What are the chances of someone making an adapter for this powerful board (apparently beats the RaspberryPi 4 in peformance. Perhaps even better an adapter from clockworkpi team itself? so expected price should be around $99 for the 8GB. Has a 260pin SO-DIMM layout (DDR5 maybe? not sure)
Anyway, this is probably the most powerful board on RISC-V side, today. Ideas, anyone into CAD stuff?
A better RISC-V core has been really high on my list of desires (along with an x86 core) for ages now ever since I got my R-01 for my DevTerm. The lack of grunt in the R-01 is rather frustrating, I’ve had a lot of fun and enjoyment out of my VisionFive 2 SBC that really shows the difference in power between them, now I just want that power on the DevTerm or uConsole.
Yes, I can relate to that, well you have on the RISC-V something to buy now, the MILK V, from my last post.
On the x86 side, not too sure if there will be a clone on RPi Module clone on x86, or a SO-DIMM on the architecture (I could be wrong here), also the battery would just be drained fairly quick, unless you’re ok having the USB plugged all the time during usage.
Either way, yes we need more power to play around either RISC-V, ARM or X86
Definitely an interesting idea and its definitely an insanely fast processor according to specs and by the looks of its a DDR4 SO-DIMM form factor.
In regards to putting it into a Uconsole or devterm given the height of the CM4 module in comparison i think there is enough room for there to be a sort of sodimm stack the biggest problem is DDR4 and 2 have slightly different external dimensions 4 being wider so having a board that flips the orientation of the LM4A module by 180deg and raises the module a bit an 8H or possibly 5.2H DDR4 connector on the far end of a DDR2 adaptor board should lift it above the clips of the DDR2 socket and have the LM4A sitting just above where the DDR2 form factor module would fit, i think it should work but the spacing seems a bit tight, a couple of FFC connectors snugged on the back for the extra 54 I/O pins of the DDR4 header wouldn’t go amiss either.
Until further specs on the detailed pinout is available these are just musings, hopefully the full specs are released soon I’m quite interested in this as well.
LM4A (this TH1520 module) wiki page:
https://wiki.sipeed.com/hardware/en/lichee/th1520/lm4a.html
LM4A is used on LicheePi 4A SBC, so you can also check it’s wiki:
https://wiki.sipeed.com/hardware/en/lichee/th1520/lpi4a/1_intro.html
And Base board Schematic download link is here, you can using it to get LM4A’s pinout:
(btw, looks like it is using jetson nano’s pinout, need some research and confirm…)
Ah thanks the sipeed LM4A’s wiki page had no details of its pinout but the dev boards wiki does i thought they hadn’t released docs since the github on the modules page was empty, haha i should look a bit harder next time thanks.
A slight update is probably overdue.
I’ve mostly worked out the adaptor, but I’m gonna be putting this on the backburner.
The main issue I’ve noted and one that alexmakura mentioned earlier is power consumption and the minimum recommended power specs for the LM4A is listed at about 24W, this is probably under operating load, but none the less the main issue is the max power output of the uconsole/devterms mainboard or more specifically the AXP228 is somewhere around 13-18W while on battery power and 17.5W while on external power exclusively meaning that even if you had it plugged in it would still be loosing charge while under load and couldn’t run on battery alone.
So the only way to make it work would be by adding a separate voltage booster to directly pull power from the batteries possibly another TPS61178 like the one used for the existing system 5V regulation, this would make it able to run on battery alone under load but would still mean that you’d have to either use it lightly or not at all while charging, due to the inbound current limit, because that 17.5W is also needed to charge the batteries themselves.
All of the above make it too impractical to viably use,
It’d be far to much effort for the limited advantages this might bring especially since it’d probably overheat and throttle anyway from the limited heat dissipation of both the Uconsole and devterm and I personally don’t need this thing to be a high performance device.
I might continue work on it once my Uconsole arrives, but until then its going to be shelved.
wow, you definitely did some work, many thanks for the update and interesting when vendor makes compatible devices, doesn’t always means it will be, but as always many factors will affect the compatibility. In this case power consumption.
I can help making a devicetree for TH1520 that limit the max CPU boost frequency so it will not consume that much power.
(But I have no power meter to check how low power consumption can reach before it’s unusable… Maybe I should try to get one before I making my own adapter…)
That’s actually a good idea, similar to the RPi (with commands to increase), and I also wonder if the AXP228 will cap at 13-18W or that is just theoretically or average, everything needs test or some try and error stuff to see the limits.
The 13-18W figure is from the theoretical absolute max current from the battery circut in the AXP228 i dont know how hard a current limit that is or how it handles the varying battery voltage hence the 13-18 part based on the stated max and min voltages its charge circut will allow and the 4A max current.
The batteries can definitely provide more current most li-Ion cells are rated at min for 1C discharge, but many are rated at higher C values so even cheap 2200mAh cells could easily output an extra few hundred miliamps per cell than what the AXP228 will theoretically allow.
The the slightly bigger issue was the max input current from the external inputs being limited to 17.5W 5v@3.5A up to 9 of those watts would be needed for charging the batteries technically you could get a few extra watts if you wired up the unused secondary power in circut which will grab an extra 1.5 Amps but thats brings you to the limit of what USB C will allow current wise at 5v and that still only adds up to 25W max, which might be enough for operating at load without batteries (if the PMIC doesn’t overheat), but would require modifications on the mainboard to accomplish.
Where’d you find this number? The docs for the Lichee Pi 4A as a whole state minimum input is 10W, recommended 15W+, and it should have a higher consumption than the LM4A on its own. Their listed power consumption tests show numbers beneath 7W, although admittedly without really fully stressing it in terms of connections & peripherals.
My own quick testing with a Lichee Pi 4A showed only about 10W draw from a smart plug, while stressing CPU and wi-fi, with keyboard, mouse, and HDMI monitor connected…
Huh fair enough I was going off the LM4A dev boards requested min power in its docs being 12V@2A thinking it’d be comparable to the ClockworkPI board and accessories in terms of net draw since I was working with the net power limit of the boards power circuitry and Sipeed add “at least” to every power spec they suggest and on the dev board they use DCDC converters that can handle 12 Amps! so I was under the impression that the 10W was probably a lower end “at least” sorta figure.
And I’m also half asleep most of the time when I’ve been doing this stuff after a usually long day so using the devboards power specs was a probable mistake, and in retrospect it’d probably work fine even if it power throttled a bit; even right now I’m tired as F’k, but at the time my thinking was if we’re shoving a high spec part into a small device I want to use every iota of performance that part has to offer, so trying to work in a overhead so not to start overheating components was also a factor in my thinking as to not catch anyone’s devices on fire.
Side note are you powering your LM4A via USBC or its DC in? because according to its purchasing page which admitably I didn’t actually look at up until now its USBC power is limited to 10W (5V@2A).
Yeah their power budgeting on the Lichee Pi 4A is a bit sus on the whole, the board clearly doesn’t have enough to run all possible peripherals at their full draw (e.g. that’d be 18W just on the 4xUSB3 ports).
Temperature is probably more of a concern. These do run hot, although they also thermally throttle themselves just fine. I suspect the main reason they’re shipped at 1.85Ghz instead of 2.0+ is that the small fan they ship with isn’t enough to prevent throttling at higher clocks.
USB-C that time, will re-do testing on 12V DC later.
Actually, it’s because normal yield can only run stably at 1.85Ghz, some very good yield can run 2Ghz+… with price of a ton of heat generated.
(kind like that AMD/Intel CPU that some can overclock to higher frequency.)
You’ll also need to verify the limits of each power domain (5v, 3.3v, 1.8v etc.) of the cpi motherboard (it combines one or more DC-DC from axp228 for each of these) against the requirement of the LM4A core board.
The AXP228 has an IPS module that switches power from battery and external inputs that then feeds into multiple separate externally accessible DCDC converters within the AXP228 PMIC for the 3v3 and 1.8v nets, but the 5v net is handled by a separate TPS61178 boost converter that pulls directly off the IPS output and it has much higher max ratings than the AXP228, so for the 5v net which is the only one the LM4A uses, theoretically can pull as much power as the AXP228 can provide.
So I re-did this on 12V input, draw under the same conditions went up slightly to 11W. I can add some USB drives and increase the draw by a few W by using them, so it doesn’t appear to be input limited here, it just doesn’t draw that much.
That’s excellent news it clears up the “at least” figures anyway meaning that the clockwork mainboard should be able to handle the LM4A perfectly fine.
Yes ! that’s great news… thank you so much… do you have any drawings yet or ideas how the board could be, or look like? probably similar to the RPi maybe?
