Ideas for processors for MNT Reform

Why am I imagining a huge SoM with a huge heatsink coming out the bottom and it’s own power supply?

I think the intention of microwatt is to fix this problem and be more risc-v like in chip size as well as making it way more lightweight in how much watts it uses and maybe better even.

as far as openpower is concerned, it’s a long ways off but the https://libre-soc.org/ project and a mnt reform module would go hand in hand and would be pretty cool. one of Their goals is a totally open openpower SOC even down to ddr4 bringup stuff. This is the dream and maybe we’ll get to see it someday!

also am i hallucinating or did powervr actually start work open source mesa drivers https://www.phoronix.com/scan.php?page=news_item&px=Open-Source-PowerVR-Vulkan? That could open up some possible SOC options, I know at least a couple higher end renasas soc’s use powervr gpu’s. Their new lower end ones use mali based gpus.

I think one of the biggest problems is all easier to source western SOC’s are networking/industrial/automotive based which may be less ideal for general purpose usage where fabless and chinese companies like rockchip/amlogic/mediatek dominate the general purpose and tvbox soc market. and the trend doesn’t seem like that trend will reverse anytime soon. (I.MX9 series aiming on edge computing, the newer ti chips with a bunch of arm core R realtime coprocessors)

1 Like

Hello! I was thinking about alternate modules for Reform and I got to wondering… if we got enough people to agree to buy, would it perhaps be worth the time of an ODM to design and produce a limited run of Reform SoMs?

It seems like a chip like Intel’s Pentium N6000, at ~6W TDP is tailor-made for an application like Reform.

There are a ton of mini PCs out there from companies like Topton that make and sell computers based on similar chips for ~$300 USD with RAM and SSD inside. The expertise is out there in spades.

I know it wouldn’t be open hardware, but it might be an interesting proposition to some. The current module is really close to meeting my needs but a lot of software, frustratingly, still does not like the ARM platform.

1 Like

I’d definitely be interested in an AMD Ryzen SoM, I looked at the one referenced way above with a view to an adaptor board, but they dont provide the relevant documentation and would probably want money and/or an NDA for it.

Yeah absolutely! The N6000 with max RAM (16gb) and 4 threads would be excellent.

If there’s more interest overall in Ryzen, it might be worth it instead to head in that direction with the Ryzen Embedded R1102G dual-core at 6W. I think when you get too hot, that becomes a huge issue, and potentially introduces scope creep because of the need for a more robust cooling solution.

EDIT: Thinking about it a little more, it seems like the cheapest solution would be to do a group buy of the AVNET Ryzen SoMs and pay a bounty to someone in the community to create an adapter board. Does anyone know if this is feasible?

Edit 2: The AVNET boards follow the SMARC standard, which appears to be well-documented! I was able to access the whitepaper for the 2.1.1 standard here. It’s 314-pin and is used by a ton of other embedded compute manufacturers, it seems! There’s a German firm that offers design services based around the standard as well, would probably be an easy job for them.

Is there anyone in the community who knows anything about SMARC?

Alright so I think this might be a dead end due to space constraints…electrically it seems feasible but mechanically it appears extremely hard to fit. In order to visualize, I cut out a rectangle the same size as the small SMARC board standard:

The wider SMARC connector would need to clear the SODIMM clips on a theoretical adapter, pushing the adapted SoM to the right. Even if there’s enough room, there would be very little depth left with a board on top of an adapter board…and then there would need to be a heat spreader attached as well… hmmmm.

If you want to go this route a fork of the motherboard would be the best way I think. Maybe there are even other interesting SMARC modules, but I don’t know how the real-world compatibility is among them.

2 Likes

you are amazing, well done for finding that. Theoretically with a pinout for both SODIMM and SMARC it should be possible to produce a passive adaptor board, which is what I was planning to have a go at (zero experience lol). Lack of space is a bear though.

ive seen a few different intels, and there’s an SMARC variant of the the SoM used in Reform that i’ve seen as well

SiFive Performance P650 - SiFive looks promising. Hopefully someone will stick it on the CM4 footprint?

1 Like

Or the T-Head TH1520 quad-core Xuantie C910 looks interesting

CM4 https: //github. com/antmicro/arvsom > Adaptor

1 Like

the blocker with that would be it doesn’t appear to have any pcie or indeed sata, so you’d effectively lose the two expansion slots on the reform board

been mulling it over and thinking about The most important things to consider for processors:

  • Open firmware and linux support
  • Openess of SOC documentation
  • Availability/Ease to buy SOCs

A little sidenote is my perspective is that of someone that likes making little development board and has a little daydream of prototyping their own module. I have 4 years professional experience in SMT assembly and have a good amount of basic MCU dev board design experience so it’s not impossible but there’s still much for me to learn for more complex application processors.

I think the CM4 adaptor is a great compromise and work around solution, it lets you use available modules vs building your own and there’s a lot of rpi clones which makes the compute modules more commercially accesible. I don’t have the money but otherwise would jump on the AM311D without hesitation.

In terms of procuring SOCs and designing for them in the west I find that NXP, TI, and then Renesas are the most open in terms of purchasing on big distributors (namely mouser and digi-key).

One of The biggest problem with x86 chips is you can’t have truly open hardware designs and schematics, which makes it a dead end to pursue in my opinion, but also this is hardware hacker world if you want to explore and try things Do it!!

Back to ARM soc’s. NXP, Ti, Renesas MPUs and application SOCs are mostly focused IOT, edge computing, industrial control and communication and HMI(human machine interface). A lot don’t even bother with integrated graphics unless you’re HMI with a graphical interface and they’re not really focused on general computing or desktop computing tasks. Not saying they can’t work it’s just maybe using a tool that’s close to right for the job but not intended for it.

The other spectrum is SOCs generally aimed at media boxes, tablets, etc. Most of these are from fabless semi companies which chinese produced chips, which i mention not in a negative way just in a degree of separation from western and english speaking world. None of the big component retailers stock these chips but stuff like Rockchip and Amlogic. These are common in maker boards from pine64, and hard kernel and etc. These companies are actually very key in bridging a gap and making the SOCs more available to the maker/hacker community. Making things like the banapi cm4 option a viable one.

Rockchip seems to be the easiest one to get freely available design and datasheet documentation from, and probably the easiest one to find SOCs via alibaba.

Allwinner is another common one but I heard they were notorious for GPL violations.

AMLogic has some very good SOCs but their website is horrendous and i think you have to deal with them directly as a company to actually get any info or purchasing which makes them feel more like one of the x86 vendors.

so with all that being said here’s some SOCs i’m thinking of looking at

TI-AM625. This is a new SOC from texas instruments. There is nothing too spectacular about it, it’s a quad core A53 clocked at 1.4GHZ with an M4 coprocessor and some ti PRUs coprocessors. A dual core variation was chosen for the new Beagleboard and this means open source reference designs are readily available. I also find TI’s documentation to be very informative for someone like me who is learning as they go as well as being available no strings attached. This would be more of a side grade howevre one interesting note is that afaik it uses a relatively new Imagination tech gpu, the AXE-1-16. Normally imagination (powervrs) is no go for open source however this the company itself has pledged to develop (already in tree) open source drivers for this GPU as well as 2 others; BXS-4-64(i believe this is found in the lychee pi4), and older PowerVR GX6250 (some mediatek socs and maybe 1 renesas rcar). Another thing is this ti SOC is CHEAP which is attractive to me as I am poor so spending money on prototyping can be tight at times so a single unit SOC at 25$ is very attractive vs nxp’s 50-100+ single unit costs for their i.mx8 socs. This imo makes it very attractive option to build a mount reform module from a diy’er perspective.

following up is rockchip, the new 3588 and related are hot news(maybe medium and luke warm by now) in maker SBC land so finding reference designs might not be the hardest and there are definitely people working on mainline support but it probably won’t be as clean as what we have for the mnt reform. Still these processors seem more focused on general computing and graphics and as far as chinese chip makers might be the easiest to get your hands on documentation as well as the actual chips themselves.

Renesas has some vehicle infotainment focused SOCs in the older RCAR line
like this one: RZ/G2M. This is a dual A57/quad A53 chip for automotive units. It has that older gpu listed in the imagination press announcement about open source drivers. I don’t know what other peripheral and driver support is like but it may be worth peaking at briefly at least. renesas likes to restrict some documentation behind log ins but at least it’s accessible. It will probably be harder to find this specific SOC compared to ti and nxp on retailer sites but still possible.

so as far as making an Reform compatible open source hardware module on a maker diy level i think the TI AM6254 is the most attractive in terms of ease and cost but the rockchip probably most attractive in terms of performance. The renesas is more of a third, weirder more curious option worth a few seconds of thought at least.

I hope this was a helpful write up, I have spent a lot of time thinking about this and looking at SOCs before closing a billion tabs and deciding not to do anything because money/time and needing to focus on more pressing things. I would like to try to make the TI module for fun on a hardware level. (I like hardware, have a hard time with firmware)

2 Likes

Thanks for posting this! The little TI chip seems like it could be good for a “value engineered” version of Reform for education, and it’s probably not that far off from the iMX8MQ, at least.

1 Like

Thanks!! I think so too? It shouldn’t be far off the base module we use. I mentioned the thread on my mastodon yesterday and lukas pointed out a new (pre production) TI SOC is an octo-core A72 that uses the BXS-4-64 gpu and stuffed full of a lot. That could be a fun one to mess around with if you have the money and the chops.

2 Likes

I have had a look at the various chips found in single board computers and those mentioned in this forum. Whilst feeling intuitively that a newer “standard” chip than the NXP i.MX8M Plus is needed but this is a field I know little about. I do wonder whether the cores do need to be so old, the Cortex A53 is 2012 so over a decade old, and the Rockchips used in Pine products might have an A55 if your lucky. These cores use ARM’s older mk8 instruction set, whereas the successor designs are the A510 and A520 using the mk9 ISA. I presume this may be something to do with the latest cores attracting higher licensing, but it seems slightly disappointing that everything available to us is so far behind the state of the art.

Which is where the Risc-V comes in. There seems a better possibility of getting a state-of-the art chip with that ISA, because RISC-V simply did not exist when the ARM cores we are looking at were developed.

Regarding performance and efficiency cores, I see myself benefitting more from a clutch of efficient cores rather than a few fast ones. Haiku (I have experience with) and Genode (which you have on Reform) let you turn individual cores on and off in realtime. This post from Genodinans shows the level of control that you can have over your individual cores. So I feel there is a lot to be said for a direct successor to the i.MX8M, providing newer cores and perhaps more of them. Would that not literally be its direct descendent within NXP the i.MX 95? This CPU consists of six A55 cores to the four A53 of the i.MX8M.

Well, currently the best availiable Risc-V SoC for Linux is the JH7110 from Starfive AFAIK. You can get a CM4 style module with it: Milk-V Mars CM | A RISC-V Compute Module in compatible form factor

1 Like

Not being interested in linux, Haiku may be the killer app behind bringing RISC-V to the Reform. Its devs have coalesced round the VisionFive mk2 which happily uses this chip (as does the Pine Star64). Haiku never really got going on ARM architecture for various reasons.

So perhaps Milk-V Mars is a Haiku compatible option that can be done (almost) straight away for development purposes?

As for a “standard fitment” RISC-V solution, my only concern is that RISC is still quite fluid and the niche OS like 9Front or Genode may be in a position where they can only support a handful of SOCs well. And they work happily on the existing i.MX8M. So we would want to be confident that the RISC-V SOC we adopt is likely to have a long service life with headroom to grow, and enough of a performance uplift that these OS benefit from migrating over. Might it be worth seeing if StarFive bring the new generation of HiFive cores namely the 600 (BIG) and 400 (little) series onto a SOC in the near future?