Intel Officially Introduces Pay-As-You-Go Chip Licensing (tomshardware.com) 141
Intel has officially revealed its Intel On Demand program that will activate select accelerators and features of the company's upcoming Xeon Scalable Sapphire Rapids processor. The new pay-as-you-go program will allow Intel to reduce the number of SKUs it ships while still capitalizing on the technologies it has to offer. From a report: Furthermore, its clients will be able to upgrade their machines without replacing actual hardware or offering additional services to their clients. Intel's upcoming Intel's 4th Generation Xeon Scalable Sapphire Rapids processors are equipped with various special-purpose accelerators and security technologies that all customers do not need at all times. To offer such end-users additional flexibility regarding investments, Intel will deliver them to buy its CPUs with those capabilities disabled but turn them on if they are needed at some point. The Software Defined Silicon (SDSi) technology will also allow Intel to sell fewer CPU models and then enable its clients or partners to activate certain features if needed (to use them on-prem or offer them as a service). The list of technologies that Intel wants to make available on demand includes Software Guard Extensions, Dynamic Load Balancer (DLB), Intel Data Streaming Accelerator (DSA), Intel In-Memory Analytics Accelerator (IAA), Intel In-Memory Analytics Accelerator, and Intel QuickAssist Technology (QAT) to accelerate specific workloads.
Might as well go all the way (Score:5, Insightful)
When you're going there, might as well go all the way and pay IBM for an actual mainframe with CPUs worth the name.
Re: (Score:3)
you mean when ibm were actually ibm and actually had systems worth the name, not the sad shadow it is today, right?
Re:Might as well go all the way (Score:5, Interesting)
The recent POWER10 systems are excellent systems. We run IBM i (used to be called AS/400) 7.5 and it's an insanely good product that allows us to take legacy COBOL and RPGLE service programs and hook them up to node.js web services and the system when the subsystems are correctly setup, purrs at everything our busiest days can throw at it. Now only that, the enhancements that they've made to the DB2 product and the RPGLE language makes it insanely easy to write new code that incredibly powerful. Digesting a webservice and doing all the required logic can even be pure SQL with minimal RPGLE or COBOL to handle any needed system parameters.
I don't deal with the z systems today but I've dealt with them and their performance is top notch and the CICS interface is as fast as ever and the IBM JVM on z is incredibly fast enterprise wise versus anything I've seen a Wintel machine able to push.
So I'm absolutely curious as to what you might be talking about as I've had nothing but incredibly good performance on the i and z systems. I'm not doubting you I've absolutely seen horribly configured systems, but I'm just curious as to how you've formed your opinion?
Re: (Score:2)
Excellent compared to what? Have you attempted the same workloads on Milan or Genoa systems? Or even IceLake-SP? You might like the tools that come with the hardware, but without performance comparisons, such testimonies don't mean much.
There were some Phoronix benchmarks featuring Raptor Talos II performance, but good luck finding anything else, even on servethehome.
Re: (Score:2)
Most of Sierra and Summit's Top500 horsepower come from accelerators. POWER9 isn't doing the heavy lifting. Fugaku, however. ..
I'll see if I can dig up some old POWER8 benches from Anandtech. If memory serves, perf/watt on POWER8 was kind of bad.
Re: (Score:2)
https://www.anandtech.com/show... [anandtech.com]
There. It's old - 2015 - but since you mentioned POWER8, there you have it. That was the spot that made the IBM system look quite bad. And it's not surprising that getting systems from IBM for comparative benches after this review (and some others of the same gen) became . . . difficult? I mean servethehome doesn't have any recent POWER benchmarks. Kind of sad.
Anyway in terms of TCO the IBM system was an albatross. At least it put up good SAP numbers.
Re: (Score:2)
well, fun fact: i actually learned my first coding at ibm back in the 80s. i never got my hands on a 360 or 370 but i have programmed extensively on s36, s38, as400 and z series for many years, i know them well and they are indeed very good, rock solid and performing systems with excellent developer and operation tools. they were also very expensive. also, best keyboards ever, hands down!
however, those are basically the evolution of the same os from the 80s with discrete improvements. i wouldn't say that "d
Re: (Score:2)
Or you can just get superior AMD enterprise hardware. Intel's performance and perf/watt deficit is bordering on the enormous. Genoa is already available while Sapphire Rapids is not (due to Intel totally screwing up the launch thanks to major hardware bugs), and Genoa is a much better product. You have to really want those fixed function accelerators in Sapphire Rapids for the product to make any sense.
Unless SDSi makes Sapphire Rapids cheap (which it probably won't), you would have to be criminally insa
Re:Might as well go all the way (Score:5, Insightful)
Or just wait until someone figures out how to unlock those extra features for free.
That actually became part of the business model with some test and measurement equipment. The manufacturer makes a single model of oscilloscope, with software locks to limit bandwidth and features. Businesses will just pay for what they need, where as hobbyists will use the hacks that unlock everything. They get to serve both market segments with the same model, at different price points.
Re: (Score:3)
I'm sure they did, but it won't stop people doing it. If Intel can unlock it, so can anyone else, it's just a case of figuring out how.
Re: (Score:2)
That implies they implemented the security features as security by obscurity, and all you have to do is 'figure it out'. My guess is they do it with signed config files, verified and acted on in the silicon.
Re: (Score:2)
It is very likely to be public key based crypto, but that hasn't stopped people hacking the various games consoles that have used it over the years. It all relies on the private key remaining private, and it turns out that is very hard to do.
Welcome to the Mainframe world (Score:5, Insightful)
So we went all the way from host-dumb terminal, to client-server, to multi-tier web applications, to VMs, to Cloud, and now back to the paying by the mips model of IBM Mainframes.
What's next? Ondemand capacity? Let you buy and install CPUs without paying it up front, only activate and pay when you needed it? (Hint, yeah, that's available on Mainframes too)
Re: (Score:3)
"What's next? Ondemand capacity? "
Yup, surge pricing for extra MHz & memory bandwidth
Re: (Score:2)
Re:Welcome to the Mainframe world (Score:5, Interesting)
Re: (Score:3)
it will take 5 years or so to see people jumping ship to other platforms.
The permanent solution will be Linux on RISC-V.
It will take more than 5 years, but we'll get there eventually.
Re: Welcome to the Mainframe world (Score:5, Interesting)
Re: (Score:2)
You mean in terms of performance?
No, I mean in terms of market dominance.
so it might be closer than you think
I hope so.
Re:Welcome to the Mainframe world (Score:4, Interesting)
Or, you know, just buy AMD and tell your vendor that you're switching to avoid hardware licensing fees.
Re: (Score:2)
Amazing. You claim 'the deal stinks', then go on to say exactly why it doesn't stink. Much mainframe software is priced by the capacity of the machine it is running on. Thus, mainframes allow you to closely tailor the capacity of the machine to your actual workload. This saves you money on both the hardware (even more importantly) the software. If your workload increases, you can instantly add capacity. One of your data centers got knocked offline? Instantly add capacity to the other one. Buying a n
Re: (Score:2)
See, when you buy more time/bandwidth from a mainframe or supercomputer, it's something they actually provide.
It's the difference between someone charging you for delivery of a package (which is fine because it's a service they've done for you) and someone making your package sit at a pick-up centre for twelve hours for no reason, and charging you if you
Re: (Score:2)
Re: (Score:2)
Of course not, and nobody claims otherwise. However, do you think the actual cost of manufacturing the chip increases significantly when features are added? It does not. The majority of the cost is in the development of the chip, not the manufacturing. And with configurable features, the customers who actually want the feature pay for that cost instead of charging people who do not want the feature. So the counter to your question is: would the overall product line (not individual chips) be profitable i
Re: (Score:2)
A NOP operation will always be free (Score:3)
Because even doing nothing at 5GHz will still yield none result whatsoever.
Re: (Score:2)
Re: (Score:2)
Some of Intel's press material seems to indicate that on-demand is what they have in mind:
https://www.intel.com/content/... [intel.com]
Looks like they'll let you pay a one-time activation fee, or they'll let you license on a demand basis.
A trap for idiots (Score:5, Insightful)
Of course by then they might be promoted for cutting costs so effectively anyway, so why should they care.
Re:A trap for idiots (Score:5, Informative)
Re: (Score:2)
Re: (Score:2)
The funny thing is, when people have been calling "the mainframe is dead!" for the last 10 years or so, I've been replying with "No it's not, Intel and AMD are slowly reinventing it with every new generation of Xeon and Opteron/EPYC".
Re:A trap for idiots (Score:5, Funny)
It's like the car manufacturers building heated seat into every car "to reduce inventory variances", but you having to pay extra to "unlock" the feature. Same with "pay extra for remote door locks." Ditto "pay extra for extra range or speed from your BEV."
Shakespeare had it wrong. "First we kill all the MBAs."
same who Put in the raid keys for on chip software (Score:2)
same who Put in the raid keys for on chip software raid and then AMD made there version an free thing
Re: (Score:2)
Re: (Score:2)
SKU reduction? (Score:4, Interesting)
The new pay-as-you-go program will allow Intel to reduce the number of SKUs it ships while still capitalizing on the technologies it has to offer.
Is that so? I always thought that the proliferation of SKUs was because Intel's production yields were totally crap so they were disabling features on chip functions that failed tests and just relabeled them as a lesser part. Are they finally confident in their production processes?
Re: (Score:2)
No. It's just more segmentation. They'll still have defective dice.
Re: (Score:2)
Yes, and to be fair it is not only Intel that does that, all the major CPU manufacturers do the same.
Re: (Score:2)
AMD to win when Intel fucks this up? (Score:2)
AMD to win when Intel fucks this up?
Re: (Score:2)
If AMD doesn't do the same thing, then it has already buring Intel. However, I can't imagine Intel doing something so stupid without assurances that AMD won't play the same game.
Either Intel is slicing its own throat, or there is an antitrust violation that needs to be prosecuted.
Re: (Score:2)
Re: (Score:2)
"Like right now, run a few VMs and 128 gb is a tight fit."
I've been saying for years that virtualization allowed many vendors to get away from optimizing their bloatware
Re: (Score:2)
There's just some software that won't run on newer OSes, so not really many options as time goes on, and RAM is far cheaper than maintaining a cranky old box with an old OS that can just give up the ghost one day.
Plus there's always things like classic games that they literally don't make any more, even if you wanted to buy a newer version.
Besides, swap is stupid when you have enough RAM. Same as hibernate. Do the calculation for how long it would take to hibernate 128 gb of ram on the fastest m.2 driv
Re: (Score:3)
There's just some software that won't run on newer OSes, so not really many options as time goes on, and RAM is far cheaper than maintaining a cranky old box with an old OS that can just give up the ghost one day.
True, but I submit that the reverse is also true here. If you are referring to Windows 95/98-era software, more than even 256MB of RAM can become counterproductive. XP-Era software commonly has issues on 64-bit versions of the OS, so 4GB is the max you can give to 32-bit XP.
So, the 'old software' examples seem a bit tricky to make the math work...you can run 8 Win98 VMs and 8 XP VMs and two Docker hosts the size of the one I run with three dozen containers for assorted chores (PiHole, Nextcloud, PhotoPrism,
Re: (Score:2)
XP-Era software commonly has issues on 64-bit versions of the OS, so 4GB is the max you can give to 32-bit XP.
Not sure this is correct. x86 has PAE, which will map a larger address space back on to 32 bit chunks. The OS is 32 bit, and the programs are 32 bit, no individual executable can access more than 4G at once, but the system as a whole can have > 4G.
I remember dealing with large in memory stuff on Linux by mmap'ing chunks in and out of the address space. Very fast, since it's all in RAM. Right pain
Re: (Score:2)
XP-Era software commonly has issues on 64-bit versions of the OS, so 4GB is the max you can give to 32-bit XP.
Not sure this is correct. x86 has PAE, which will map a larger address space back on to 32 bit chunks.
It's "effectively correct". XP Home and XP Pro only did 4GB of RAM, period. Even if you had 8 in the slots, you'd only get 4GB available to the OS. If you're emulating 32-bit XP, you'll never see 8GB of RAM in it regardless of PAE existing.
XP did have a 64-bit release of XP Pro, but it was relatively rare and some 32-bit (or even 16-bit) programs were finicky-at-best, and 64-bit hardware drivers were relatively uncommon (though admittedly this is less of a problem with virtualization). AutoCAD 2006, Adobe C
Re: (Score:2)
It's "effectively correct".
Ah OK. I knew PAE existed and worked under Linux. I remember it being supported in Windows, but I didn't know it didn't work effectively under XP, which was by far the dominant OS.
Re: (Score:2)
I want my next build (sometime towards the end of the decade) to have the following:
Obviously no swap, no hibernation, same as today.
RAM is cheap. Human t
Re: (Score:2)
I can assure you that you’re doing it wrong if 128gb is a “tight fit”. Once upon a time a dozen users could be banging out code on a PDP-11/70 with a single megabyte of ram. A System/36 from IBM could have over 100 users in six megabytes of ram. Software today is bloated shit with abstraction layers slowing things down further. Hell you could run virtual machines back on 1970s IBM S/370 machines. Real concurrent copies of operating systems too. They even had an OS where a new VM was create
Sure, sure, Intel (Score:2)
Given your "stellar" security track record... you really think any intelligent IT person is going to want a processor that, one way or the other, is almost certainly going to be accessible via the network?
Re: (Score:3)
Yep, bunch of assholes. Of course it will be network-accessible. How else are they going to turn features on and off - RFC 2549 - TCP over Carrier Pigeon?
"Gee, sorry, network outage, CPU now disabled, try again later."
Rentier economy indeed. Bet you some id-10-t Harvard genius will suggest extending this to power supplies, UPSes, graphics cards, motherboards, a whole licensing stack. Because "what could go wrong", same as recommending central banks holding crypto.
Re:Sure, sure, Intel (Score:5, Informative)
Intel already has a vulnerability due to the EFI interface that allows software to change voltage: Plundervolt
https://plundervolt.com/ [plundervolt.com]
Will Linux playalong with this and can the DMCA be (Score:2)
Will Linux play-along with this and can the DMCA be used to Lock in say windows server if Linux just happens to bypass / auto lock the cpus at full?
Re: (Score:2)
Yeah, if they copyright the code that is bypassed/circumvented.
Re: (Score:2)
and then use that to lock in windows booting only
Re:Will Linux playalong with this and can the DMCA (Score:5, Informative)
I'd be very surprised indeed if they actually implement any of the license validation or direct hardware control at the OS level when they have the Management Engine available to do that sort of thing; the driver provides a mechanism to enumerate features and to hand keys to the firmware so that you don't have to mess around with some nonsense preboot interface to do it; but doesn't seem to do any sort of license checking itself, just hands the key off and lets the firmware do whatever it sees fit.
will supermicro and others be forced to lock UEFI? (Score:2)
will supermicro and others be forced to lock UEFI to do this and maybe even lock IN MS boot keys / block loading your own keys as loading an NON MS ok OS may just let you unlock the cpu?
Cracked CPU's in 3,2,1.. (Score:3)
I, for one, welcome crackable free upgrades.
Re:Cracked CPU's in 3,2,1.. (Score:4, Interesting)
I, for one, welcome crackable free upgrades.
what's the crack for enabling RAID 6 on HP Proliants?
Re: (Score:2)
apt install zfs ;-P
Re: (Score:2)
Enabling IP over Infiniband switches would be cool too. 40gb links for pennies on the dollar.
Re: (Score:2)
Enabling IP over Infiniband switches would be cool too. 40gb links for pennies on the dollar.
Isn't that already doable?
Re: (Score:3)
Besides very specialised applications, is anyone still bothering with Infiniband when 40GbE and 100GbE are commonplace?
Re: (Score:2)
Second hand Infiniband parts are much cheaper. All the beefy boxes on my home network are 10gb fiber now. With Infiniband you don’t have to run IP, it’s a licensed option on some switches and the ebay market is a crapshoot.
will the basic UEFI need direct internet access (Score:2)
will the basic UEFI need direct internet access to check licensees now? and will this maybe open an back door to the basic firmware that can be used to hack into the system?
I don't think that UEFI bios may have stuff like login into an VPN (and if so basic auth only) login into an web proxy (and if so basic auth only) to call intel own linces check server and if the network is down? the system is hosting an router / proxy / VPN in an VM? (you may have an COLD start up issue)
Re: (Score:3)
will the basic UEFI need direct internet access to check licensees now? and will this maybe open an back door to the basic firmware that can be used to hack into the system?
I don't think that UEFI bios may have stuff like login into an VPN (and if so basic auth only) login into an web proxy (and if so basic auth only) to call intel own linces check server and if the network is down? the system is hosting an router / proxy / VPN in an VM? (you may have an COLD start up issue)
Embedded blockchain to inventory the billablel microtransactions even when disconnected from the internet.
When it get's to a certain threshold, you get an alert to enter a code on the OEM site and get back a reset code.
If not done within a specified time, the firmware will begin turning off features & reducing capability.
Intel, more like Injail (Score:2)
This is an excellent development. I can't wait to buy a 14th gen i9 14900K and use it to emulate a Commodore 64, therefore the upfront price will be - if on par with certain nano scale SBCs - around $5.
Maybe I'll decide to do more with it once the latest kernel supports it better. Until then it's Dam Busters, Silent Service and Pools of Radiance.
I can see it now... (Score:5, Funny)
C:\> calc.exe
ERROR: Opcode FMUL not available. Please insert Credit Card to continue.
Re: (Score:2)
FS0:\> EFI\Microsoft\Boot\winload.efi
ERROR: Compute time expired. Please insert credit card to continue.
FS0:\> EFI\ubuntu\grub.efi
ERROR: Loader not authorized. Please choose an Intel approved loader to continue.
FS0:\> EFI\ubuntu\vmlinuz
ERROR: Non-enforcing kernel. Please choose an Intel approved Linux distribution to continue.
FS0:\> EFI\Intel-Rent-Manager\vmlinuz
ERROR: Compute time expired. Please insert credit card to continue.
FS0:\> reset -s
Thank you for choosing I
Re: (Score:2)
I avoid the cloud (Score:5, Insightful)
I avoid the cloud because it's someone else's computer. The last thing I need is for my computer to also be someone else's computer.
Re: (Score:2)
>The last thing I need is for my computer to also be someone else's computer.
It's not even your computer anymore. The icon on the desktop has been changed from "My Computer" to "This PC" to show this.
Re: (Score:2)
I don't do Windows. It's still my computer.
Re: (Score:2)
Re: (Score:2)
I'm mostly referring to cases where renters really should own. My computing needs are fairly steady, ownership is the sensible approach, but Intel wants to make me a long term renter instead.
Of course, in cloud computing, once you add up all the nickels and dimes, you might be surprised how little you would need to utilize your own equipment to make ownership cheaper than cloud rentals.
Imaging if using that rental truck for a weekend cost 1/4 what it would cost to buy the truck.
Barf (Score:2)
Between this CPU, the subscription seat heating from BMW, and the faster acceleration from MB, I feel like we're living in some kind of dystopian sci-fi world. The assholes who thought this crap up, and the execs who approved it should all be drug out into the street and shot.
But from a business perspective, I'd really like to see the numbers. I don't understand how selling the exact same product at different price tiers can actually save the company money. Unless, of course, the difference in production an
Re: (Score:3)
Between this CPU, the subscription seat heating from BMW, and the faster acceleration from MB, I feel like we're living in some kind of dystopian sci-fi world.
It's only a dystopian nightmare if you're required to actually buy and use that stuff.
For now, anyway, most of us can happily ignore the pay-as-you-go-subscription-model nonsense and buy products that operate under the traditional "you bought it, you own it" model, and most likely watch these "innovative business plans" wither and die from lack of interest.
If/when this sort of thing becomes legally or commercially mandated, then would be the time to look to Katniss to deliver us.
Re: (Score:3)
How long? (Score:5, Interesting)
Furthermore, its clients will be able to upgrade their machines without replacing actual hardware or offering additional services to their clients.
How long will it be before the 'upgrade' will be time-limited unless you pay an additional fee? In other words, how long before you'll pay for the chip up-front AND pay a rental fee to maintain those 'extra' accelerators and features?
From there it's a natural (to psychopathic corporations) step to have a literal EOL. Intel could decide that it's time for you to purchase new hardware, and hobble or brick your existing hardware to force you to pony up for the latest iteration. Anybody who thinks that can't happen really hasn't been paying attention.
Re: (Score:2)
purchase new hardware
rent new hardware
Re: (Score:2)
The other question is, will the fees to unlock drop over time, as newer generation CPUs are released?
If it's 5000 bucks to unlock something now, will it be 3000 bucks in a year, and maybe 200 bucks in 5 years?
And how long will the unlock services in Intel side run for? Can you unlock the CPUs in 10 years time if you are still using that hardware?
Thanks (Score:2)
Maybe I'll buy an ARM chip from one of the many out there doing 100+ cores and massively wide busses. Then attach my own accelerators (GPUs). If I want to spread my costs out instead of a large initial investment, I would have taken out a loan. Intel going the IBM mainframe support model seems like a dumb way to repeat history.
Another company joining the hated club (Score:2)
Pay extra for.. (Score:2)
Re: (Score:2)
well, this is just perverse (Score:2)
but some dipshit is definitely buying it.
Bad ideas in security (Score:2)
If you can turn them on, some enterprising individual can turn them off.
Pencil Trick (Score:2)
Back in 2000, there was the AMD Athlon, which used broken trace to lock the clock multiplier. Using the "Pencil Trick", you reconnected a trace, and that unlocked the multiplier and allowed overclocking.
Anything like this happening again?
I would love to say "Bye bye Intel" (Score:2)
Rental future (Score:2)
Do we really want a future where most people are renting not just their home but all their possessions? That leaves them vulnerable to increases in rent, maybe even forces them to work their whole lives with no chance at retirement and unable to get health/nursing home care if they ever need it. .. I'm rich .. I own all my stuff" .. well guess what .. if a majority of the population rents and has no chance of feeling secure about their future they are going to come for you
So you might think "haha, who cares
Re: (Score:2)
What people seem to miss is that the entire concept of private property rests on the fact that anyone can own private property, just as a feudal social structure rested on every level's sovereignty of their charges, down to a peasant being sovereign over his wife and her over their kids. Because that's how you justify a moral principle to your average person, and it's why so many not at all well off people in the US stand with the
IIRC Intel planned similar in the early 2000s (Score:2)
But with regular chips that all consumers could use.
All the enthusiasts hated the idea, and derided it, as did the online tech sites.
I think it helped push enthusiasts into being even more prone to buying AMD. Perhaps there's irony there in that some suits and bean counters thought it would be a swell way to make up for revenue that was getting lost to AMD.
Re: (Score:2)
IBM actually did this back in the 1970s -- they had a range of mainframe computers that were sold at various price points but the CPU unit was identical, just underclocked and with partially disabled functionality on the lower-end models. The owners could buy an on-site upgrade if they needed more CPU capacity at which point an IBM engineer would arrive at their premises, open a panel and set a switch or two and job done. This was colloquially known as the Golden Screwdriver" upgrade.
I can definitely see th
Re: (Score:2)
Re: (Score:2)
One issue going forward is that the chip manufacturers are pushing towards "zero defect" wafer production. They won't manage zero, of course but a lot more fully-functional chip dies will be produced per wafer in the future if they succeed. That would mean fewer lower-end spec devices available to the consumer since the lower-spec devices are based on high-end dies that fail part of their testing and have the failed cores and other operating units fused off. A partially-defective die with 4 working cores ca
You don't own the hardware (Score:2)
This is a trend that's been developing in several areas, and not just in IT. The most obvious example is in automotive leases. You don't own what you're paying for, you only rent it. With IBM tossing their hat into that particular ring, they're signaling their willingness to exploit that approach as a means of boosting their revenue streams.
Several others have already made the mainframe comparison, so instead of piling on, I'll just say that people should be grateful that trend is self-limiting...because if
Why compete? (Score:2)
Why compete when you can squeeze your customers instead? Here is a market opportunity to show your value, instead you push me on price. Hello AMD.
Re: (Score:2)
Intel will probably charge more for these when limited than AMD will for unlimited processors that are faster than these are when fully unlocked, and AMD has already made big inroads into the server space lately, they have some really beastly CPUs. I think this is going to turn out to be a big foot-shooter, nobody wants this on PCs. It makes sense on mainframes because the costs are so high, but nobody wants PC processors to be as expensive as mainframe processors either.