Intel Confirms "On Demand" Upgrades With Sapphire Rapids (Software Defined Silicon)

[Ironmask]

I so want a hacker to find a way to unlock this shit, teaching Intel a lesson to never pull this crap again...

Unfortunately, Intel is probably well aware about this type of scene (then again, Intel being aware of their own security holes is a bit ironic). This is for chips that are solely for the datacenter market. That is, you're not gonna have hackers bypassing these limitations, these chips are going to be owned ("owned") by corporations who are legally obliged to adhere to Intel's ToS or else they can get sued.

Linux really should have rejected these patches, this isn't even immoral, it's just downright backwards. It reminds me of the old days when IBM used to sell mainframes and would put usage counters on their tape readers so they could bill the corporations/governments using them like how a power company would check your meter to bill you. Except IBM had a legitimate case there given how prohibitively expensive the hardware was and the fact that it would break down so often that it was honestly beneficial that IBM owned it given how often they'd have to maintenance it. But this, this is just pure greed. Intel doesn't have to own anything, it's a chip the size of a postcard. They have nothing to maintenance, it's all solid-state technology.

It's not immediately a bad deal. Again, this is for datacenters. These bills are for giant corporations to absorb. On the surface, that's not an issue at all. Honestly, they deserve to be taken advantage of. The real issue is the matter of Intel seeing success in this venture and then seeking out civilian markets to employ this tactic with. And thats why Linux should have rejected these patches and told Intel to try shilling their DRM somewhere else.

[NeoMorpheus]

I so want a hacker to find a way to unlock this shit, teaching Intel a lesson to never pull this crap again...

No need, simply dont give Intel any money.

But somehow, these days, we no longer consider ourselves customers, instead, we are corporation worshipers that no matter how bad the corporation treat us, we still want to give them money.

[Nuc!eoN]

Same thing Tesla does with their full-self driving (and IIRC at one time they also did this with a certain pair of models, where they had the same sized physical battery but the lesser model was software locked to be a smaller sized battery). Hopefully this isn't the new trend in hardware pricing, I hate how everything is going to a subscription model, and this is even worse than that.

Interesting. AFAIK teslas run on free energy internally and their "battery charge" is just artificial. Gotta pay for something right?
So what you are saying would make sense in that regard.

[ryao]

You mean like how the industry learned not to use DRM when DVD encryption got hacked? No, the only thing industry will learn is that they need to try harder, next time.

To recap some previous discussion threads about this, some of the dangers of this approach are:

• Flash has a limited lifetime and maybe there are circumstances where the hardware keys could become corrupted before the end of the CPU's usable life.
• Intel could transition to a "leasing" model, where you have to pay an ongoing subscription fee for your cores/capabilities. This also means they can increase subscription fees as/when they choose, basically turning you into a hostage.
• Hackers could use this capability to cripple a host (i.e. by erasing or scrambling installed keys), if they can find a set of exploits which enable them to tamper with the keys from within a guest VM (or any other way they manage to gain access).
• Even if you buy an Intel SKU that's factory-locked to a certain feature set, they'll quite likely use the same underlying mechanism. That means it could still be vulnerable to points 1 & 3, above.

It was probably inevitable, but that doesn't mean we have to welcome it. Due to the last point, we should prefer to buy CPUs from manufacturers which don't employ such practices at all.

They could go with fuses that permanently activate the functionality once they are blown. All your concerns would be handled by that.

I have not checked the public details, but are you sure that they used flash?

[ryao]

I'm reminded of Stallman's stance on firmware. If it's upgradable, then it's software and you should demand open-source. If it's locked into a specific version/configuration at the factory, then it's hardware.

It is possible to hardcode the logic into transistors and make the upgrades occur when fuses are blown. Then there would be no software at all, yet it would be upgradable with the right key.

[ryao]

so conflicted about this, On one hand it's pretty scammy, on the otherhand, a lot of stuff that is already shipped is just hard locked instead of soft locked anyways

I recall IBM using this business model around 2015 give or take a year. They would ship more hardware than you purchased and then charge you for a software unlock. I am not sure if they still do it. I have not paid attention to what they are doing.

[ryao]

Most servers are bought by companies that wouldn't apply any cracks even if they had one, anyway. The same way they don't pirate software even though they could if they wanted to. I don't think it would hurt Intel nearly as much as the OP thinks.

You would probably have OSS developers doing it on their own personal hardware to save money. The principal of first sale comes to mind.

Of course, that assumes that there are any outside of Intel still using Intel hardware given the nonsense product segmentation Intel does with ECC support.

[coder]

I'm reminded of Stallman's stance on firmware. If it's upgradable, then it's software and you should demand open-source. If it's locked into a specific version/configuration at the factory, then it's hardware.

That's not sufficiently nuanced. If it's stored in flash, then it's vulnerable to decay and exploits which can still manage to modify it. Hiding it from the user then becomes harmful.

So, "locked into a specific version/configuration at the factory" really needs to be: "truly ROM-based". Except, I'll bet virtually nothing is ROM-based, any more.

[coder]

Getting the end user to pay is dumb.... What they should do is charge software developers to access stuff.

Imagine if Nvidia locked half of their GPU cores unless a game's developer pays for an unlock signature for the game. All that money would add up much more!

1. Games publishers would simply pass the price on to consumers. Some might offer "premium quality" add-ons, to give you more of your GPU. The main difference is that the games publisher will sell you your GPU capacity after taking extra profit for themselves. So, it'll end up being even more expensive.
2. Would hurt independent game developers, who have less clout to negotiate GPU pricing than big publishers.
3. Would make Nvidia into bridge trolls, like how people complained that ISPs are, when they can charge content providers for access to end users.
4. It limits the end user from getting maximum possible performance from their hardware (which, even if the base price was low, they still needed adequate cooling & power supply to support), on some games.
5. Would probably reduce the amount of time you can still buy/play older games.
6. Could have a chilling effect on the modding capabilities, in some games.

I could probably come up with a few more "negatives", but I think that's enough.

Basically, any time you take away end-user choice and put it in the hands of an intermediary, there are downsides.

[coder]

This is for chips that are solely for the datacenter market. That is, you're not gonna have hackers bypassing these limitations, these chips are going to be owned ("owned") by corporations who are legally obliged to adhere to Intel's ToS or else they can get sued.

Do we actually know whether this will be limited strictly to "datacenter" chips or could it affect server CPUs, more broadly? You certainly have a variety of businesses with "on prem" servers. These servers eventually tend to find their way into people's "homelabs".

I'm also imagining a manager facing budget constraints searching for hacks to avoid paying Intel. Or maybe it's an employee that does it behind their manager's back. When licensed software was mostly on the "honor" system, companies weren't so honorable about paying for all the software licenses they actually used.

Linux really should have rejected these patches, this isn't even immoral, it's just downright backwards.

I think there would have to be specific terms in the charter which forbid such functionality. You don't want a situation where these decisions are being made on a case-by-case basis, because that opens up the process to accusations of favoritism and the whole ecosystem can factionalize very quickly.

But this, this is just pure greed. Intel doesn't have to own anything, it's a chip the size of a postcard. They have nothing to maintenance, it's all solid-state technology.

But you expect them to develop drivers, compiler support, kernel support, and firmware updates? Yes, those are mostly front-loaded activities, but a lot of work (i.e. expense) goes into patches for functional errors & exploits. And that's called "maintenance", even if it doesn't involve on-site support.

The point about size is also laughable, as if you want us to go back to the days of room-sized computers.

It's not immediately a bad deal. Again, this is for datacenters.

Regardless of where it starts, this is the camel's nose under the tent.

A lot of datacenter hardware has a second life, if mainly in poorer countries. Shutting off that channel (or squeezing it to a trickle) could have consequences difficult to foresee.

Honestly, they deserve to be taken advantage of.

They'll pass these costs on to their customers. Anything which takes control away from a user tips the power balance in favor of the vendor. Even if you aren't the direct customer, it makes the system less efficient, and that translates into pricing increases for end customers of these businesses.