The website has a clear and concise explanation of "Intel On Demand":
> What Is Intel On Demand?
> Align your infrastructure investment with your business demands with Intel’s API-enabled business offerings that deliver flexible consumption and configurations.
Or the other way, "Features that were previously fused off due to market differentiation but the same die otherwise can now be accessed without throwing away the current product and re-buying different SKUs (e.g. server/workstation Xeon chips)"
How "Acceptable" this is really depends on what features are going to be behind this paywall.
Is it really? If it's not retroactive, and you knew it going in and perhaps got a discount for it at purchase time is it really "you already brought". It's more like instead of destructively fusing off capabilities of functional products (ie irrecoverably ruining perfectly good hardware to enforce market segmentation), they make it easy to upgrade later. This isn't really that bad an idea.
As long as I can buy it fully unlocked from the get-go and am never forced into a subscription (as opposed to one-off) on my home workstation I really don't mind.
I had the exact same thought. They clearly have no intention of answering the question. They clearly don't want the reader to have any idea what it is. Why even include the question?
Confusion and fear of dwelling in a lower intelligence bracket (because maybe smart people understood the answer) is a simple way to hand-wave the problem away. You don’t understand? Perhaps you can’t!
Only a few will call bullshit. Likely on niche sites on the internet. Probably not in such a dramatic way that the masses take note.
There might be a funny meme image gif.
The question is asked so that an answer may be provided, so that the question has been answered. That avoids anyone getting surprised by the question and giving the wrong answer. The right answer doesn’t have to mean… anything.
Delay, deny, distract. Until the avoidable is inevitable. Then we are all to blame, not just some poor corporation that meant well. After all, they told us what they were doing and we didn’t object!
This happens in other markets just fine, one that springs to mind is oscilloscopes. Often you will find that a the hardware in one model is the same as the hardware in scopes in the same "family", what differs is software / locks set. This allows the manufacturer save on costs during manufacture but place the R&D cost (and profit from that R&D) onto the customers requiring those needed features.
Its quite common to buy a lower end scope and unlock some extra sample size, seom extra bandwidth, some extra decoders by following a guide found on the net. (My personal theory as to why manufacturers don't lock these down more is because it gets home lab engineers more inclined to buy a brand they know they can hack and get the best bang for their buck in the hopes that will help drive sales when it comes to their companies requiring new lab gear).
If I were to give Intel the benefit of the doubt, They currently have a sizable SKU count in any generation of processor that are basically the same silicon under the hood just binned/hardware locked differently based on any manufacturing defeats and customer demand. If the locks already exist (be it hardware locks atm) why not move those locks to software and allow customers to unlock extra features as required?
As long as Intel don't a) convert to a "pay as you go"/subscription model or b) pump up the base price to gouge customers on common features I'm not going to loose any sleep over it, however but it wouldn't be the first time a company has fucked over customers.
> This happens in other markets just fine, one that springs to mind is oscilloscopes. Often you will find that a the hardware in one model is the same as the hardware in scopes in the same "family", what differs is software / locks set.
There exist quite some makers who solve this "oscilloscope tiering" by installing a hacked or alternative open source firmware on the oscilloscope.
My personal fav (while not a scope) is the Flir hacking years ago (Not sure if they have locked them down since). Things like converting a Flir E4 to an E8 via software alone (and vastly improving the resolution) https://www.eevblog.com/forum/thermal-imaging/flir-e4-therma...
I'm not going to disagree I mean after all Bill gates is quoted as saying back in 98
> Although 3 million computers get sold each year in China, people don’t pay for our software. Someday they will, though, and as long as they are going to steal it, we want them to steal ours. They’ll get sort of addicted, and then we’ll somehow figure out how to collect sometime in the next decade.
Though by the time XP came around MS was no longer happy with the casual copier although VL still made it pretty dam easy to pirate windows. I still enjoy listening to Dave Plummers talk about windows product activation - https://www.youtube.com/watch?v=FpKNFCFABp0
I don't get the outrage. Normally, these features would just be fused off during the CPU binning phase. Even working areas of the chip are fused off for market segmentation. If you have a 4 core chip, it's probably a 6 or 8 core internally.
All this seems to do is make the fused off area, that was previously unusable, now usable. For the average consumer, it doesn't change anything. But now a business can purchase a CPU that doesn't support, say, AVX-512, and enable it on-demand when they need it. Is there a reason we should be more upset about this than the CPU binning that's been happening for decades?
Side-note: Intel's market segmentation on ECC memory is, however, garbage.
As long as this is a one time fee to unlock each subcomponent, it is fine, actually even better because it means less waste and more flexibility.
If however it goes pay-as-you-go, then the outrage will be warranted.
Intel could have very well gone with this as, purchase an i3 now, and upgrade to an i5 by paying the difference, and help reduce waste. All they'd have to write is "upgradeable up to i7".
Maybe they could charge different prices for different products?
I understand the binning concept but what I don’t agree with is turning off working hardware with the only purpose being price. 6 out of 8 cores work — sell it as a 6 core chip. AVX-512 works just fine — don’t turn it off so you can sell it as a ‘subscription service’.
> ... what I don't agree with is turning off working hardware with the only purpose being price.
Except, that already is what's happening. This changes nothing negatively. The issue Intel and AMD have is that it's too expensive to have multiple masks, so they end up only making one or two. This has the effect that, if a feature works, but they need more of the lower-tied SKUs, they fuse off the working parts. Your 4 core part might actually have 8 of the 12 cores work fine, but they were fused off to sell it cheaper. Whether you find that despicable or not doesn't change the fact that it already happens.
The only thing this changes is that the hardware lockouts (fuses) are now "software" lockouts. You can now pay to access those working features you already had.
Oscilloscope manufacturers have been doing this for years. Rather than stock 8+ different SKUs for a series, each with their own custom PCB, they stock two or three (based on supported channels). Then, everything is software locked. For example, Tektronix's new MSO4 series has bandwidth capability of 1.5 GHz in the hardware, but it'll cost you $27000 (for MSO44). They also have a 200 MHz model for $8420 (also MSO44). There's literally no hardware differences between the two; the bandwidth is entirely software limited. Afterwards, you can call and get a license key to unlock the higher bandwidth without buying a new scope.
> You can now pay to access those working features you already had.
Judging by my interpretation of their website you can have access to these features on a time-based basis, like when you have a peak load on the database server or something. May be wrong because they seem to be intentionally vague but that’s a lot different than a software update to an oscilloscope (which is also kind of dodgy business practices).
Sorry but what if I want a 4-core and your yields improve and you make 6-working cores? Should I have to pay more? There is no clean answer that doesn't suck for someone.
IDK, one would think they could sell the 6 core cheaper when the yield improves and encourage people to spend a couple bucks on a better product. The only difference is the silkscreen on the top of the chip. I seriously doubt they lose money on selling 6-core as a 4-core so this is just a price gouging exercise IMHO.
I’m obviously not the target market for these types of shenanigans because I’ll just take my consumer dollars elsewhere instead of playing their silly games.
I don't even hate this as much as I should. I showed the proper amount of outrage when BMW decided that heated seats should become a subscription, but this time I don't.
Any time that a specific feature brings a lot of value for a small crowd, it's bad business to commoditze it. It's how ECC memory kept being restricted to Xeons, and how hotels were able to charge for Wi-Fi back in the day. Majority of people didnt really need it, but the ones that do will insist.
But here's the thing: those people that absolutely rely on that feature will pay such ludicrous amounts that it subsidizes the product for everyone else. Heck, this one is big enough for Intel to create a sales channel and launch a product for it. That's $50M out the gate. They expect returns that change the bottom line, and significantly so.
Only question is whether the feature in question is a seat heater or ECC memory. If it's something that basically everybody will want, it's rent-seeking. But by the looks of it, I'm not going to unlock this on my parent's computer, so well done Intel?
While you are right in principle, the example with the ECC memory is not really appropriate.
The case with the ECC memory, which has been created by Intel when they have segmented the market between Pentium and Pentium Pro, is an example of a market segmentation technique where the vendor brainwashes the naive part of their customers to believe that they do not need a certain feature and that they might get cheaper products without that feature, but in fact the vendor sells the products with the disabled feature at the same price as before, to the naive customers, while using the fact that they disable the feature wanted by the experienced customers as a means of extortion to force the latter to pay much more money for the non-crippled products.
ECC is a feature that should have been mandatory by law on any computer, at least on all models where the memory is not soldered, but on modules inserted in sockets.
If there is no ECC, that guarantees that the computer will make mistakes from time to time. Even if the consequences of computer errors can have very different costs, depending on the application, nobody needs a computer that will make mistakes.
Even for someone who uses a computer only for games or for watching movies, so that a computer error should have very little importance, the lack of ECC still introduces a very small risk that, if the user is not careful to keep backup copies, some actions like copying some game assets or movies could produce corrupt files.
For someone who uses a computer creatively, the risk of computer errors can become non-negligible.
I think almost all problems with my computers so far have been bad memory or OCZ SSDs. But the performance hit you take for games and some other things is also very real and big. The very best DDR4 ECC memory on the market is DDR4-3600 CL16-19-19-39, which a) is not a good bin to start with, almost certainly with bad tertiary timings and b) costs more than twice as much as equivalent non-ECC.
That's just more market segmentation. If Intel had smartly made ECC a standard feature, we'd just have high speed ECC kits with good timings aimed at enthusiasts. There's no technical limitation on ECC that stops you from running the chips fast at increased voltage.
Well sure, but that's because there's little demand for fast ecc memory by the gaming crowd due to the aforementioned market segmentation. There's no substantial limitation on ddr speed due to ECC itself.
Wow that's a name I haven't heard in a LONG time...
Is there a name for the phenomenon where you don't hear about something for years, but you don't notice until it's mentioned, and suddenly you go "Wait...what happened to them?"
I'm fairly sure I had a 128 GB OCZ SSD back in the day.
I had a couple of OCZ drives. They were pretty flaky and unreliable. The Intel X25–M drives were really the first production ready consumer drives that I ever used. I still have one going to this day 13 years later. My Intel hardware in general has been as bulletproof as hardware gets. I actually believe Intel is underrated and underappreciated. Their quality assurance while not perfect, is the best. My current system is 100% Intel and Samsung.
Empirically that's true, though. Consumers don't seem to care if their machine glitches a few times a year due to memory, especially when their software crashes more often.
For me, regardless of the knowledge level of the user, their time and files are priceless. As a result, the device they use should be reliable.
Of course, a faultless system is unimaginable, but I don't think that multi-year uptime our servers afford should be reserved to these higher end systems.
Even Windows is stable enough to be run for years without major problems, excluding the update & reboot cycles. Hence, I think that every user deserves a system which at least has ECC memory.
If we give consumers Machine Check, SMART, thermal monitoring/throttling, Persistent storage (HDD/SSD) error correction and other goodies, we can give them ECC. It should be a non-issue. Having more ECC RAM around will bring the prices down, too.
Gamers will say that ECC will add latency. They can disable it, or maybe industry will evolve to provide non-ECC systems at a premium for that 2% speed advantage.
I'll vote for an enable/disable switch for ECC, though.
I think the hypothetical gamers from your post would be wrong. AFAIK ECC should, at most, add an insignificant amount of latency into the memory controller. It doesn't add any to the actual memory transactions because the extra bits move in parallel. Registered memory will add latency, but that's a different technology that just tends to be paired with ECC.
You may be right. I did no latency comparisons for these systems. As long as the ECC is handled in the memory controller and not offloaded to the CPU, it will be completely transparent to everything on the platform sans the machine check subsystem, which will log and report these errors in the first place.
Intel and other players offer in-line ECC which interleaves the ECC data with regular data, as such working on regular DRAM configs.
However: in-band ECC is only enabled for Alder Lake on embedded and chromeOS systems. Not available in the SKUs for regular Windows laptop/desktop use.
I look forward to when industry takes this logic further, and requires an additional "business" license if their analytics detect your vehicle use exceeds allowed "personal/family" parameters.
TBH, I don't hate the concept of one-time unlocks of features that are significant differentiators but would be a disproportionate cost to distribute to everyone.
Annual subscription for AVX512? No.
Annual subscription for ECC? lolno
Paying a one-off to enable ECC on your consumer chip? I prefer AMD's "just have it" approach, but I suspect if the market wants it enough, Intel's approach will rapidly change when people start to want ECC more and go with AMD purchases instead in the markets between "personal system" and "megacorp that just pays for it", so whatever.
What'd be really amusing is if the goal is that Intel can sell chips with these features disabled if they were broken, and then send you a replacement gratis if you try to buy the unlock and they go "oh no". That's really the major market value add that I can see Intel getting from this, is being able to have better yields on these by killing rarely-used features and patching over it when someone has a broken part and wants it.
I've seen an enterprise where machines were bought with N CPUs. They actually contained 2N CPUs, and the enterprise could hire and enable extra CPUs on a monthly base. They could also buy them outright.
Year after year, the enterprise under-provisioned these machines. Severely, as in basically using 2N CPUs most of the time. 'Yeah, last year we had to pay a fortune, but that was exceptional. Surely this year nothing unexpected will happen the whole year. We made a plan, but unlike last year, it is perfect'. Engineers pointed out severe holes in the plans year after year, but management didn't want to hear it: It would break the cost-cuts in the budget.
Intel will make a fortune on this in the Enterprise.
The System/360 machines were typically leased, not sold. And the leases had different tiers for different clock speeds. If you decided you needed more power, you'd call up IBM, accept a more expensive contract, their service rep would show up, unlock a padlock on the machine, open a panel, turn a knob, close and lock the machine again, and leave.
Indeed it feels like that - Intel makes only one part, sells it at one price and, if you want to enable the dark silicon, you pay Intel for a key and the feature is enabled.
It's better than what they used to do, when they killed parts of the CPU to match the SKUs they needed to sell. At least you can pay Intel to enable things.
> when they killed parts of the CPU to match the SKUs they needed to sell
Or that part of the CPU was dead to begin with; every wafer of silicon has some defects, but if these defects hit a part of the chip they can disable, they can avoid wasting a whole chip. Of course, if there's high enough demand for the SKU without that part, and low enough demand for the SKU with that part working, they might kill that part even when it's defect-free.
(Of course, if the part can be enabled "on demand", it means that the disabled part must be working, so they cannot be reusing partially defective chips; it smells like a cash grab.)
OTOH, if all I need is a powerful CPU and I have no use for any of the accelerators right now, I can pay a lower price than I would be able to, without Intel needing to make a different part for me.
For Intel, it also helps to fine tune the product lineup by getting more detailed usage information.
Price elasticity will play a role here, but, all things equal, you'll get a slightly better part than the one you'd get with the units permanently disabled.
You should organize politically to compel businesses to behave ethically. They understand the power of political organization - they've built the whole lobbying industry for it.
B2B: You think more granular control of how their goods are used will benefit Intel's business consumers? That Intel won't carefully adjust every parameter to extract the maximum value from each business?
Reminds me of when Microsoft and Sony where going head to head with consoles, and Microsoft tried to introduce a marketplace for selling or sharing a game. The resulting advert from Sony; One guys says to the other "Here just take it" and passes a physical disk to his friend. That ad, basically sunk the idea with consumers.
Post COVID cloud spending is down. Companies are starting to wind in their IT budgets, why would they want to pay yet another subscription?
Assuming that this means certain features on Intel CPUs are only available after purchasing a one-time license: sure why not, masks become increasingly expensive so it might make sense to have a single one and differentiate via software. We could finally have ECC in consumer hardware.
Assuming this means one needs to buy a license in perpetuity every year: fuck no.
Align your infrastructure investment with your business demands with Intel’s API-enabled business offerings that deliver flexible consumption and configurations.
If you are interested in leveraging Intel On Demand please speak with your equipment provider or Intel representative."
Could someone please explain what this actually means? I'm completely lost.
I originally thought this would be some sort of FPGA that can download a "capability" (special purpose accelerator), probably signed with a device unique ID to require license for each copy, so customers can on demand purchase special accelerators.
So, Intel is going to put transistors on commodity chips, then lock them behind a paywall.
This ends one of two ways. (1) AMD provides them without an upsell, or (2) like TSX and persistent memory, they don’t get sufficient real-world usage to have the bugs ironed out in a generation or two, and Intel eventually drops support for them.
I wouldn’t touch any of these features in production, even if Intel gave me a free site license. (Even if that included dev laptops and github / AWS CI, which it would not).
Doesn't matter for b2b, company needs to stay complaint. Anyway experimented with similar unlockable CPUs in past, I wonder what the results where, since they are only now going back to the idea.
> What Is Intel On Demand?
> Align your infrastructure investment with your business demands with Intel’s API-enabled business offerings that deliver flexible consumption and configurations.
That answered all my questions.