Intel updates mysterious 'software-defined silicon' code in the Linux kernel Intel has updated the code it says allows the implementation of "software-defined silicon" (SDSi). Chipzilla dropped some code for SDSi into the Linux Kernel in September 2021, describing it as tech that allows users to activate dormant features in silicon. The code outlined a process for enabling new features by verifying …
TBH, sounds like a waste of waste of time and money.
There are already numerous versions of each chip released so price differentials already exist.
The competition is hot so maximizing performance always is required to stay competitive.
And then there is the cost of enforcement. They might be able to work something out with Windows or IOS, whose every customer is under telemetry surveillance 24/7. Maybe impossible with Linux.
Cellphones of course use that strategy - e.g., deliberately crippling embedded AM/FM radio in the US, perhaps to force music seekers to use online music options instead. Notably, the switch is hardware set by region, not a software switch. After some fatalities in CA fires, there was some talk of requiring radio to be enabled in the US so the radio can be used to get emergency info about what roads are open, etc, in disaster situations where the cellphones towers are down and a raging inferno is headed for town, but of course that plan never went anywhere.
Why bother manufacturing several versions of each chip instead of making a single, high-end one then cripple it with software locks ? Doing this has the advantage of forcing the customer into pay (constantly) to play.
As for enforcement on Linux, don't count on that. Intel will simply withhold technical specifications and APIs from open source developers, like Nvidia does with their graphic cards. There will be no open source software for high-end features.
It's not quite that simple... a lot of the lower-end chips are the higher-end chips that didn't make the cut due to manufacturing issues, and so had some of their cores disabled, or were down-clocked because they couldn't handle the higher frequencies in testing. This is called 'binning' where the CPUs are manufactured, and then sorted into 'bins' based on their capabilities.
So, Intel isn't going to only sell high-end CPUs that have to be unlocked, because they would have to just throw out a lot of their lower-end CPUs if that were the case, and they'd lose money.
More likely, this is for things like enabling VROC via software instead of requiring a hardware dongle. Hardware dongles suck, they take up space on the system board, some OEMs won't add the socket for them and simply won't enable the feature, etc. If Intel can unlock such features via signed licenses, it saves on hardware cost AND becomes much easier for both the OEMs and the customers (imagine having to open 500 servers to install VROC dongles, vs. applying a software license via API).
You mean like buying a VAX 750 with 4K of memory, and when you want to upgrade [after shelling out $$$], the fiend (field) engineer, opens the panel, and moves a single DIP switch - presto, 16K.
This has been standard practice for decades.
Hah... going to install it in a Dell server? The newest ones can be 'fused' to the OEM on install, so after once installing it in a Dell server, you will only EVER be able to install it in a Dell server. Put it in a SMC or HPE server and it will simply refuse to boot.
Intel and AMD both make a good product, and they both do some things that really piss me off.
You know, this sounds suspiciously like what IBM, Amdahl and some of the larger mini-computer manufacturers were doing in the '60s, '70s and '80s with microcoded processor extensions.
Pay a fortune, have an engineer turn up with a super secret floppy disk, and get a whole load of new instructions added to your processor.
The only difference now is that we have the Internet, so they have to protect this code with cryptography, to prevent it leaking.
"Customers also lost value as the 'golden screwdriver' enhanced vendors' ability to price discriminate. Customer purchased the machine at the market cost, and then later, if and when they needed the additional capacity, they paid more money for the same hardware. Pure revenue to the vendor with no cost-based justification."
FTFY.
The consumer doesn't benefit, unless you believe that vendors would forgo profitable sales so they could maintain higher standardized pricing. And you shouldn't believe that because vendors that try such a forgoing-sales strategy will lose to competitors who will trade volume for margin. That's of course what ultimately happened with commodity PCs.
We had a 407 accounting machine at college for listing cards. I noticed it was missing every 3rd cycle and I noticed the S1 and S2 relays counting the cycles. It was the S1 or S2 relay which I pulled and the machine ran full speed. The IBM rep said, "Don't do that." So I didn't.
I'm pretty certain this is capacity. You can (or at least could, not sure now) buy, at a discount from the full price, servers with inactive processors and memory, and later buy the licenses to use this extra capacity, and add them without even having an engineer visit. This was called Capacity Upgrade on Demand CUoD)
IBM also provide 'time limited' licenses, which were there so you could license and use the additional resources for a short period, presumably to get you over humps in the workload. This was confusingly called the very similar Capacity on Demand (CoD).
Interestingly, for the Power 7 775 supercomputer clusters, IBM over-delivered on purchased capacity, in a system they called "Fail in Place". The idea was that the customer could use the extra capacity, but then when devices failed (CPU or Torrent hub chip), the failures would be counted against the 'extra' capacity, and not repaired.
Simple failures like memory or adapters would be replaced, but more complex failures would be left, so long as the capacity did not drop below the purchased capacity. At various points in the life of the system, the remaining capacity would be evaluated, and if it was deemed insufficient for the remaining life of the system, repair actions would be initiated.
This worked close to the theory, but it was found that when certain "Fail in Place" components failed, like the Torrent chip or the optical links, because of the complicated mesh network topology, it could affect communication speeds between adjacent octants (hardware partitions within single drawers).
This is a problem with step-locked HPC tasks, and any jobs including any node in the same compute drawer in the cluster could be slowed down. This meant that they ended up fixing things that they thought they didn't have to when the systems were designed.
But someone has to pay for the development costs of that extra silicon. Should you charge everyone for it from the first release, even if most people don't need it? Or should you allow people to only buy what they need? A scheme like this means cheaper silicon for the basic users, and people who want better performance simply pay more for it.
Isn't this a pretty standard thing in a lot of industries? Cars? Software?
Software, yes. Cars, not so much. It isn't unusual to be given a code to unlock additional software features as those features are typically already installed and just need to be enabled.
However, I cannot recall the last time I was given a code to unlock 2 additional cylinders or enable LED headlamps. Car model trim prices increase because additional items have been physically added to the car.
I have a feeling Intel is looking for a way to get on the yearly licensing income gravy train without having to ship anything.
>Isn't this a pretty standard thing in a lot of industries? Cars? Software?
I was reading just yesterday about car manufacturers seeing a bright future with software subscriptions as a significant income stream.
https://jalopnik.com/a-carmaker-s-23-billion-plan-to-keep-you-paying-long-a-1848172449
This sort of article is useful because we're looking for a replacement car and this gives us a hint about what manufacturers to avoid. I don't see this as "buying what you need" but rather a seductive profit option that sidelines product development in favour of putting effort secure paywalls of existing product. The result is stagnation (we're seeing this with entertainment media at the moment).
>Your computer is now emulating a BBC Micro until you pay up, scum.
That is going to be interesting, what with online payments and exchange of certificates etc, I suspect that (paying up) is going to take a few days and that's if the web page and payments processing work in BBC Micro mode (complete with dial-up Internet?)...
Sparked a memory of the cost-reduced IBM 407 accounting machines that rented at a lower price but ran at 2/3 speed.
Many universities had them (for listing ones card decks before submitting a job, among other tasks).
It seems that if a small piece of card stock accidentally managed to find its way between a certain set of contacts,
the machine would run faster.
But I'm sure this sort of thing was not done at my school's computer center (other than maybe maybe between 2300 and 0500)
in pulling this proprietary functionality from the kernel. I mean, if there is no plan and it may never be used, what concern to Intel could that possibly cause?
But if Greg, Emperor and all are rash enough to let it stay, then how long it will be before the magic keys are hacked, made publicly available, and regenerated on-the-fly by the kernel whenever it needs them. ISTR something like this happening once in the past, but brain fade gets us all in the end.
"Your car is in rescue mode due to your recent posts in anti-Oceania Government chat rooms abroad. You can only follow one of the preset paths to one of the nearest police stations for an informative chat."
"Your car runs in rescue mode due to your Vaccine Pass expiration. The nearest vaccination point is 0.6 km from your actual position (6 min on feet, based on your walking speed profiled data). Follow the map that was sent to your mobile phone."
"The number you dialed is associated with an expired Vaccine Pass. Please try again later."
"Sorry, Winston. Your mobile phone hasn't been authorized to film the police officers' actions during this protest".
...etc...
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
