amount of battery runtime there. Another "small" quirk is that if your input voltage (from power brick) is higher than the set output, then it'll bypass the boost converter, which I assume is more efficient (not that anyone would care particularly here). So the UPS feeds my motherboard straight 19V when my 19V power brick is plugged in, and 12V when it's not. That works for me because my thin-ITX board (IMB-1240-WV) accepts a wide voltage like that, but a PicoPSU might not like it.

Now with that, it actually just works fine. Charges the battery and runs the system while power brick is plugged in, and seamlessly continues to run off battery when unplugged. I can also query the status (like capacity and stuff) over USB. But there are two considerable issues I'm facing with it currently.

First, I mentioned the issue of vampire draw in the Devotion thread. My IMB-1240-WV also draws ~5.5W while off, so this must not happen while the system is on battery. The UPS can be told to shutdown, which will totally cut power to the system as I desire. However, the conditions to do this are not intuitive. It can only be shut down if the power brick is already unplugged. Having the power brick in will prevent it from being shut down, and will also wake it. This means that the simple laptop use case of turning off the system while it's plugged in, and then unplugging it and putting it in your bag, will keep the UPS on and draining battery power. I was able to run my cells flat in about three hours just by doing that - I wasn't joking about 5.5W being "really substantial". At least I can say there's a workaround for this though if you exercise diligence. I would attempt patching the firmware so that it doesn't behave this way, but for the life of me, I can't seem to find the firmware sources - presumably because they don't publish them.

Second is simply that the battery integration with the OS doesn't work out-of-box for me, at least on Kubuntu. It's kind of reported as a battery, but doesn't really say anything useful (like remaining capacity or AC presence). I briefly looked at it in Windows, but it seemed to kinda just show 100% and not respond when I unplugged my power brick. It is supposed to support Windows properly though so maybe I've done something wrong there. This is an area where abratchik's Arduino-based USB HID-compliant UPS literally just worked, on Windows or Ubuntu (albeit with a udev rule I had to add there), so I'm a bit confused about it. Again, if I could build my own firmware, I could possibly do something about it, but I can't. I'll see if the OpenUPS is any better here. Or maybe re-flash firmware...or it might be the crappy soldering I did on the USB cable, which I am going to replace shortly...so stay tuned for the next episode of Nobody Asked...

Anyway, I'm looking at it because it's probably closest to what I need to design for the production Devotion. The design being accomplished with some hired help, of course. Suffice to say that I don't think the OpenUPS2 is quite optimised for laptop use as-is.

Your friend is quite right, chaining multiple charge/discharge units isn't safe without additional protections. One reason Lithium batteries go spicy, is the cells not being evenly charged. That said, 18650's are safer than flat cell lithium. Simply due to the hard casing and fail-safe vents, if they do go critical.

That is a super cool platform MNT is building. I think you could adapt part of what they've built to your needs. Something I noticed though, is the lack of a 12v output rail. They use the LM62460 and MAX1837EUT33 converters to supply 5v, 3.3v & 3.3vSB, which is good. You do require those voltages but 12v as well.

Thus you'll need to add another DC/DC boost converter. Keep in mind all of your high draw components(CPU) will be on the 12v rail. My research says, it's typically not recommended to draw more that 3.5amps per 18650. If you've got eight in a system that translates to ~120w peak. So to be safe, you'd want to limit the CPU to ~80 watts.

Keep in mind, most laptops these days have CPU's between 10-45watts. Partly for cooling but also to reduce battery wear. High performance laptops will sometimes have charging bricks up to 180w but they won't pull that much unless it's plugged in and under high loads. On battery the hardware knows to set limits.

Here the DC/DC boost converter I'd probably start with. You can dial it at 12v up to 100w. https://www.jacobsparts.com/items/DCMOD-D

Hope that helps!

Yeah this basically got me back to my drawing board of maybe using an openUPS, my plans at this point is fully build the laptop and have the picoPSU do some stress testing, see what my peak wattage is, and assuming it doesnt exceed certain limits we go from there.