• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

News It seems Intel wants to revive Intel Upgrade Service

moinmoin

Platinum Member
Jun 1, 2017
2,775
3,673
136
You remember Intel Upgrade Service? https://en.wikipedia.org/wiki/Intel_Upgrade_Service

Phoronix reports Intel is now pushing a driver into the Linux kernel with the same purpose:
The Intel Software Defined Silicon (SDSi) driver is for supporting the "post-manufacturing mechanism for activating additional silicon features."

While this is apparently "only" for Xeon server chips for now this is a worrying precedence. I honestly hope its inclusion into the kernel is rejected.
 

igor_kavinski

Senior member
Jul 27, 2020
331
153
76
The original Intel Upgrade Service was a greedy attempt on Intel's part to make people pay for some extra performance that they were used to getting through overclocking. I would welcome SDS if it means cheaper processors with features disabled that most people don't use or need. Intel is already doing this segmentation with the K and non-K processors. I don't hear people making a big ruckus about why K processors exist. It's coz not everyone needs an unlocked multiplier to be happy.
 

jpiniero

Lifer
Oct 1, 2010
10,111
2,367
136
The original Intel Upgrade Service was a greedy attempt on Intel's part to make people pay for some extra performance that they were used to getting through overclocking.
Read the Wikipedia page. Had nothing to do with overclocking. It was mainly aimed at giving a slightly higher bin to very low end processors.
 

zir_blazer

Golden Member
Jun 6, 2013
1,062
292
136
I have conflicting feelings about this. For one, we know the basics of binning and die harvesting: Each die has to be tested to measure its properties, like how high it can clock, or if it has broken Cores/Cache/whatever, so you know in which SKUs you could actually use that die in. For example, you aren't using a Zen 3 CPU 8C Chiplet with a broken Core in a 5800X or 5950X since they require fully functional 8C Chiplets, but with 7C it can work in either a 5600X or 5900X by disabling another Core (In which case, you would have an otherwise functional Core wasted as dark silicon). Regardless, such testing means that at some point during the manufacturing process, there should be a system that keeps track of how functional an individual die is. You obviously don't get to see that data, as it is factory only, yet any system involving upgradeable Hardware should include actually burning that data into the die itself so you have a way to know what other parts of the die are functional and could be enabled, but are shipped as disabled.

Also, market dynamics play a role in the fact that otherwise perfect dies go to low end models if the top SKUs aren't selling well but you have major low end demand. This seems to be the case if yields are good but they don't sell at a premium. We had cases like with the 45nm K10.5 where you had Athlons II and Phenoms II unlocking Cores and Cache L2, and while it may have potentially hurt higher end AMD Processors sales (Not that much because at that point the gap with Nehalem was much closer, but on low end AMD was superior), it was limited to certain Chipsets, so it at least helped to give an incentiving feature to buy the latest generation of AM3 Chipset and Motherboards. But, if the vast majority of dies are fully functional, and they DO sell (Case in point, current AMD Zen 3. No low end parts to speak of), there is no reason to have low end SKUs at all.

I'm not inherently against having a way to potentially unlock the remaining functional features and dark silicon that are there. But I know that big corporations being big corporations, they will attempt to monetize it in a way that is inherently anti consumer, like with the upgrade being far more costly than a proper factory configured model (Which was one of the reasons why the original Intel Upgrade Service completely sucked). Or selling features like Hyper Threading and AVX that I don't think that can be individually broken without crapping the entire CPU Core, but are popular features to do market segmentation with.
 

jpiniero

Lifer
Oct 1, 2010
10,111
2,367
136
At least for now it sounds more like rando features on Xeons that they don't want to make specific SKUs for.
 

igor_kavinski

Senior member
Jul 27, 2020
331
153
76
Read the Wikipedia page. Had nothing to do with overclocking. It was mainly aimed at giving a slightly higher bin to very low end processors.
I was referring to unlocking extra frequency through the upgrade cards. Had it taken off with the low end CPUs, it could have meant the end of overclocking the higher end SKUs for free. Thankfully, we got Turbo boost instead.
 

NTMBK

Diamond Member
Nov 14, 2011
9,407
2,890
136
I'm all for this. Multiple different SKUs are already made from the same die, forcing people to throw out a functional CPU to get an upgrade is just wasteful. Much better to let people unlock an extra 10-15% performance down the line, and squeeze another year or two out of old systems.
 
  • Like
Reactions: Tlh97 and Zucker2k

coercitiv

Diamond Member
Jan 24, 2014
4,546
6,262
136
Much better to let people unlock an extra 10-15% performance down the line, and squeeze another year or two out of old systems.
I doubt it's intended for consumer hardware, but I guess I'll keep an open mind.
 

moinmoin

Platinum Member
Jun 1, 2017
2,775
3,673
136
I'm already looking forward to this feature having the usual Intel bugs and malware exploiting them by locking down some for the customer crucial CPU features and keeping the licenses hostage...
At least for now it sounds more like rando features on Xeons that they don't want to make specific SKUs for.
Intel being too shy to add even more SKUs? Honestly can't imagine that.
 

NTMBK

Diamond Member
Nov 14, 2011
9,407
2,890
136
It would be cool if Intel did a "try before you buy" scheme. Give you a 48 hour trial license to run AVX-512, and see if it actually makes your workloads run any faster. If it does, great, pay for the upgrade- if not, leave it turned off.
 

Shivansps

Diamond Member
Sep 11, 2013
3,366
978
136
Imagine Intel moving to a "PPI" model... (Pay Per Instruction), you get the basic x86 stuff working with 2 cores 4 threads for free but you have to pay for every time you use AVX, or more than X threads/cores.

It would be like, free to play, but for CPUs.
 

Doug S

Senior member
Feb 8, 2020
789
1,150
96
I doubt it's intended for consumer hardware, but I guess I'll keep an open mind.
I agree, this is straight out of the IBM mainframe (and then down the road POWER) playbook for enterprise sales.

Consumers hardly EVER upgrade their computer. They won't even upgrade the OS unless Microsoft practically forces it on them like they did with Windows 10, let alone something like RAM (why do you think only people in tech forums whine about Apple soldering RAM to the board and regular consumers don't know or care) Making it easier to do something that consumers have never wanted to do isn't going to be worth the trouble for Intel, the consumer market for this would be pretty small.

I'm skeptical they'd even do it on the desktop at all as most of the same arguments apply for corporate PC fleets, this will be a server thing. This sort of thing has existed in the high end market for servers forever, and not just with IBM. You could buy an HP or Sun server configured with x CPUs but have fewer turned on. They had various options for later paying to enable them as your needs grow, or even temporarily enabling them for peak load needs.

Basically this provided the flexibility "cloud" provides today before cloud was a thing. There's less demand for it now because of that, but there are always some functions a business may not want to put in the cloud for security, reliability, or performance reasons so there's still going to be a market for this sort of "upgrade service". And you can charge an enterprise customer who doesn't want the downtime and hassle of upgrading server hardware more than you could squeeze from 100 consumers for such an "upgrade".
 

NTMBK

Diamond Member
Nov 14, 2011
9,407
2,890
136
I agree, this is straight out of the IBM mainframe (and then down the road POWER) playbook for enterprise sales.

Consumers hardly EVER upgrade their computer. They won't even upgrade the OS unless Microsoft practically forces it on them like they did with Windows 10, let alone something like RAM (why do you think only people in tech forums whine about Apple soldering RAM to the board and regular consumers don't know or care) Making it easier to do something that consumers have never wanted to do isn't going to be worth the trouble for Intel, the consumer market for this would be pretty small.
Opening your PC up and upgrading hardware is scary. Clicking the "more performance upgrade" button and paying $20 isn't scary.
 

SAAA

Senior member
May 14, 2014
522
118
116
Opening your PC up and upgrading hardware is scary. Clicking the "more performance upgrade" button and paying $20 isn't scary.
Download -buy more ram on the go!

Anyway I'm all against it. Just like with games... pay once and enjoy, can't stand seeing virtually locked virtual content, figure real silicon.
 

Doug S

Senior member
Feb 8, 2020
789
1,150
96
Opening your PC up and upgrading hardware is scary. Clicking the "more performance upgrade" button and paying $20 isn't scary.
What exactly would they be getting? Enable more cores? Raise the turbo window?

Intel would be flooded with complaints "I paid for more performance but I can't tell any difference". Because if there is some sort of performance problem it will be software related, or disk fragmentation related (for the dwindling number of PCs sold with hard drives instead of SSDs)

Adding cores or raising the turbo window does not improve perceived performance for the typical user. It doesn't even improve ACTUAL performance unless you are doing repetitive tasks and timing them. i.e. stuff people running benchmarks or fanboys bragging about performance do, but ordinary PC users don't do. Unless you are a power user doing particular tasks and know it will improve your workflow, you won't measure the difference. And if you do, why would you not have bought that faster performance up front, or upgrade your whole system for a much bigger bang.

Because no way Intel would sell something like this for $20. It would be a bigger increment than what you could have paid when you bought the CPU - otherwise people would buy the cheaper one and upgrade once they received it...
 

VirtualLarry

No Lifer
Aug 25, 2001
51,987
6,909
126
Because no way Intel would sell something like this for $20. It would be a bigger increment than what you could have paid when you bought the CPU - otherwise people would buy the cheaper one and upgrade once they received it...
This, 100%!
 

NTMBK

Diamond Member
Nov 14, 2011
9,407
2,890
136
What exactly would they be getting? Enable more cores? Raise the turbo window?

Intel would be flooded with complaints "I paid for more performance but I can't tell any difference". Because if there is some sort of performance problem it will be software related, or disk fragmentation related (for the dwindling number of PCs sold with hard drives instead of SSDs)

Adding cores or raising the turbo window does not improve perceived performance for the typical user. It doesn't even improve ACTUAL performance unless you are doing repetitive tasks and timing them. i.e. stuff people running benchmarks or fanboys bragging about performance do, but ordinary PC users don't do. Unless you are a power user doing particular tasks and know it will improve your workflow, you won't measure the difference. And if you do, why would you not have bought that faster performance up front, or upgrade your whole system for a much bigger bang.

Because no way Intel would sell something like this for $20. It would be a bigger increment than what you could have paid when you bought the CPU - otherwise people would buy the cheaper one and upgrade once they received it...
How about the unlock all the way from an i5 to an i7, for the price difference between the two? Enable HT, unlock more L3 cache, boost the clock speed. Or enable AVX-512 to speed up Photoshop. Or unlock some EUs on your integrated GPU to improve laptop gaming.

There's a lot of potential things they could offer. It makes it easier for users to get precisely the configuration they want, without Intel having to stock and distribute hundreds of different SKUs.

Sure, some of the features will cost more than $20. But it gives a lot more flexibility. How many of us have bought the i5 because it's good enough for current games, then 4 years later wish we had forked out for the i7 because now it's getting bogged down in the latest games? Imagine if you could pay a reasonable upgrade fee and extend the life of your system for another couple of years. I know I would do it on my Haswell i5.
 

zir_blazer

Golden Member
Jun 6, 2013
1,062
292
136
How about the unlock all the way from an i5 to an i7, for the price difference between the two? Enable HT, unlock more L3 cache, boost the clock speed. Or enable AVX-512 to speed up Photoshop. Or unlock some EUs on your integrated GPU to improve laptop gaming.

There's a lot of potential things they could offer. It makes it easier for users to get precisely the configuration they want, without Intel having to stock and distribute hundreds of different SKUs.

Sure, some of the features will cost more than $20. But it gives a lot more flexibility. How many of us have bought the i5 because it's good enough for current games, then 4 years later wish we had forked out for the i7 because now it's getting bogged down in the latest games? Imagine if you could pay a reasonable upgrade fee and extend the life of your system for another couple of years. I know I would do it on my Haswell i5.
The problem with this approach is that they are selling you good silicon ahead of time for potentially no profit. An upgradeable SKU that is unlockable to +X more Cores/Cache L3 means that you are giving away extra known-good silicon on lower end models, which works like giving a loan that you don't know if you will recover. I don't see a good point in doing so.

The only way for this to be flexible AND potentially useful for an end consumer is simply that each individual Processor has its own internal table of what other parts of the die are disabled on that SKU but otherwise good to go, yet without any guarantee than that particular SKU has anything at all to unlock. Upgrade prices should be variable based on what can be enabled, thus completely different on every individual Processor. It should be intended to monetize otherwise dark silicon, not to create more SKUs with good silicon as a loan with no collateral.
For example, suppose that you purchased an Athlon II X3, which could be based on either the Deneb die (4C with Cache L3) or Propus die (4C with no Cache L3). Maybe the disabled Core works, maybe it doesn't, maybe the Cache L3 is present and works, or maybe it doesn't, or it could not even be present at all. Worst case scenario was that you couldn't unlock anything, best case scenario that it was a fully functional Deneb (Thus a Phenom II X4). Since all that stuff gets tested at factory anyways, silicon vendors could write that directly to an internal ROM on the die or whatever, so you could potentially unlock some extra stuff that the vendor tells you that it IS good, unlike with the Athlon/Phenom II generation where you could unlock everything but you still had to stability test and such.
 
  • Like
Reactions: Tlh97

NTMBK

Diamond Member
Nov 14, 2011
9,407
2,890
136
The problem with this approach is that they are selling you good silicon ahead of time for potentially no profit. An upgradeable SKU that is unlockable to +X more Cores/Cache L3 means that you are giving away extra known-good silicon on lower end models, which works like giving a loan that you don't know if you will recover. I don't see a good point in doing so.
Intel already does this, with NO way of recovering the lost revenue. Do you really think that Intel's yields are so bad that they couldn't enable Hyperthreading on every part if they chose to? I expect that 95% of chips could already be run as the highest SKU for that given die, if Intel wanted to. (Barring things running at crazy high clocks.) Intel has to segment things to provide a range of price points, so that they can extract maximum sales and revenues from the market without producing 200 different physical dies.

The only way for this to be flexible AND potentially useful for an end consumer is simply that each individual Processor has its own internal table of what other parts of the die are disabled on that SKU but otherwise good to go, yet without any guarantee than that particular SKU has anything at all to unlock. Upgrade prices should be variable based on what can be enabled, thus completely different on every individual Processor. It should be intended to monetize otherwise dark silicon, not to create more SKUs with good silicon as a loan with no collateral.
For example, suppose that you purchased an Athlon II X3, which could be based on either the Deneb die (4C with Cache L3) or Propus die (4C with no Cache L3). Maybe the disabled Core works, maybe it doesn't, maybe the Cache L3 is present and works, or maybe it doesn't, or it could not even be present at all. Worst case scenario was that you couldn't unlock anything, best case scenario that it was a fully functional Deneb (Thus a Phenom II X4). Since all that stuff gets tested at factory anyways, silicon vendors could write that directly to an internal ROM on the die or whatever, so you could potentially unlock some extra stuff that the vendor tells you that it IS good, unlike with the Athlon/Phenom II generation where you could unlock everything but you still had to stability test and such.
I think it would be reasonable enough to just have the SKUs have a certain set of things they could guaranteed unlock. That's how they did it in the past, with Intel Upgrade service- "this SKU can unlock this much extra". So they would need slightly tighter binning on their SKUs to allow it, but that should not be a problem for anything but the highest clocked parts.
 
  • Like
Reactions: Tlh97 and Zucker2k

Doug S

Senior member
Feb 8, 2020
789
1,150
96
How about the unlock all the way from an i5 to an i7, for the price difference between the two? Enable HT, unlock more L3 cache, boost the clock speed. Or enable AVX-512 to speed up Photoshop. Or unlock some EUs on your integrated GPU to improve laptop gaming.
Unlock from an i5 to an i7 "for the price difference between the two"? Wow, dreaming much?

More likely they'd charge 3x the price difference, because of the worry they'd sell more i5s to people who say "that's good enough for now" and not only that but give themselves a kicker down to the road to i7 (as if it would be possible to notice without careful benchmarking, but if people are paying for it they'll convince themselves they see it) and figure that's good enough to delay buying a new PC another year or two beyond when they otherwise would have. Intel is in business to make money, not encourage its customers to spend less up front and upgrade less often!

And who do you think this would be targeted at? Does the average consumer know the difference between i5 and i7? Hell, I DON'T KNOW, and neither do you - because the difference between the two varies quite a bit depending on what i5 you are starting with and what i7 you want to end up on. All anyone knows is "the higher the random marketing number Intel has assigned, the better it theoretically is". That's hardly a recipe for mass consumer acceptance lol!
 

ASK THE COMMUNITY