Intel reveals pay-to-play Xeon features with software-defined silicon Intel's software-defined silicon service will let organizations pay money to enable features that are hardwired into future Xeon server processors such as Intel Software Guard Extensions, signaling a major shift in how users pay for computer chips. The x86 giant's software-defined silicon (SDSi) service is officially known as …
Yep. Much as I don't mind not paying for what I'm not using. I'm obviously paying for it if you're shipping it to my door regardless and then making me pay again to use it.
It taking all the worst parts of cloud, all the worst parts of on prem, and mashing them into one hideous middle finger to the customer.
AMD has a wide open door to walk through here at least. Made massive inroads in desktops and consumer. Intel is just letting them take servers now.
You have to design in all the features because it would be prohibitive cost-wise to make chips with and without stuff like SGE. In the past they would fuse off some features and have a gazillion SKUs with various combinations of features at various price points which was doing customers no favors either. This lets you pick and choose what features you need and only pay for them, rather than finding whatever SKU includes what you need.
Compare this buying a new car today where you have stuff like "premium package" that includes the stuff you want plus a bunch of expensive things you don't, versus being able to select what you want option by option so you could have heated seats that aren't leather or get the upgraded stereo without paying for an onboard nav system too.
Cars adopting this model is a different matter, at least for hardware features where they are incurring additional cost to install that hardware in every car sold hoping people will upgrade later (Tesla including all the hardware for "full self driving" whether you buy that feature or not benig the worst offender) This sort of thing will clearly increase manufacturing cost as you build in a bunch of features people aren't paying for, hoping they decide they want those features later - or maybe the second owner does.
The problem is this is quickly going to go from "one time fee to enable SGE" to a yearly subscription model. Wall Street gives higher multipliers to companies that can generate ongoing income versus product sales income - just look at Apple going from a 10-15x multipler to 20-25x once their subscription income started becoming a meaningful portion of the overall picture. iPhone revenue is still half the business, so this was more of a psychological change but companies are simply responding to what Wall Street rewards - because CEOs care about stock prices because their income depends to a large extent on rising stock prices.
If the parts are only a few dollars (heating elements for seats are probably a good example) then sure. But what about all the extra gear Tesla installs to support their "full self driving" package whether or not it is paid for? There's no way there is less cost installing all those cameras and the computing horsepower to utilize them.
If the functionality is already built into the device but intentionally and reversibly deactivated by the manufacturer, then the rightful owner should have the right to reactivate it themself, without any obligation to the manufacturer, and to tell others how to replicate the process.
If the functionality is already built into the device but intentionally and reversibly deactivated by the manufacturer
I have a very vague memory of IBM doing something akin to this with their mainframes - shipped with full functionality but with the features that the customer hadn't ordered turned off.
An 'upgrade' would consist of the engineer turning up and tweaking the setup to enable the feature requested. And adding a few more digits onto the monthly support cost..
Oh, they still do that with the hardware for the iSeries, IIRC.
And IIRC, if we need to unlock more CPU on it, it's "receive feature key, input into specified place in the OS, and IPL when it asks to" or some such; I honestly don't know, that's one of the few systems I don't admin for [RedactedCo].
That kind of pricing model with cars is intentional, you can basically throw in useless options costing something like $50 to install and calculate $200 MSRP for it and people will have to just deal with it if they want to have another $50 safety feature in $3000 package.
For some reason you end up picking everything if you want to have all safety features available. And for some reason you end up paying around 1/3 from base model price extra.
Luckily there is competition so base models might be making loss to the manufacturer. So they are moving forward and linking subscriptions into accounts and such that means every owner needs to purchase licenses of their own.
But in short customers will be paying larger lifecycle costs and companies are getting higher returns. Bad calculation for the consumer.
If these CPU features are priced nicely for the manufacturer, costs will be passed direct to consumers with comfortable margins.
Just like with the BMW, where additional features which you don't want to pay for weight a significant amount and have to be dragged around with you everywhere affecting MPG, in the case of the chip if the power gating isn't perfect the feature circuitry will be using energy and producing unwanted heat you have to pay to take away.
In the case of the power of a car, the fee for the upgrade could be considered as a warranty against the greater probability of something breaking due to use of the increased horsepower. Similarly the extra cost of an extended range Tesla is a sensible hedge against early battery failure.
However, an extra charge for fitted capabilities that have no bearing on reliability is just an excuse to milk the buyer.
> considered as a warranty against the greater probability of something breaking due to use of the increased power
That sounds exactly like an admission that the vehicle is not suited for that extra power, i.e. the manufacturer would implicitly be admitting that it should never have been offered with the extra power.
Does this mean the system with an SDSi CPU has to have a service running to communicate with Intel HQ and enable the chip on each reboot? Sounds needlessly complex, never mind the security risks. If problems crop up, especially vulns, I could see this costing more than just hard wiring chips to be forever enabled or not (and why disable when you've got AMD on your heels?).
Subscriptions for broswer soft - I can see that because the browser is sandboxed. The CPU is not.
It's going to affect their yields, as it means selling chips where the features have been verified as working, but Intel don't know whether the customer will ever actually pay for the feature or not.
I predict that 99.9% of customers either enable it at initial purchase or never, thus effectively turning it into an expensive way of having multiple SKUs.
A customer cannot test them.
A bad CPU module usually intermittently fails at random, causing either a crash (which they will blame on the software) or worse, wrong answers.
Which Intel will also blame on the software.
At least initially, and for long enough that the "extra" features gain a reputation for being unreliable and killing the product name forever.
So you're probably right.
This reminds me of the early days of software licencing. Problems with validation, lost keys, paid-for but did not work, second hand sales, licence transfers etc etc. all left users very frustrated and angry.
I use, and supply, a used servers and I wonder how this is going to play out? Will a processor with licenced features enabled still have those features? If licences have lapsed will the CPU just lock-up? What about software the requires a licenced feature to be enabled but isn't - what then?
If we think the Wintel conspiracy is over then wait for Microsoft to use the "can't live without" features in their software which will force users to pay Intel a bit more than they budgeted for.
It's a minefield and, hopefully, AMD will use the debacle to get more sales.
How is this in any way different from how IBM has been marketing their mainframes for the past several decades?
The "chip" (size of a DVD case) can scale from a handful to dozens (if not hundreds) of processors.
It is cheaper for IBM to make identical chips, and allow people can pay for whatever power they need. This allows small organizations to buy only what they need, instead of IBM having to price to the median.
If you need more power, a mainframe box can scale quickly and massively through 'just' a simple software key download (all it takes is $$$$$$).
No; never worked FOR IBM, but spent the last 10 years of my career managing the mainframe capacity and upgrade cycles of a company with four large-ish mainframes.
"Enable x% of the identical cores in the box" is much simpler.
They will have multiple SKUs with different "up to" figures, so they have space for yield and the customer knows up-front how far they can upgrade without swapping hardware.
Aside from that, mainframe volumes are pretty low and entry level prices rather high.
There's no list prices, so it's easy enough to ship an SKU with enough working headroom and price in the risk of going in and swapping a CPU block if that particular customer ever buys more than the number that actually work.
Mainframes and servers are very different beasts.
Mainframes have always been tailored to customer needs, and they have always been very expensive. That is why servers took off in the first place - you could have a server for a pittance (compared to a mainframe) and their power, reliability and feature set has never stopped growing.
Now, Intel is trying to shoehorn a subscription model to its line of servers.
Good luck with that.
Mainframes ARE servers, much like in the way of any other server - they use the same Von Neumann architecture. Where they differ is in scalability and resiliency (and multiple embedded accelerators, and the different operating systems, etc), but to call them "not a server" is simply incorrect. Even IBM calls them servers (or, to be more modern, both mainframes and Power and called "Systems").
The ability to ship a full machine with resources deactivated and activate using software keys is part of the mainframe for decades. But every other commercial system has been offering a similar capability for ages. IBM I (former AS/400) had this in the 90s. IBM AIX-based systems (pSeries, System p, Power Systems, depending on the year) introduced this capability in early 2000s. HP Superdome had it. I believe even the Sun Fire 15K had it. It is nothig new, has been old news for ages, except that it is news now because Intel wants to do it. It is news for Intel, not for the IT world.
And yes, it works fine, and it is a way to make a server more affordable. And it does not make it more expensive (maybe slightly if you go for full activation, but without the feature, the manufacturers would simply charge the full price on it).
Moreover, these days the HW cost is not the important factor. If Intel charges, say, $2000 to activate a core, Oracle charges $47,000 to license it to run Oracle Enterprise. So, if no all cores are needed, turning them off brings a significant costs savings. This is the major motivation for the capability. The second one is the instant upgrade that it allows, with zero or near zero downtime.
I worked for IBM on technical pre-sales for Power Systems (still work for IBM in another area). I designed many solutions using on demand capabilities on Power Systems, and several large companies use it regularly, along with dynamic relocation of resources, to manage their production systems.
It is cheaper for IBM to make identical chips
I seem to recall that Intel used to (may still do) have one chip-manufacturing process - they would make all the chips capable of the highest Ghz rating. Then the ones that fail at that frequency would be tested at the next highest and, if they passed, would be sold at that frequency.
Rise and repeat until the chip fails all the testing frequency - at which point it gets crunched.
Its called Binning and they most certainly still do it. It saves them money by being able to sell a processor that they'd otherwise have to destroy. I don't know how common it is for other manufacturers but I know Nvidia does the same and Id imagine Samsung and TSMC do it as well.
I'm assuming the SKUs with features disabled will be much cheaper, to undercut AMD, with the full-featured SKUs coming in at their usual price. Over time, if Intel starts gaining back on AMD, I suppose the full-feature SKUs will grow exorbitant in price, so Intel will get to keep their "you can always pay for fully unlocked" promise, but making it financially impractical to do so, forcing customers to get the gimped versions and pay for the 1-2 features they actually need, which will then come up to around how much they used to pay for full chips before, with room for extra revenue for Intel down the line should those customers' needs change.
That, or sell a full SKU and then start charging subscriptions in 5 years for the SDSi SKUs.
So the company incurs the full cost of developing and producing powerful hardware and software and then sells an inferior product instead. Doesn't sound like a great business decision to me. I'd rather buy from the competition who sells a fully functional product.
To sell a deliberately inferior product sounds like a bad idea to me. It might work if you are the only supplier in town, but it is not a way to build and gain market share in a competitive environment. I believe these managers should go sell razor blades, printer ink, or coffee pods instead.
I guess this also gives Microsoft further slots to charge for "extra" capabilities in Windows that makes better use of Intel's "extra" hardware. As if Microsoft's SKUs and terms weren't already confusticating enough.
Seems rather stupid, really. Or rather, it smells/stinks strongly of IBM.
We desperately need an IT-savvy government somewhere to intervene to prevent exactly this kind of thing.
If not with a law banning it outright, then with a law specifically protecting every user's right (1) to defeat any software-based system that attempts to restrict, for the purpose of price differentiation, functionality already built into a product they own even if they only paid for a cheaper version ostensibly without this functionality, and (2) to publish instructions to accomplish this for the benefit of others.
If a hi-fi manufacturer in the 1960s had released a "mono but upgradeable by authorised dealers" FM stereo tuner that sold for a price significantly less than its stereo sister model, but actually had the stereo demultiplexer already fitted, and the "dealer upgrade" for which a hefty premium would be payable later consisted of snipping a wire link between the left and right outputs and removing a black cover over the "stereo" indicator lamp, I cannot imagine consumers being very happy.
How about a non-charitable reading instead:
- give them a way to spread out equipment costs Well now, that is the job of bankers and lease companies, no? Intel is trying to get in front of the finance services and collect cash directly.
- scale capacity Nope. The new features are incremental at best, and their performance impact will be microscopic compared to all the other work being done by these CPUs. Microbenchmarks are not the way to evaluate them, by the way. You get one guess how they will be marketed.
- throughout the lifecycle of the Xeon processor. No way Jose. The lifecycle of the Xeon processor in the original customer's hands. If you sell it or transfer it, for sure the licenses for features will not transfer with the CPU and subsequent users will have to pay pay pay. See Microsoft, IBM, Dell, HP, and other vendors for precedent.
- deliver what they [customers] truly want See all comments from other customers, above. Nope.
Other items not mentioned in the article:
- Software costs. It's bad enough that not all current CPU offerings include features that lots of customers like (for video transcoding for example). But these new "enterprisy" features will soon be mandatory for "enterprisy" software like Oracle, SAP, and other because Intel will convince these vendors that what is good for Intel is good for them.
- Hardware lifecycle costs. The news that AWS is keeping their 2017-era CPUs in service is an example of a successful monoculture with 99% feature compatibility for several product generations that is useful for many AWS customers. Add a completely orthogonal set of optional features and the monoculture is then fractured into numerous sub-cults that won't interoperate because software.
- Hyperscaler problems. You and I don't face this but Google and Facebook do. What optional features should they buy? Should they buy in bulk? It multiplies their choices and they may well opt to vote with their feet towards ARM and AMD.
In the test equipment industry.
I am not defending Intel here as they are in a (relatively) volume market compared to test equipment.
A fully featured, high end oscilloscope (capable of analysing / displaying multigigabit signals or analysing complex systems for timing requirements) can cost upwards of £200K when everything (particularly the probes which can cost £25K each for the highest performance parts). That is but one example - there are many.
Now there are some who would like to have an upgradeable piece of equipment but simply don't have the budget for one of those all shiny top of the line [1]. For them, it can make sense to buy something that is capable of that level of performance but that level of performance will not be required for a year or more so they pay a lower price but still have an upgrade path.
There are many pieces of test equipment that are sold this way but the high end test equipment market does not have the volumes for economies of scale. Paying for an upgrade is less than buying the two pieces of equipment.
So it can make sense in some areas.
[1]. There are less expensive pieces of equipment, certainly, but those are not upgradeable, generally speaking so that will always be limited to its given specifications.
Not sure why you're downvoted, as this is very true. The T&M industry has, for a long time, more or less a racket where they charge you extra for functionality that already is baked into any of their test instruments - both in hardware and software.
Which is probably why the traditional T&M vendors like Keysight or Tektronix (who's been struggling to make a decent scope since the end of the analog era) are getting eaten alive from from the bottom end by Chinese manufacturers like Siglent and Rigol.
...occur to me.
* Will software need to be recompiled with/without these extensions to take advantage of it...? I can for see an infrastructure manager unlocking a CPU feature but the software not being compiled to use it and not benefiting.
* Seems Intel are heading towards granular licensing of CPU features / microcode / cores as well as billing based on consuming a feature. This is a mature business model both in Cloud and mainframe / minicomputers going back decades. It is just new to the server space.
Not a fan of this, but for Intel to keep their investors happy, they need to explore new ways of getting more and more money from customers - even if it is unpopular.
> ...select fully featured premium SKUs or the opportunity to pick and choose the features that matter most to their business at any time throughout the lifecycle of the Xeon processor.
So is the mouthpiece saying users don't have to pay for this? Of course not. The only choice being offered is pay or don't pay (i.e. don't get), which is much like now. More worrying is the implication that there will be a continuing rental payment.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
