Ideas for processors for MNT Reform

Or other mnt reform projetcs, maybe even then one I mentioned in my other thread about the mnt reform frankenstein project.

Would either of these processors work for this?

LS2084A and LS2088A

or, LS1046

there isn’t exactly an information site on which arm processors can be used without any insecurity blobs/blobs in general.
IF nothing else, I just decided to throw these to you as possible processors for the future. I assume you already looked at these, but yeah, I am curious if you have tried. And if the possibility still exists in these or even newer a73+ cortex based processors.

Anyways, feel free to reply.

AFAIK the main problem is that both LS2084A, LS2088A and LS1046 don’t have embedded video core.

LS1028A is the only that has it.

Layerscape series is mainly for network switches, as far as I got.

So I guess while it’s possible to make SoM with e. g. LS2088, you couldn’t use it without external video card.

1 Like

Alright, so I take it there isn’t anything with Quad Core A72 that is as libre and fast as the imx8m that also can use 4GB of ram or more while still being as functional?

I wonder if someone ever plans to reverse engineer A75 cortex processors or later in the future.

Would be nice, ah well, I can dream. :wink:

Anyway I think even dual core LS1028A with A72 is better than quad core A53.

Also @mntmn is currently working on an adapter for Raspberry Pi Compute Module 4. It has quad-core Broadcom with A72 cores. But probably it will require active cooling.

1 Like

Hmm, I thought it was very blobbed though? the raspberry pi 4 module aka…

Question though, how much faster is A72 then A53? I am guessing it must have more than twice the speed at least. Am I right?

Or no?

It has blobs. But for me it’s “acceptable”.

A72 is out-of-order core so the performance is much better.

Better to just look at Pi4 vs Pi3 benches (both have quad-core processor: A72 vs A53, just like in our case)

But are the blobs like ones found in the imx8m? Or do they have real security issues? I guess that is what I wondered.

Aka, by imx8m blobs I mean the ddr4 controller blob. The arm chip is usable without blobs from what I remember…

Unless I am confused?

Also, how much faster is the a72 then the a53? The cortex, I mean.

EDIT: I just looked and it appears the speed of a72 might be on a whole another level from what I see… usb speed, ethernet, clock speed, etc…

2x or more it looks like… :slight_smile:

You mention USB and Ethernet speed, but these don’t depend on C72 vs C53. These are mostly constrained by bus/Network-on-Chip architecture and chosen IP for these functions of the individual SoC, which varies greatly between chips and vendors.

The next “official” MNT Reform CPU module is the one with LS1028A (dual Cortex-A72) and 8GB or 16GB DDR4. The sources are already there: Reform / MNT Reform Layerscape LS1028A SoM · GitLab

Another interesting bit in the LS1028A is the cache coherent interconnect. I can’t say yet how performance and bandwidth differ in detail though.

The remaining problem is that we don’t have a prototype of the module yet, as the processor shipment to RBZ gets delayed all the time.

2 Likes

Note that the Pi4 CM adapter is more of an experiment. I personally wouldn’t use it as a daily driver, as I prefer the i.MX8MQ for its architecture. Also, there is only a USB2.0 root port on the Pi4 CM, no USB3. The DSI output is constrained by secrecy/missing documentation, so we have to go through a HDMI->eDP converter for the internal display. Moreover, the main CPU in the system is actually the VideoCore which requires binary firmware if I’m not mistaken.

1 Like

So your saying the Pi4 CM thing is not really important, the layerscape 1028A is however important. :slight_smile:

As for 1028A though, do you know which runs faster for you when you have compared their speeds of processing? I am curious what you have to say.
Also, is getting 16GB even have a purpose if you using an arm processor like Layerscape LS1028A? Or would it make more sense for most people to get 8GB of ram with it. Just wondering?

It’s too early for me to answer these questions. I’ll write more about this topic once we have the hardware prototypes.

Out of order execution also means complex secret firmware on the chip for selecting instructions. I guess it’s alright though as long as we can choose the old ones.

Hopefully he has taken this into account and its not a threat?
Idk?

Yeah, it is kind of early… by the way, can you answer the 11th response from Bibilocar?

is there any complex secret firmware required that is a security issue?

Respond whenever

Proprietary firmware does not mean that it’s insecure by default. It could be dangerous then you keep features like Intel ME or AMD PSP turned on, but it’s not a case with ARM processors AFAIK.

In my opinion most privacy and security issues come from proprietary OSes and apps.

I can suggest you to create your own threat model at first. Do you really scare relatively rare hardware security issues?

I think open source firmware is a really cool thing when you really know what it does and want to experiment without the surprises that black boxes usually have.

Depends, sometimes your right, sometimes your wrong, isnt there some sort of arm type risk called the Arm Platform Trust Zone?
I recall somewhere hearing some arm processors have something close to equivalent…

My threat* model is this, does it dial when your connected or not connected to the internet, to places other than what I am connecting to currently.
Aka, are there any remote issues with having out of order execution?
I know you can break into most hardware easily if you have physical access, that isn’t my issue. I am speaking specifically, about over the net, or remote security issues as a whole. That’s my threat model. :wink:

Seems I had a typo, also I looked it up and it seems the trust zone can be controlled by the user? So maybe its not a big issue? Idk… heh.

AFAIK there is no known CPU firmware exploits to hack your PC over net except Intel ME and alternatives.

Out-of-order execution means that it could be vulnerable to a newer SPECTRE like attacks I guess. But all of them only allows an attacker to read bytes SUPER slowly from CPU cache. While it’s still dangerous, this attacks are mainly very limited.

I guess then there is no big issue with the LS1028A then…

I still await the future. I am sure he will have more info on his other projects and the LS1028A, etc… soon.

The i.MX8MQ module with in-order Cortex-A53 cores will always be available. The new modules are just alternatives. The idea is to create a lively ecosystem of modules around MNT Reform.

5 Likes

I guess in case of success you will just provide an option to order a new MNT Reform with chosen module?