Thoughts on Intel software-defined silicon

Some of the LWN comments suggest this is similar to previous/existing gating of features being fused off for example or only some cores being enabled. I think this looks very significantly worse for multiple reasons though. While clearly much of it is pure segmentation for pricing, fusing features/cores can have genuine yield justification too. If some silicon is simply off for good then it doesn't need to work let alone be stable, so if testing reveals defects that can be a reasonable way to salvage a chip. That the lower probability and thus lower count defect free chips would cost more made sense. And fusing also at least requires commitment: once done, the chip is permanently less valuable, which is at least some check on overly aggressive artificial monetization.

The perverse incentives here seem a ton worse. There is direct incentive to be very aggressive with it, because even if they get the balance wrong they can adjust it later and employ all the marketing stuff done with software (50% off sale on vector math acceleration 2 days only!). It'll by definition always be used with features that the chips are fully capable of and thus even further pushes towards tiering capabilities without any connection to yield. The centralized control requirements moving further into underlying hardware are concerning as well from a redundancy and security perspective. I'm a big believer in businesses following the incentive tracks they create long term, for better and for worse. With this development, Intel's are diverging further from their customers'.

About the best that can be said is that at least the market now is a lot more competitive. Between Intel's fab woes, TSMC, AMD and both Arm and burgeoning RISC-V efforts, Intel is more on than I think they've been in a very long time. Maybe ever? Which does make it all the more curious to me why they're trying this now of all times. Seems like 5-7 years ago they'd have really has us. Now though it's just another thing in strong competitors favor. The end of a lot of core x86-64 patents is starting to kick off too. Or maybe Intel thinks this will let them do sticker prices that are drastically lower and thus look better vs AMD et al even as they can still extract more money in the end? It's worrying. In fact now that I think more on it, Intel weakening but using this brings its own extra worries. If they feel the need (or are forced) to juice returns for shareholders and themselves down the road if their competitiveness falters, or even if they regain dominance, they will have an installed base with a capability to squeeze it.

Will this result in an illegal number situation once the keys are found?

Also, to steel man Intel’s position, reducing hardware skus can improve yield and increase quality (due to the known silicon lottery phenomenon), that is one abstract benefit to consumers.

This is basically taking the tiered SAAS and trying to manifest that business model ability into hardware.

I’m rather a fan of this sort of stuff but only with my middle finger firmly raised at the vendor. Electronic test equipment has been doing this for a long time. You can buy a base model item for $1000 and crack it to the $5000 part in thirty seconds.

It may be easy to dismiss Intel's move to SDSi by saying "just buy AMD instead," but if this model turns out to be lucrative enough, there's no guarantee that AMD wouldn't follow suit. We'll have to see how this plays out and how it will affect the x86 landscape.

lots of slippery-slope-type ideas come to mind, including time-limited subscriptions that weren’t mentioned in the article

i think that environmental/ recycling/ reusage argument can be a good basis for a campaign for a blanket ban against any such measures in any customer electronics. it seems like a very important battle

Haven't Intel tried this before?

A quick search hints at [1] circa 2010, but I think I recall a previous attempt circa 2004

I'm pretty sure it was unpopular and short-lived.

[1] https://en.wikipedia.org/wiki/Intel_Upgrade_Service

> an off-the-shelf processor might be unable to run anything more demanding than "hello world" until additional licenses have been purchased.

That's probably a bit extreme, but generally the right train of thought.

Sounds like what IBM is doing with mainframes - you always get a standard CPU module full of cores, and then pay for license to selectively unlock them. It’s been like this for… two decades now?

This could be good for consumers if implemented reasonably. If Intel can save money having just one product line, selling i7s at Celeron prices with cores just disabled in software, it could be good for everyone. Their method of disabling will probably take some time to crack, so people who buy these at Celeron prices can sell them as i7s, flooding the used market with good-performing CPUs.

We already see this with other hardware... Think Dell iDRACs and HP iLOs, which just need a few tricks to get all those licensed features once they're out-of-support and the vendor isn't keeping track of them as part of the support contract.

I'd also like to think this will lower the bar on upgrades during sales negotiations. HP, Dell and Lenovo are trying to get you to buy 100+ servers from them instead of their competitors, one of them might offer a free software upgrade for the CPUs as part of the deal, since it really doesn't cost anyone anything to do so... We've all seen deep discounts on software sales/licenses because their per-unit cost is nominally $0, now that model may be extended to hardware.

Of course the counter-point that comes to mind is IP-KVMs. If you've seen them being sold used for $20, that's because the hardware is useless without an ongoing subscription from the vendor. I suspect that only persists because IP-KVMs are too small of a market to attract the kind of effort needed to crack the mechanisms.

We'll have to wait and see.

CPU manufacturing is already impossible to compete with. This, I believe, will give Intel too much power. They will be able to sell processors at a loss if they can remove features after the fact (make the money back in subscription after the "free trial") and undercut any possible competitor out of the market.

For example, Intel could offer, say Zhaoxin's level of performance for free for a short time, making sure companies like Zhaoxin never make into the consumer market.

This is a great move for Intel but terrible for the consumers.

Intel has a better story for not being stomped on via process than they did a year ago, but that chip still isn’t like on AWS.

Maybe get your shit together on being eaten alive on one side by aarch64 and Threadripper on the other before inventing some new way that what used to be called a nasty zero day is now a “feature”?

> And, of course, there are dodgy web sites out there demanding payments for access to their content.

Nice self reflection (:

I’ll simply boycott Intel if they follow through with this. AMDs offerings have been far better value for some time now but Intel was always an option at a lower price point. Now it’s AMD or nothing, and so I think in fact this strategy gives AMD too much power.

This leaves a bad taste in my mouth, as it feels potentially quite predatory.

I think this will almost certainly be used to either kill second hand processor sales (by either burning the processor to a given motherboard), or to allow Intel to double dip post sale, much like EA’s “project ten dollar” did with used video games.

We're going to end up using vetted RISC-V FPGAs, aren't we? It could be that there will be a significant speed / security divide that will only grow; not a fun world to live in, but one which might become a necessity.

Welcome Intel's patches for their planned use of the facility. Then welcome anyone else's patches that allow unplanned use of the facility; and tell Intel to piss up a rope when they whine about that.

My only real concern here is whether the manufacturer may be able to unilaterally _limit_ already-enabled hardware capabilities in the future. If they can do so without your express consent _at the time that the capability is downgraded_, well then one can fairly say that the capability is not owned. That would be tragic.

As usual, consumer-level customers are here complaining about the things they don't understand. Do any of you run a large 1,000+ core cluster? If not, please spare me your "middle fingers" and "bad taste in your mouth". This isn't going to affect your $500 Chromebook.

Intel is behaving as a monopolist, but they do not have a monopoly. This move will fail spectacularly because no one will be forced to adopt the feature, and the engineering effort going into software defined silicon is mostly waste of human capital.

This is really the beginning of the end then. This is the same route that IBM went before it’s failure in the business market with vendor lock-in and other nonsense. I had hoped that intel could recover but they don’t seem to want to.

So I saw a comment once which suggested all intel cpu's are the same and they just disable certain features before its sold.

So this article would make sense when thinking about the above comment, only instead of features being factory fused or disabled, they become a payable option which might be a one time fee or an option which can be switched on for a period of time.

From a recycling point of view, it would make sense, people dont have to throw away a computer or shelve it when they want to upgrade, but I cant help but think what about the realisation over time people may come to if this instruction set on/off service/rental/purchase becomes an option? Would that then harm intel or not if it became known that and would that spawn some new hacking interest in microcode or whatever it is that controls the enabling of different instruction sets?

I'm interested in what the legal basis for this is. If someone found a way to circumvent the check (and presumably they will), what law would they be breaking by unlocking everything in sight?

Seeing only negative comments here, a bit of a counterpoint: I welcome this, if it means the sales price of the CPU will be significantly lower. This could be better than charging everyone a higher price, even for those who never touch the to-be-paywalled features.

Although it does depend on what kind of features are paywalled. I can see many ways how this could turn into a disaster for consumers.

I can't wait until people finds exploits to get those features for free

Fully-functioning silicon being wasted on paywalled features. When there's a semiconductor shortage. It's a greedy and wasteful idea in any day and age, but right now it comes off as tone-deaf, too.