Libre/Open-Source POWER10 Hardware Systems Unlikely Until At Least 2022

[madscientist159]

These aren't off-the-shelf SoCs IBM can ship at a moment's notice. They come off a separate line fused on-ROM with different (Raptor's) master keys. And looking at how Intel couldn't book new GPUs with TSMC all the way through 2021, I'm guessing no one is willing to book special/small scale 7nm fabs right now. After all, Global didn't gave up on this node for no reason.

Sorry, but I do need to correct this...while I cannot make any statements about POWER10 (this is because it isn't released yet by IBM), I can absolutely confirm there is no special handling, key fusing, or anything along those lines required for POWER9, and that POWER8 was the same. OpenPOWER CPUs do not come with any kind of fused key; the end user needs to program their own key into the CPU if desired, and even then that process just programs an EEPROM inside the CPU if physical presence is asserted or the booted firmware has a valid signature as determined by that particular CPU. This is what the "Secure Mode Disable" physical jumper is for, fundamentally, to allow recovery of a system to which the signing key is lost via physical presence at that system.

Raptor in general believes fused key systems to be dangerous on multiple levels (e.g. how does key rotation work? does the vendor retain control or is all control passed to the owner?), and either avoids hardware that requires them or deactivates them prior to shipment with a public well-known key as required, across all of its product lines

Make sense?

[rmfx]

Sounds a bit like openwashing.

We try to get a open reputation for the product but in fact, open just get the crumbs years later meaning that we let them fix the old issues and we focus on the new cool stuff.

[starshipeleven]

These aren't off-the-shelf SoCs IBM can ship at a moment's notice. They come off a separate line fused on-ROM with different (Raptor's) master keys. And looking at how Intel couldn't book new GPUs with TSMC all the way through 2021, I'm guessing no one is willing to book special/small scale 7nm fabs right now. After all, Global didn't gave up on this node for no reason.

fusing different keys into electronics isn't exactly a very hard thing to do and I don't think it is a foundry thing. Microcontrollers do it all the time, they are manufactured empty and if the key memory is empty or the one-time fuses are not programmed you can just program them from the microcontroller's software, or through dedicated chip programmer hardware.

[starshipeleven]

Raptor in general believes fused key systems to be dangerous on multiple levels

Yeah, the main issue is key revocation. Only complete and absolute morons design a security system under the assumption that no key will EVER be compromised, so you must have a way to change the keys.

Heck, even garbage like the HDCP (HDMI "antipiracy" extensions) has some form of key revocation in place, although without a central server it has little chances to actually do much.

Microcontrollers get a pass on this as you can easily replace the whole system with a new one when you need a new key.

[onlyLinuxLuvUBack]

Sounds a bit like openwashing.

We try to get a open reputation for the product but in fact, open just get the crumbs years later meaning that we let them fix the old issues and we focus on the new cool stuff.

Welcome to IBM ?
Their whole lets get rid of fishkill fab in NY, and lets get rid of thinkserver hardware and give to lenovo, and let's only make money by selling support/services/licenses. So their philo is lets make tons of money and everyone else can do all the work.

No skin in the game, they'll drop you like a fishkill fab....
I am lining up to buy.

[lorabe]

While i don't know why the raptor implementation will arrive later, I have my own theory.

Power 9 came when there was not OpenPOWER foundation but OpenPOWER org, a consortium conformed by IBM's partners and there wasn't an open sourced architecture.

With POWER 10 things changed, now there's an open source ISA and finally the first implementation which is closed as expected. Now what's going to happen with IBM's partners like raptor?

Probably there are new conditions to use the particular implementation designed by IBM, and this will make things harder for Raptor.

I might be wrong tbh but as far as I know IBM expect companies to use the new open source ISA to make their own OpenPOWER processors...

[edwaleni]

The OpenPOWER Summit is less than a month away. I would wait until their partners have their say after the embargo ends.

https://events.linuxfoundation.org/o...north-america/

[sophisticles]

Someone please explain to me what "open source" hardware actually means, or more accurately, what is the benefit of it? Unlike with open source software where you can take the source, modify it and then build a custom copy, once a piece of hardware is built you can't really modify it (outside of silly things like flashing the firmware or overclocking), I see no benefit of open source hardware.

[andyprough]

Someone please explain to me what "open source" hardware actually means, or more accurately, what is the benefit of it? Unlike with open source software where you can take the source, modify it and then build a custom copy, once a piece of hardware is built you can't really modify it (outside of silly things like flashing the firmware or overclocking), I see no benefit of open source hardware.

I'm assuming this is a joke post? If not, you may want to read up on a fellow named Edward Snowden.

[starshipeleven]

Someone please explain to me what "open source" hardware actually means, or more accurately, what is the benefit of it? Unlike with open source software where you can take the source, modify it and then build a custom copy, once a piece of hardware is built you can't really modify it (outside of silly things like flashing the firmware or overclocking), I see no benefit of open source hardware.

in this case it means the firmware and microcode needed to run it and the drivers needed to run OS on it are opensource.

It's actually quite a bit for security (you can fix any issue that is found regardless of what the manufacturer does) and means you can trust the hardware to a much bigger extent than x86 where you have like 30MB of firmware blobs between the OS and the hardware.

it's as close as we can realistically get to opensource hardware with high-complexity designs like this, since even if the actual hardware design was open, good luck at getting it manufactured in the low low volumes of less than a few millions.