/tech/ - Technology★

POWER9

DIY SoC implemented on a FPGA.

>inb4 botnet in FPGAs

Take that old Nintendo wii out of your attic, and BAM, you've got a PPC computer ready to go.

>>936293

this

>>936292 (OP)

>What considerable alternatives are there?

Besides PIC https://en.wikipedia.org/wiki/PIC_microcontroller there's only IBM openpower with POWER9 and the Talos motherboard.

From what I know it's the only and least evil solutions (since we still don't have the blueprints nor the vhdl).

Second part:

OpenPOWER (Supported Distros: Fedora, Debian, CentOS/RHEL, OpenSUSE, Ubuntu, BSDs):

Raptor Engineering sells POWER9 workstations, that may soon be getting RYF certification.

They're expensive as fuck, but probably the most powerful non-botnet computers that exist. Comparable to Xeons/Epyc.

https://www.raptorcs.com/TALOSII/

PowerPC (Supported Distros: Gentoo, Free/Open/NetBSD)

The company that still makes this is NXP

https://www.nxp.com/products/microcontrollers-and-processors/power-architecture-processors

Here is a project for a Libre PowerPC laptop using NXP, shooting for RYF certification.

https://www.powerpc-notebook.org/faq/

EmbeddedPlanet has several PowerPC SBCs, most using NXP.

https://www.embeddedplanet.com/product/single-board-computers/

If you have old PowerPC Apple products, they use OpenFirmware, which is here

https://code.coreboot.org/p/openfirmware/source/tree/HEAD/

Recommended models if you want to buy one: late PowerBook G4, PowerMac G5

MIPS (Supported Distros: Debian, Gentoo, NetBSD):

The /csg/ of desktops. Lemote is a chink company that sells libre MIPS boards, using PMON firmware.

http://www.lemote.com/html/product/

EmbeddedPlanet also has MIPS boards with processors from Cavium with U-boot firmware.

https://www.embeddedplanet.com/single-board-computers/processor/cavium-oceteon-ii/

GnuBee has two low power NAS devices. They're cheap, they use MIPS, and they're going for RYF!

https://www.crowdsupply.com/gnubee/personal-cloud-1

https://www.crowdsupply.com/gnubee/personal-cloud-2

SPARC (Supported Distros: *BSD, Gentoo, Illumos):

Legit server stuff here. Open ISA and Open Boot firmware.

https://code.coreboot.org/p/openboot/source/tree/1/

Below link is a big list of options, from ancient 80s shit, to modern servers released this year with 5.0GHz and 32 cores

And of course, everything in between

https://en.wikipedia.org/wiki/SPARC#Implementations

RISC-V (Supported Distros: FreeBSD):

Only SBCs here. SiFive has some.

https://www.sifive.com/products/freedom/

There's also LowRISC

http://www.lowrisc.org/

For Routers, check out LibreCMC and its supported list.

https://librecmc.org/

https://gogs.librecmc.org/libreCMC/libreCMC/wiki/Supported+Hardware

>>936299

>PIC

I just recently discovered retroBSD, which can be installed on a PIC32. Kind of interesting, but obviously impractical for anything modern.

>>936294

This is the way to go.

>>936293

>Big endian architecture

Too much of a hassle.

>>936294

This tbh

>>936294

How fast would that be compared to a standard IC?

>>936333 (checked)

Just use FPGA for prototyping, then use

https://en.wikipedia.org/wiki/Multi-project_wafer_service

>>936334

You design it then hand of the designs to some unknown for fab? That doesn't sound very good, especially if it's anywhere that has Chinese or communists involved, like Germany.

>>936335

Who cares if they steal your design? Signals intelligence agencies won't be able to design bugs into it in the time period from you handing over your design to it being created.

And even if they had some wonder team that can do so in days (they don't), they wouldn't bother to do it for small student projects being produced on only a few wafers.

Now, if you do this over and over and they are aware you are creating non-compromised CPUs, then they might spend some time to design bugs in future versions and approach the fab, but more likely they would approach you first, so you would know they know about your operation.

Easy way to avoid this is just make all you need in a couple of years and then sell them AFTER they've been fabbed.

Also, since you're putting them together, you could image them under a microscope and see if the layout matches your design.

>>936338

Do you have any proof about what you are saying, because otherwise doubt is not a reason for trust.

>>936335

Also, the

>inb4 botnet in FPGAs

is dismissing the fact that all large FPGAs are botnetted!

It's like saying you're buying an intel cpu for a 'secure desktop' you're building and then saying

>inb4 IME botnet

>>936339

Learn how much time it takes to design a complex integrated circuit, and then determine the time required to modify the design in a way that compromises its security and is difficult to detect.

You are implying they can do so in a short period of time, which is a much bigger assumption IMO.

Also, since you can view the final product under a microscope and match it to your own design, the point is moot.

>>936341

I am less worried about them introducing new designs and more worried about them finding flaws in the design and exploiting it that way.

>>936342

That's a good point. Only a verifiably correct and bug free design (not sure if possible for hardware and would probably make design take 10x longer) could alleviate that risk.

And even if one were able to produce verifiably correct hardware on a design level, I just discovered this

http://sharps.org/wp-content/uploads/BECKER-CHES.pdf

which would make microscope analysis of the finished product useless for detecting introduced bugs.

>>936344

Still though, I think FPGAs with coprocessors on them are a bigger risk.

http://www.clifford.at/icestorm/

The ice40 FPGA series are the only FPGAs (there are no CPLDs or ASICs ones) who can be programmed with an entirely free toolchain (from a HDL to the bitstream you send the FPGA) meaning you don't need to use any huge, bloated, and proprietary IDEs by the manufacturer.

The microprocessor I'm currently designing is targeting one of the higher end ice40s. It's kind of a fun project and am planning to at least support VGA and PS/2 connectors so I can use it as a simple computer.

>>936301

Thanks for the list. Raspberry Pi isn't listed, and I also remember reading somewhere that it has something like Intel's management engine in it. Do you have any information about it?

>>936387

For that one, it needs a binary blob to even boot properly, afaik

>>936390

Is there somewhere I can read about this?

>>936333

Probably about 100-400 MHz.

>>936292 (OP)

There are several threads in the catalog on this exact subject

>>936402

Ehh. I wonder how bad they are after coupling that with their IPC.

>>936301

>>936302

Librebooted ThinkPads are unstable in terms of performance (probably the lack of microcode). That's why I'd recommend a POWER9 workstation over everything.

>>936293

Basically this and only this from a practical perspective, everything else is RPi-tier in terms of performance.

>>936330

Power8 and newer supports both BE and LE you mongoloid, the CPU firmware selects which mode at boot and by default selects LE. Thats why Canonical and such list PPC64LE

>>936334

Even using MPW you are still looking at thousands of dollars per 1MM^2 for old processes like 40nm and tens of thousands for newer processes like 14nm.

>>936335

TSMC and other large fabs deal with thousands of customers a year for their shuttle service, do you really think they are going to destroy their reputation by going out of their way to reverse engineer a CPU designed by some nobody just so they can insert hardware backdoors into a CPU which is going to be used by a grand total of one person? You are delusional as all hell. Reverse engineering even a small CPU from the masks is like reverse engineering a compiled binary by hand without knowing what each instruction actually is.

>>936333

Slow, I remember seeing something about Nvidias massive clusters of FPGAs they use to prototype GPUs and they run at a fraction of the speed that the final product does.

>>936402

Probably on the lower end of that for anything larger than a Microblaze or generic MIPS processor.

>>936455

Industrial espionage is a real thing.

>>936458

>Industrial espionage is a real thing.

I know, but last time I checked the guy I was responding to probably isn't Jensen Huang or Lisa Su and it was TSMC being mentioned and not some no name fab in China. TSMC and Global Foundries aren't stupid enough to physically alter designs because their business depends on large companies trusting them with their IP.

>>936461

Also a reminder TSMC (Taiwan Semiconductor Manufacturing Company) is NOT Chinese no matter how much the putrid soulless Chink drones try to take over Taiwan's national identity and turn it into a puppet state

>>936682

Sorry meant to reply >>936390