28 ms·
Thoughts on Intel software-defined silicon
- benreesman 5y agoIntel 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”?
- MrYellowP 5y ago> 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.
- NickNameNick 5y agoHaven'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 https://en.wikipedia.org/wiki/Intel_Upgrade_Service
- pabs3 5y agoYeah, the LWN comments mention that Intel attempt and some other ones like the RPi video decoders.
- benreesman 5y agoI hate to be a hater, I grew up on Intel. But they’ve been doing increasingly desperate cash-grab stuff for years and years now. Oh great, let’s default all the ML stuff to the SIMD extension that doesn’t de-clock you. How did I ever live without it. I want them to win on some level, but this isn’t how to do that. If I had to I could tweak/fix/touch-up re-build all my stuff on ARM and life would go on.
- valyagolev 5y agolots 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
- silvercove 5y ago
- trasz 5y agoSounds 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?
- p_l 5y agoShorter, used to be that you still had different CPU books (still might be so) but the leasing model where it often makes sense for IBM to ship more hw than less is very popular with clients. Prime difference might be that from money point of view the way IBM handles it does end up in savings for the client (as it allows a pay-for-use/capacity rather than pay-for-hw model, and you can go 90% of the month in cheaper configuration and only pay for higher one for end of month/year ops). Another difference is that technically with IBM you're paying a license for microcode running on the CPUs, and you mix and match differently licensed versions to match your workload.
- classichasclass 5y agoIBM does this with their Power iron too. I had to get a Capacity on Demand (more like Crippled by Default) code to get the second CPU in my AIX POWER6 turned on. When the system planar had to be replaced, it had to be regenerated. What a pain. OpenPOWER and PowerNV systems running on the metal don't do this, but if there's an LPAR involved (AIX, i5), you could be on the hook. The moral of the story is to buy your Power gear from a company like Raptor unless you must run AIX.
- ComradePhil 5y agoCPU 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.
- Laforet 5y agoIt's not like they have not tried. At few years back in time, the market was flooded with cheap x86 tablets because Intel was pretty much handing out pallets of Bay Trail SoC for free to any willing manufacturer in an attempt to kick start an mobile ecosystem. Suffice to say it never got any traction.
- diegocg 5y agoI can't wait until people finds exploits to get those features for free
- h2odragon 5y agoWelcome 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.
- jesprenj 5y ago> And, of course, there are dodgy web sites out there demanding payments for access to their content. Nice self reflection (:
- 19h 5y agoArticle subscriber-exclusivity expires after some time on lwn and subscribers can share links that bypass paywall, seems pretty reasonable to me..
- xoa 5y agoSome 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.
- tomxor 5y ago> 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 [...] they will have an installed base with a capability to squeeze it. I think this is the most concerning possibility... people building servers or devs and gamers will be able to wise up to the shenanigans and then pick a competitor; but punters walking around a computer store are not going to appreciate the differences if Intel get away with making big claims on the sticker (possibly even spinning "software defined silicon" as a way for them to upgrade hardware over the wire) - for the average consumer this would making an already difficult to achieve informed comparisons hopeless. Reality may be different though, it's easy to assume the worst with the blue chips - perhaps this is just an optimisation on their current practices rather than a squeeze on customers, the concern is of course if they will exploit this new capability to the full potential we are all speculating on.
- hughrr 5y agoI’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.
- benreesman 5y agoI like your style, but if Intel has phabbed a run of 1000 chips that are worth un-breaking because someone got EUV moving I missed that news.
- garaetjjte 5y agoI think some vendors are doing it for purpose, there are trivial to use keygens available for years and they are doing nothing to stop it. That way they can increase value for hobbyists (who couldn't afford to pay for more than base model anyway, but they will pick it over competition which cannot be easily unlocked) but still earn more from big customers who have money and aren't willing to use keygens.
- snackypoo 5y agoI’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.
- 5e92cb50239222b 5y agoUnless they follow Intel's lead like that did with IME.
- xattt 5y agoAMD is working towards integrating the Pluton security processor.
- snackypoo 5y agoThat’s also a problem. I had high hopes for ‘free’ RISC-V alternatives while not really being a fan of the ISA, but Intel’s ‘foundary innovation’ chiplet program is totally unacceptable - because it keeps a company I simply don’t trust (ME) controlling the root of trust in silicon where the most desirable quality is the absence of exactly that. I may end up using Ampere chips in spite of a distaste of ARM. I’m also seriously pursuing an OpenSPARC based SoC, with a test design through Skywater. At the latest process node SPARC V9 is plenty competitive and there’s a clear market for a high core count SoC absent of additional IP. I just need a modern memory controller. I’m going to document the process at silicon.engineering although that site is just a placeholder now.
- pabs3 5y agoWhat OS are you going to run on these chips? ARM has reasonable support, but SPARC is declining, for eg Debian only has an unofficial port, so no stable release, no security upgrades.
- Tsiklon 5y agoThis 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.
- voakbasda 5y agoYup. Like Tesla, I am certain that extra features will not transfer during resale. If they already plan to lock us out of working hardware, you can be damn sure they plan to charge us for those same features every time that chip is resold. Seriously, does anyone believe otherwise? This is straight out of the rent-seeking playbook.
- vardump 5y agoSeeing 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.
- atq2119 5y agoAny feature that's a serious benefit for raw performance takes up serious area. At Intel's volumes, it's better to produce a different (and smaller) die for performance-based market segmentations. Expect this to be used for features that are comparatively cheap in terms of area but valuable for enterprise. Virtualization features, ECC, that sort of thing. This will slow the adoption of those features in software ecosystems since many developers still quite reasonably focus on stuff that works on their local machine.
- throwaway2048 5y agoYep, Intel clearly hasn't learned a single thing from the disaster that has been price segmentation on features like AVX-512, the result is the vast majority of software never bothering to use it, because it can't be guaranteed to actually be present.
- namibj 5y agoI'd argue exposing the IEEE 754 binary64 (aka f64; aka double) FMA unit's `((u52 * u52) & (1 << 51)) + u64` and `((u52 * u52) >> 52) + u64` (low half/high half, respectively) in the AVX-512IFMA instruction set was quite effective for the cost being comparatively cheap bypassing of exponent/sign handling and corresponding instruction decoder additions. Note: these help for wide multiply, as needed most prominent by ed25519 and RSA. They do apply for "just" 256bit vectors using 4 lanes of 64bit (where multiplicands are just the low 52bits), without the AVX-512-typical clock penalty. In similar spirit, it would have been nice to get emulated AVX-512 on AMD and Intel's E-cores, as it uses just one instruction to process twice the width, reducing the frontend throughout requirements. Also said AVX-512IFMA is rather nice, regardless of vector width.
- mensetmanusman 5y agoWill 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.
- pmorici 5y agoDepends on how they implement it. They will probably do it in a way where the license is linked to the processor’s serial number via cryptographic signature using public / private key pair so it can’t be installed on more than the part it is meant for. In that case, that number has no use except with that specific processor. This is different than how DRM content protection works because the nature of that problem is different.
- le-mark 5y ago> This is basically taking the tiered SAAS and trying to manifest that business model ability into hardware. IBM did it for years with mainframe pricing, they had excess mips installed and allowed burstable performance setups for a fee, on hardware you “owned”. Amazon took this strategy over, albeit without the ownership sham. Now Intel wants a slice. It’s turtles (ie greedy MBAs) all the way down.
- protomyth 5y agoIBM also did it with their POWER AIX boxes. The big difference was those instances were Business To Business contracts and not consumers. I get the feeling the consumer market will react differently to Intel.
- chaosite 5y agoThe threat in business to business contracts, in addition to "we'll sue you" for licensing violations, is "that's an unsupported configuration, you're on your own."
- 5y ago
- ris 5y agoI'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?
- coobird 5y agoIt 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.
- voakbasda 5y agoI cannot imagine this bet paying off for Intel, because I will be choosing AMD in all of the workstations for my engineering group. Beyond my own group, I will use my influence to be sure my entire organization avoids them as much as possible. If anything, I expect this will hurt them where it counts. IT folk will not be amused and find other places to spend their money. The decision to lock up working silicon is radioactive and will have long lasting consequences on buying. I like to imagine this sort of decision being made across the world, as other engineers learn of these anti-consumer shenanigans and make waves in their own organizations.
- ac29 5y ago> I will be choosing AMD in all of the workstations for my engineering group Too bad AMD has largely ignored the workstation market. The last threadripper release was mid 2020 and is a Zen 2 product. Newer Zen3 based chips have been credibly rumoured, but AMD themselves havent said anything about it.
- IYasha 5y agoStill using DRM-less FX and Opteron machines today. Hot but no surprises ;)
- theevilsharpie 5y ago> Too bad AMD has largely ignored the workstation market. The last threadripper release was mid 2020 and is a Zen 2 product. Intel has also been ignoring the workstation market (although perhaps not by choice). Despite the current-model Threadripper being nearly two-years old and last generation technology, it is still far and away the performance leader in that segment in terms of perf/watt, perf/dollar, and oftentimes outright performance.
- flenserboy 5y agoWe'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.
- ryukoposting 5y agoFully-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.
- notriddle 5y agoThe semiconductor shortage is caused by shipping bottlenecks, not raw materials.
- colejohnson66 5y agoCPU binning (with fuses) already wastes “good” silicon. An i3 is just an i7 that wasn’t as good and had some features disabled. See also: the “silicon lottery” for more info. It’s been this way since the beginning. All this is doing is replacing unrepairable fuses with ones that can be disabled after the fact. Theoretically, this is just giving you access to the features that would’ve been permanently fused off, but still good. Is it scummy? Yes. But is it wasting any more silicon than before? No.
- Terry_Roll 5y agoSo 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?
- BeetleB 5y ago> So I saw a comment once which suggested all intel cpu's are the same and they just disable certain features before its sold. Really not true. What happens is that during manufacturing of a particular design, you get a distribution - not all units will have all features working. They analyze the data, and partition the units into segments based on (common) features that do work. Those on the tail end of the distribution (on the positive side) are the most featureful and sold with a premium. And so on from there. If they actually tried to manufacture different CPUs for high and low end features, then each of those would have their own distributions, and they'd waste a lot of units for not meeting the spec. This way they can simplify the production and not waste the units.
- togs 5y agoOh interesting, I didn’t know this. So an i3 is just partially botched i7
- BeetleB 5y agoI believe so. Not saying they may not also do what the GP said, but from what I've heard most of it is my explanation. I think his comment is more applicable to the data center chips they make, and less for the consumer chips.
- indymike 5y agoIntel 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.
- axiolite 5y agoThis 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.
- burnte 5y agoThis won't save them money, it'll just extract more from us. It'll make upgrades easier, but at the expense of witholding things from the start. They'll still bin CPUs so they sell the top performers at top prices.
- mlindner 5y agoThis 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.
- ponow 5y agoMy 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.
- phendrenad2 5y agoAs 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.