Hacker News Reader

Comment by johndoe0815 | Hacker News Reader

Calling this "Software Defined Silicon" instead of what it is - rent-seeking - is audacious...

Intel can make parts to fit a SKU by (a) sorting them based on what elements are working when they come off the proverbial press, (b) physically disabling working elements, or (c) disabling working elements in software with the option to re-enable them.

I totally understand the argument that (c) is more objectionable since the chip you buy is prevented from using all the elements it left the factory with. However I think it really shouldn't feel different than (b)? The only difference between getting 8 core part with 8 more cores burned out on its way out the door and an 8 core part with 8 more cores disabled in microcode is that you can double the core count of the latter in place for a price that's presumably proportional to the price difference between the 8c and 16c SKU?

Of course there's a slippery slope to a hardware-as-a-service pay-us-x-per-unit-time-or-we-turn-off-your-CPU, but with competition from AMD on x86 and in-house ARM chips from Apple and hyper-scalers, one would think Intel wouldn't want to make their product even _less_ attractive.

I think the issue is that the economy is supposed to in theory serve people, which most people would think means that if there is no material difference in effort to produce a part, the best most useful technology should be produced and in consumers hands. The theory of market competition says that moves like this should be impossible because it would open Intel up to being undercut by a competitor nearly instantly. However, we can see that monopoly capitalism does not function the way the legitimizing theory does. In fact, it is populated with rent seekers that manipulate the public using market segmentation strategies to profit.

Price discrimination happens all the time even when there's no material difference in providing the good/service (e.g. Uber surge pricing).

It only happens when there is not robust competition (even 5-6 big players is not enough), which is the case in pretty much every non-commodity market today.

I have seen price discrimination/segmentation at hotels even though there are tons of hotels (senior citizen discounts, military discounts, AAA discounts, etc), and grocery stores and restaurants price discriminate heavily via coupons.

At least surge pricing makes sense - when there are more fares than drivers, the fares have to pay more to be picked up first.

With CPUs, it's less intuitive - The people with DRM-hamstrung CPUs are supposed to be getting a cheaper price subsidized by the people buying the product at full price. The cheaper models otherwise would have to be sold at the full price.

That's not to say that Intel's prices are necessarily fair at any tier...

Surge pricing of a limited resource is not exactly "price discrimination" in the way that term is used traditionally in economics. Per wikipedia: "Price discrimination is a microeconomic pricing strategy where identical or largely similar goods or services are sold at different prices..."

In the case of surge pricing, there is just not enough goods or services to sell, so you simply dynamically increase the price in response to demand akin to an auction, not selling it more expensively to a subset of customers simply because they can pay and to others for lower prices (which presumes a supply surplus exists). In other words, an "Uber ride at rush hour" is not a substantially similar product to "Uber ride in off-peak time".

Apply this to software, and everybody will be forced to pay the Office Pro price when they only need Office Home features.

For a certain price, you can buy an Intel CPU with certain known features. For a higher price, you get more features.

This is a concept that’s used in many product classes. The fact that it’s also done for HW is just an irrelevant detail.

> Apply this to software, and everybody will be forced to pay the Office Pro price when they only need Office Home features.

Or rather Microsoft makes less money producing pretty much the same products and selling close to the lower price, which is more likely, since their margin is high enough to allow it.

The only reason this works is that it's backed up by government enforcement of intellectual property rights. If the government didn't do that, then consumers wouldn't pay anything for office software, Microsoft Office wouldn't exist, and the whole landscape would be very different. The natural price for the second copy of a piece of software is $0.

Something being known and taught in product classes doesn't make it any more acceptable. Planned obsolescence is taught in economics classes, and it's still awful despite that.

> Microsoft Office wouldn't exist

I believe you're wrong. It would absolutely exist, either in cloud form, or as a rented screen-time service before that.

Productivity (and other) software has a real economic value, it's not a movie you can skip and lose nothing.

Similarly, if my aunt had wheels, she’d be a bike.

I don’t see the point in discussing lala land hypotheticals.

Is the marginal cost of providing a copy of software really $0, or does the marginal user ask questions, have feature requests, uncover bugs with new usage patterns, ...? Does that non-zero marginal cost vary with the features included in the software? Isn't support a major part of most software product differentiation strategies?

That's true, and that's continuous labor. However, it's also possible for software to come with no support too (e.g. abandoned games). As software developers, we care that our labor is paid for, but there's no reason there should be some kind of eternal dividend for doing nothing. Nor should there be for bushiness people generally.

Which is why intellectual property rights are time-limited? One can make a strong argument those rights are protected for _too much_ time, and the length of that restriction can be attributed to regulatory capture. But the remedy is shorter copyright terms and patent exclusivity periods, or higher costs to maintain protections (like property taxes on real property), not abolition of intellectual property as a concept.

Given that most are pushed into the yearly SaaS subscription instead of the more expensive one time Home license, that is already here, no need for future tense.

> The theory of market competition says that moves like this should be impossible because it would open Intel up to being undercut by a competitor nearly instantly. However, we can see that monopoly capitalism does not function the way the legitimizing theory does.

Can you quote single mainstream economist claiming that market entry barriers don't exist?

Mainstream economics argues that the market tends toward equilibrium, not that it equilibrates instantly, and that the structure of specific markets controls the speed of that process. Modern semiconductors are fantastically capital intensive, and so respond slowly.

An Intel leveraging their position to under-invest and earn outsize gains from their leadership position in x86 ... would lose business to competitors (like they have been by AMD), innovators (like they are to Nvidia) and substitutes (like they have to Apple Silicon).

> Modern semiconductors are fantastically capital intensive, and so respond slowly.

That's my point and de-legitimizes the economic model that grants them private control.

There's a substantial leap here from 'Intel might be temporarily able to / trying to exercise a degree of monopolistic pricing power' to 'market capitalism is illegitimate'. The response to a firm with mono{pol,pson}istic pricing power is to encourage competition, require cooperation, or force divestment.

I think the latter two cases are different because it ultimately comes down to property rights; (b) may be harder to reverse but if you can do it, you still own it (remember the AMD pencil mod? https://www.overclockersclub.com/guides/tbird/) (c) is, being a software decision and thus tied to licensing and such, feels more like "you can't use the hardware because we said so", and that arguably is what people hate. It's a slippery slope in the same direction as the "you will own nothing" movement.

...and what makes you think AMD and others are not going to try the same thing?

isn't ARM a commodity market? seems to require a lot of coordination from the entire industry.

>is, being a software decision and thus tied to licensing and such, feels more like "you can't use the hardware because we said so", and that arguably is what people hate. It's a slippery slope in the same direction as the "you will own nothing" movement.

SaaS offerings doesn't seem to gather the same type of hate though? When was the last time people were outraged that the "basic" tier lacked features in the "premium" tier, even though the code for implementing such feature is literally on the same server and gated behind an if statement?

With SaaS, it's pretty clear that you don't own anything.

> ...and what makes you think AMD and others are not going to try the same thing?

Overwhelming competitive pressure? AMD and Intel may be a duopoly on x86 parts, but Apple Silicon has proved, and AWS Graviton has a strong argument, that consumer and cloud compute workloads can transition to ARM. Any attempt by Intel to leverage its market position just accelerates that transition to substitutes.

Ignoring the broader economic arguments that paradoxically justify this as productive, (b) seems worse because it’s literally destructive. They’re producing something of value and then deliberately damaging it.

Is creating something of value, then prohibiting someone from using it any better though?

FWIW, I find both b and c objectionable.

It may feel antithetical to the very core of your being, but that's where engineering meets capitalism. Intel is a business (because it has to be), and businesses want to maximize profits, and where b and c aren't materially different, taking something and making it worse is the same failing of capitalism as communism - human nature.

I can make 10 widgets for 10 customers to buy, but between those 10 customers, 2 of them want the cheapest version, 6 of them want the medium option, and the last 2 want the most expensive option. The way to maximize profit is to produce that same distribution, divorced from the actual yield of the production process. Pricing a product is a dark art but looking a bit closer at it, there's a huge psychological or human nature element to it.

but with competition from AMD on x86 and in-house ARM chips from Apple

Competition only works, if people compete on that market point.

This can be a juicy revenue stream. If intel gets little blowback, expect AMD and others to salivate, then jump on board.

You raise an interesting point: Should Intel be allowed to destroy value it produced, in an effort to make more money?

Arguments could be made that it's ecologically criminal to destroy useful things for profit.

This is an interesting take. I think it's clearly unethical to destroy production as a price support for human essentials. Industry shouldn't incinerate insulin, dump milk[1], trash food, etc. in pursuit of current or future profit. I have a hard time extending that argument to processors. Would it be nice for Intel to pass unexpectedly high yields on to society in the form of better parts or prices? Sure, but I have a hard time finding an ethical obligation for them to do so.

[1] of course in choosing this example, we admit that _government_ might ethically choose to destroy production to maintain price floors so that domestically-produced products remain available.

As a sibling comment notes, Intel can (and has a history) of doing this physically. Where a manufacturer price discriminates by offering two products, a normal one and an inferior one, it may be the case that the cheapest way to do that is by simply making twice as many “good” ones and intentionally disabling half of them.

There is a very famous economics paper by Deneckere and McAfee that outlines exactly this behaviour from Intel in the 386 (…?) days, and how it may even lead to Pareto improvement - and increased consumer welfare. I can’t find it at the moment on mobile, but it should be relatively easily surfaced.

Edit: found it, “Damaged Goods” - https://www2.econ.iastate.edu/faculty/langinier/teaching2005...

I thought the normal and inferior products lines are usually the result of binning, not of intentional production. In which case, it's a way to make good on products that would not normally sell under your brand and would otherwise be disposed of.

If your manufacturing efficiency has advanced to the point that you no longer have low quality parts coming out of your binning process, then I think customers are right to wonder why that savings and efficiency aren't being passed on or being used to further advance the product line.

In any case, if the purpose of passing the inferior products off under an alternative label is to avoid brand damage, then this seems to be the opposite of that.

In many cases yes, it is the result of binning, but the 486 SX (my first comment incorrectly said 386) was an intentionally disabled 486 DX. From memory, you could then purchase a 487, which was actually just a full-fat 486 DX to slot into your co-processor slot as an 'upgrade'!

My understanding is that binning is typical at the beginning of the product line, but usually process tweaks increase the yield as the product continues. At which point things become complex.

If you no longer have N core chips because all of them test good at more cores, but don't want to sell the N+1 core chips at the old N core price, you've raised prices with all that entails.

Your last point seems sound, but it strikes me that this model might only possible since there is a limited number of competitors in the market. I would expect more competition to drive the prices down while still increasing product core counts.

A supplier of voltage regulators sold the same (physically and functionally interchangeable) part to Ford and to Cessna. Cessna required 100% testing of the parts. Ford accepted statistical process controls.

Solution: make the parts on one line, inspect the greater number of how many they’d need to inspect for Ford and Cessna, stamp all the inspected parts and sell any part to Ford and only stamped parts to Cessna. All three companies can benefit from that arrangement.

Yay! I had a Cessna voltage regulator fail after over 30 years. So i would say this system works. :)

Well, “works” in the sense that your Cessna is (and mine was) flying around the very best inspected of 1960s technology…

It's often done for binning, but you also get working hardware that has been disabled. Many AMD CPUs have been unlockable with a reasonable success rate[1], and IIRC all of certain models of X800 GPUs could be upgraded from 12 to 16 pixel pipelines (which from memory was sold as the X800XL variant).

I can see how in theory it might make sense for market segregation. Say you're designing a 2-core and a 4-core CPU - the design simplifications and manufacturing efficiencies of just making one chip (and partly disabling some of them) _might_ drive the price down for both market segments compared to manufacturing two designs, even if on the surface the 2-core user seems to be buying more hardware than they are given access to.

With that in mind, SDS doesn't seem like such a bad idea. Buying an n-core chip that can later be upgraded to a 2n-core chip without buying a completely new part seems preferable in some ways to buying an n-core that is actually a 2n-core chip that has had half its cores laser-sliced off (and thus can never reasonably be reactivated) to fit a market need.

[1] https://www.cpu-world.com/info/AMD/Unlocking_cores_and_L3.ht...

The thing is with those disabled aspects of AMD gpus, AMD is not trying to sell it to you later.

Also it could have been disabled because it was faulty when testing so binned differently or just to meet unit numbers for a SKU. Thing is you don't know.

Here intel is selling you a pice of hardware but by telling you can pay later to unlock it. They are telling you it this extra functionality works. The thing is once intel sells a device they no longer own it. So what rubs a lot people wrongly here is that for something you own your now having to ask intel to use some aspect of it. For things you own you should not be asking 3rd parties for permission to use your property.

If you are asking a 3rd party for permission do something/use something “you own”. This implies the seller has managed to retain some aspects of the right of exclusion. The issue here is the sell of tangible property normally involves the transfer of all rights. If the seller wants to retain some rights normally a contract is needed.

Now with this situation intel using a couple things to basically retain this right of exclusion. First cryptography since you don’t have the private keys anything controlled by those keys intel effectively controls. Second is software copyright. A type of intangible property were the right to copy/reproduce is by default legally retained unlike other forms of property. So any software tools to enable this additionally functionality intel can keep a tighter grasp legally on its reproduction.

Anyways the short and sweet if your having to ask the seller for permission after the sell to do something with something you “own” it implies the seller is some how retaining some property rights. This is going to rub people wrongly for tons of reasons.

There's also just straight up demand for worse products at a lower price. To wit: many luxury goods 'outlets' now sell lines produced exclusively for the outlet market rather than unsold merchandise from the primary retail channel. If there isn't product in the outlet stores, that segment's demand is met by a competitor.

Of course, there's an argument that this is just to protect the exclusivity of the primary retail channel, and that increased prices in that channel cover any losses from destroying, rather than down-binning unsold merchandise. The same argument might also apply to HEDT processors :)

This is a great reference, thanks for digging it up!

Our customers all get the same executable file. Yet if they've paid for it, they get more features as the license will unlock modules.

The code is all there for everyone, compiled and ready in case the customer decides to upgrade their license.

Is what we're doing "rent-seeking" as well?

Of course we could make differentiated builds, just add in a bunch of ifdefs and provide a new executable to the user when they upgrade their license. Would that be any different?

The result would be the same, the user would still pay more to get access to the additional functionality.

The cost of copying your executable is near-zero. The cost of producing a CPU die is non-zero. Suppose that ECC support takes a certain die area, then either everyone is already paying for it while buying the CPU or everyone who licenses the feature is paying for the silicon for everyone who doesn't enable it.

So, if you are licensing a CPU feature, you may be paying twice for it. Of course, whether you actually do so, depends on the relative cost of producing that area of the die vs. the development cost of that functionality.

Another issue is that this probably would not happen in a healthy, competitive market. In fact, the opposite would happen, CPU vendors would try to provide more features at a price point in order to compete (which is probably why AMD does support ECC on many Ryzen SKUs, as an underdog AMD still has to compete). However, if you need a x86_64 machine, you only have two options. Or if you are unlucky and required to purchase from a large vendor like Dell, you may only have one option.

> The cost of producing a CPU die is non-zero.

While strictly true, the actual die is a relatively tiny cost when producing a CPU. The majority of the cost is the upfront cost of designing the masks. Once you have those, using this or that mask doesn't make a huge difference.

> Suppose that ECC support takes a certain die area, then either everyone is already paying for it while buying the CPU or everyone who licenses the feature is paying for the silicon for everyone who doesn't enable it.

As mentioned the majority of the cost of the ECC support is in the R&D, not the physical manufacturing. As such, moving this to a software level can be a win-win for both parties. By selling ECC support to more customers, they can afford to price the bare CPU cheaper.

That's exactly what happens with our software. There's no way we could sell to single-person companies at the price we do without having features behind "license walls".

> Another issue is that this probably would not happen in a healthy, competitive market. In fact, the opposite would happen [...]

One of the reasons people love cloud computing these days is that the cost scales with activity. Get more customers generating more activity? No problem, just add some extra capacity from your cloud provider. Need some additional functionality to service a new market? No need to pay for years for something you don't use until you need it.

Wait until you realize that the entire SaaS industry and the subscription model is based on that same concept you're alluding to. I can understand service (say a barber) that needs to have a subscription model. But, most products today such as Adobe Photoshop should just be a one-time fee. Yet, we have come to accept that everything is $/month. You own nothing.

The most eggregious of these is browser-based apps. You have zero control over upgrades and whether you want features or not. With a desktop app, atleast you can ignore updates.

In practice Adobe Photoshop was effectively a subscription long before the pivot to SaaS. If you used it professionally, the old versions couldn’t open files from the new versions, so if you were part of a creative pipeline that you didn’t fully control, you got pushed into upgrading within a few months every time a new version came out.

The switch to SaaS didn’t really end up costing it’s pro users much more since they were spending $600 every 2-3 years on it anyway. But I suspect it did convert at least some of the people who pirated a copy from work into paying customers.

Yeah, SolidWorks also does this. If you get a file from a later version from a customer, you cannot open it on the older system even if it is just a year old. Entire machine shops get sucked into this - they already run on thin margins and it is probably one of the most expensive aspects of running a shop.

This is where the JetBrains pricing model is exactly right. If they sold licenses, people would upgrade every few years, after some needling^W marketing. Instead, they sell as a subscription with the option to cancel and keep only the features available as of your last renewal. So someone that would upgrade every release pays about the same average cost and gets a more continuous flow of new features, someone that would buy once can just buy once, and everyone gets the optionality to stop paying and 'keep' what they've already paid for.

It's hard to have this discussion until we know what the features are.

If it's hardware that implements a (non intel) patent and you have to pay to use it, but not everyone would want the feature, seems reasonable to let those who want it pay for it. Like MPEG-2 support on (early?) raspberry pis.

If it's something frivalous, it's frivalous.

A main use is probably core count.

AKA sell parts for a lower cost with a low core count, and when people need more performance, let them turn the additional cores on. IBM has had this model on the mainframe for quite some time.

Intel gets the benefits of fewer valuable physical SKUs to manage inventory of at distributors... and of very high margin revenue later.

IIRC Intel did almost same thing at around 2011.

I mean... that feels scummy, but on the other hand, when they've got a mask that has N cores and the yield is good, they fuse off working cores to get the lower core SKUs the market wants. There may be many customers that the IBM model works for.

It used to be you could do some fiddling around and get disabled cores undisabled to see if they were disabled for good reasons or not...

Price discrimination would also be an apt description.

IBM was famously past master at this: they’d increase your clock rate temporarily, at a price.

They still offer stuff like this in the mainframe machines, right? I recall that their z machines have full duplicate processor drawers idle and locked out so that if one dies, IBM can remotely activate the idle one to maintain service?

I for one will have no compunction about:

a) abandoning CPU vendors that do this, forever if needs be b) pirating features in my CPU

Why? It's the exact same thing as binning, only that now the binning is done in software.

I would have much more of a problem with this if Intel was still the only game in town. I get the feeling Intel is effectively ceding parts of the CPU market to ARM and focusing on their high-margin products.

It's actually, "functional pricing."

How exactly is this rent seeking?

Reply on news.ycombinator.com