Intel Finalizes 'Intel on Demand' Pay-As-You-Go Mechanism for CPUs

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,785
Reminds of the Intel Upgrade Service. This can't be good.


1667513900561.png


"From previously uncovered beans about Intel's SDSi, or Intel On Demand if you prefer, we already know that the program enables access to the interface in the CPU to allow silicon features with an Authentication Key Certificate (AKC) and Capability Activation Payload (CAP) license. We also know that the program allows us to enable the specific feature on a particular CPU socket, not across all processors in the system or the data center itself.

Meanwhile, the fact that the software will need to discover which capabilities are physically supported by a processor and hide those not supported means that not all Intel Xeon Scalable 'Sapphire Rapids' processors will be created equal. Some CPU models may not gain support for certain features even by using Intel On Demand software.


Not all users will require AMX, DLB, DSA, IAA, and QAT at once. But which will be enabled by default on all SKUs and which will have to be activated using the IOD software is something that Intel will probably reveal on January 10, when it launches its next-generation Xeon Scalable CPUs officially."


https://www.tomshardware.com/news/intel-finalizes-intel-on-demand-pay-as-you-go-mechanism
 
Pretty sketchy... and reminds me of what many car companies are doing where you buy a car but then have to pay extra to "unlock" features such as heated seats, etc. But, this may open the door to being able to find workarounds to "unofficially" enable some of these features.

And none of this is really unprecedented on the hardware side at least. For example, when you compare a Ryzen 5900X and a 5950X. They are literally the same, except the 5900X has two cores disabled on each CCD. In many cases that is because some of those cores didn't pass testing, but it's also done sometimes even to a fully functional chip just to fill demand where that demand exists. And we've had 2 decades now where Intel has released chips where basically the only difference is that one has hyper-threading enabled, and the other has it disabled. So what if you had these exact same hardware scenarios, but you could pay to enable those 2-4 extra cores, or pay to enable Hyper-threading? Would that be worse than where we are already at?
 
Honestly, I look forward to it, I buy a fair amount of Xeons and navigating their SKU list for specific features is a PITA, being able to just buy something with a set of the new Sapphire Rapids chips in there and just purchasing some licensing for additional features rather than mixing and matching the actual CPU's is going to be easier for me in the long run and won't really change what I pay so if you are in enterprise procurement this is a win win.
 
I mean here is a list of the current Xeon suffix list that requires thought and mixing and matching and they are all different SKUs for Intel so they have to estimate demand and plan their production schedules accordingly.
https://www.intel.ca/content/www/ca...0059657/processors/intel-xeon-processors.html
Options1st Generation Intel® Xeon® Processors
(formerly Skylake)
2nd Generation Intel® Xeon® Processors
(formerly Cascade Lake)
3rd Generation Intel® Xeon® Processors
(formerly Ice Lake-SP or Cooper Lake)
BN/AMinor electrical specification changes compared to non-B SKU of the same processor number (with no change to core count, frequencies, and features and no impact to platform thermal requirements)N/A
CN/AN/ASingle socket
FSupport for Fabric version of the processor that has an integrated Intel® Omni-Path ConnectorN/AN/A
HCustomized SKU for specific customers; details are Intel ConfidentialN/AProcessors for use in 4- or 8-socket configurations only. These processors were formerly codenamed Cooper Lake. All other 3rd Generation Intel Xeon Scalable Processors (without the H suffix) are formerly codenamed Ice Lake and support 1- or 2-socket configurations. Cooper Lake and Ice Lake processors are not interchangeable within a system.
MLarge Memory Tier (See Memory Capacity table.)N/AProcessor specifications optimized for AI and media processing workloads.
NN/ASpecialized for Networking/Network Function Virtualization workloadsProcessor specifications optimized for communications/networking/NFV (Network Function(s) Virtualization) workloads and operating environments.
PSupport for integrated Intel® FPGA N/AProcessor specifications optimized for IaaS cloud environments such as orchestration efficiency in high-frequency VM environments.
QN/AN/ASupport for a liquid cooling solution. Temperature inlet to cold plate = 40°C. TTV ψca (case-to-fluid inlet resistance) = 0.06°C/W.
RN/ADesignates that a SKU has been refreshed from the prior generationN/A
SN/AOptimizations for Search Engine ProvidersSupport the maximum Intel® Software Guard Extensions (Intel® SGX) enclave capacity of 512GB (per processor) with up to 1TB Intel® SGX enclave capacity in a two-socket configuration. 8368Q also supports up to 512GB Intel® SGX enclave capacity.
TThermal and Long-Life Cycle SupportThermal and Long-Life Cycle SupportSupport for up to 10-year reliability and support for higher Tcase. These SKUs are often used in operating environments with long-life use requirements and require Network Equipment Building System (NEBS)–Thermal friendly specification support.
UN/ASingle Socket OptimizedSupported in one-socket configurations only. 8351N is also supported in a one-socket configuration only.
VN/AVirtual Machine (VM) Density Value SpecializedProcessors specification optimized for SaaS cloud environments.
YN/ASupport for Intel® Speed Select Technology - Performance Profile (Intel® SST-PP)Support for Intel® Speed Select Technology - Performance Profile (Intel® SST-PP) 2.0. In addition, 8352S, 5318S, 5318N, and 8352V SKUs also support Intel® Speed Select Technology – Performance Profile 2.0.
Suffix may follow indicating options



Suffix may follow indicating memory capacity

Memory Capacity1st Generation Intel® Xeon® Processors
(formerly Skylake)
2nd Generation Intel® Xeon® Processors
(formerly Cascade Lake)
3rd Generation Intel® Xeon® Processors
(formerly Cooper Lake - these SKUs have an H suffix)
3rd Generation Intel® Xeon® Processors
(formerly Ice Lake - these SKUs do not have an H suffix)
No SuffixUp to 768 GB per socketUp to 1 TB per socket
(except 9xxxx SKUs that support up to 3 TB per socket)
Up to 1.12 TB per socketUp to 6 TB per socket
LN/AUp to 4.5 TB per socketUp to 4.5 TB per socketN/A
MUp to 1.5 TB per socketN/AN/AN/A

This program replaces that list with a series of license codes to turn things on and off, much easier for everybody involved.
 
I am thinking warez for Intel chips lol.
It's funny but really if I am paying $16,000 for a CPU then a few hundred for turning on Thermal long life support and VM density isn't really going to turn me off.
Now if Intel tries this for Consumer CPU's that isn't going to fly, could you imagine the malware/rootkit potential in that.... Yuck, it gives me chills thinking about that dystopian nightmare.
 
I mean here is a list of the current Xeon suffix list that requires thought and mixing and matching and they are all different SKUs for Intel so they have to estimate demand and plan their production schedules accordingly.
https://www.intel.ca/content/www/ca...0059657/processors/intel-xeon-processors.html
Options1st Generation Intel® Xeon® Processors
(formerly Skylake)
2nd Generation Intel® Xeon® Processors
(formerly Cascade Lake)
3rd Generation Intel® Xeon® Processors
(formerly Ice Lake-SP or Cooper Lake)
BN/AMinor electrical specification changes compared to non-B SKU of the same processor number (with no change to core count, frequencies, and features and no impact to platform thermal requirements)N/A
CN/AN/ASingle socket
FSupport for Fabric version of the processor that has an integrated Intel® Omni-Path ConnectorN/AN/A
HCustomized SKU for specific customers; details are Intel ConfidentialN/AProcessors for use in 4- or 8-socket configurations only. These processors were formerly codenamed Cooper Lake. All other 3rd Generation Intel Xeon Scalable Processors (without the H suffix) are formerly codenamed Ice Lake and support 1- or 2-socket configurations. Cooper Lake and Ice Lake processors are not interchangeable within a system.
MLarge Memory Tier (See Memory Capacity table.)N/AProcessor specifications optimized for AI and media processing workloads.
NN/ASpecialized for Networking/Network Function Virtualization workloadsProcessor specifications optimized for communications/networking/NFV (Network Function(s) Virtualization) workloads and operating environments.
PSupport for integrated Intel® FPGAN/AProcessor specifications optimized for IaaS cloud environments such as orchestration efficiency in high-frequency VM environments.
QN/AN/ASupport for a liquid cooling solution. Temperature inlet to cold plate = 40°C. TTV ψca (case-to-fluid inlet resistance) = 0.06°C/W.
RN/ADesignates that a SKU has been refreshed from the prior generationN/A
SN/AOptimizations for Search Engine ProvidersSupport the maximum Intel® Software Guard Extensions (Intel® SGX) enclave capacity of 512GB (per processor) with up to 1TB Intel® SGX enclave capacity in a two-socket configuration. 8368Q also supports up to 512GB Intel® SGX enclave capacity.
TThermal and Long-Life Cycle SupportThermal and Long-Life Cycle SupportSupport for up to 10-year reliability and support for higher Tcase. These SKUs are often used in operating environments with long-life use requirements and require Network Equipment Building System (NEBS)–Thermal friendly specification support.
UN/ASingle Socket OptimizedSupported in one-socket configurations only. 8351N is also supported in a one-socket configuration only.
VN/AVirtual Machine (VM) Density Value SpecializedProcessors specification optimized for SaaS cloud environments.
YN/ASupport for Intel® Speed Select Technology - Performance Profile (Intel® SST-PP)Support for Intel® Speed Select Technology - Performance Profile (Intel® SST-PP) 2.0. In addition, 8352S, 5318S, 5318N, and 8352V SKUs also support Intel® Speed Select Technology – Performance Profile 2.0.
Suffix may follow indicating options



Suffix may follow indicating memory capacity

Memory Capacity1st Generation Intel® Xeon® Processors
(formerly Skylake)
2nd Generation Intel® Xeon® Processors
(formerly Cascade Lake)
3rd Generation Intel® Xeon® Processors
(formerly Cooper Lake - these SKUs have an H suffix)
3rd Generation Intel® Xeon® Processors
(formerly Ice Lake - these SKUs do not have an H suffix)
No SuffixUp to 768 GB per socketUp to 1 TB per socket
(except 9xxxx SKUs that support up to 3 TB per socket)
Up to 1.12 TB per socketUp to 6 TB per socket
LN/AUp to 4.5 TB per socketUp to 4.5 TB per socketN/A
MUp to 1.5 TB per socketN/AN/AN/A

This program replaces that list with a series of license codes to turn things on and off, much easier for everybody involved.
I always knew Intel sku was convoluted but never realized just how convoluted it is. I can definitely see your point from a procurement perspective that pay for feature proposition is better as there is less headache for everyone involved trying to figure out which sku to get.
 
It's funny but really if I am paying $16,000 for a CPU then a few hundred for turning on Thermal long life support and VM density isn't really going to turn me off.
Now if Intel tries this for Consumer CPU's that isn't going to fly, could you imagine the malware/rootkit potential in that.... Yuck, it gives me chills thinking about that dystopian nightmare.
Definitely, since Intel did try to do that with the Pentium G6951 that led to no where.
 
I always knew Intel sku was convoluted but never realized just how convoluted it is. I can definitely see your point from a procurement perspective that pay for feature proposition is better as there is less headache for everyone involved trying to figure out which sku to get.
Yeah looking at the current lineup there are well over 50 different Xeon SKU's for the current 3'rd gen lineup, this would let Intel replace 50 CPU's with like 10 CPU's of varying core and memory counts, then leave the rest to software unlock.
 
It's funny but really if I am paying $16,000 for a CPU then a few hundred for turning on Thermal long life support and VM density isn't really going to turn me off.
For enterprise, I agree, it would make things much more efficient overall.
It is a slippery slope however...
Now if Intel tries this for Consumer CPU's that isn't going to fly, could you imagine the malware/rootkit potential in that.... Yuck, it gives me chills thinking about that dystopian nightmare.
I warned everyone here of megacorps and the dark cyberpunk future! :borg:
 
I warned everyone here of megacorps and the dark cyberpunk future! :borg:
Not even that, but as it is now it’s stupid hard to not get the general public to not give their banking info to “Microsoft Support” when they call to let you know your computer has a virus. I can only imagine how much of a field day they would have with the general populace having an FPGA they could write to.
 
This is a classic example of an excellent engineering firm, allowing their business end, to make crappy, ridiculous, greedy, marketing/product segmentation decisions. Intel loves doing crap like this.
 
  • Like
Reactions: erek
like this
That's probably the case, but one of the main reasons AMD has been picking up server market share from Intel is by not following them.
Yeah but no, AMD offered a simple yet superior platform for the sub $50,000 market for the past few years. The AMD Epyc platform is the perfect combo of simple and effective especially if you are looking to host 4+ VM’s on a single host. If you are looking to host VM’s AMD is your platform of choice, and that gained them market share.
 
If it's already all on the chip, which is required for this sort of thing to work, I don't see an advantage for the consumer to turning this or that on/off based on licensing, enterprise customer or not. You are always going to pay full price for the chip. The licensing will just be a revenue add for Intel, while cutting their SKU count. The customer is not benefitting from this.
 
If it's already all on the chip, which is required for this sort of thing to work, I don't see an advantage for the consumer to turning this or that on/off based on licensing, enterprise customer or not. You are always going to pay full price for the chip. The licensing will just be a revenue add for Intel, while cutting their SKU count. The customer is not benefitting from this.
That's the neat part! This isn't mean for the consumer. 🤷‍♂️
 
  • Like
Reactions: erek
like this
How many days before that gets hacked?

"Internet connection required during cpu usage to validate license keys"
I can see that coming next, and then KMS servers kind of thing for on prem Enterprises to use that do not allow Inet to servers, and that, as you noted is how it will be gotten around.
 
  • Like
Reactions: erek
like this
It's funny but really if I am paying $16,000 for a CPU then a few hundred for turning on Thermal long life support and VM density isn't really going to turn me off.
Now if Intel tries this for Consumer CPU's that isn't going to fly, could you imagine the malware/rootkit potential in that.... Yuck, it gives me chills thinking about that dystopian nightmare.
It would also happen in the enterprise as well, as it already does around the world. Plenty of countries use pirates software to run entire companies, and they will do the same for CPUs
 
Seems poorly timed given the intensifying competition. This is something you do if you have a monopoly.
 
  • Like
Reactions: erek
like this
I mean here is a list of the current Xeon suffix list that requires thought and mixing and matching and they are all different SKUs for Intel so they have to estimate demand and plan their production schedules accordingly.


This program replaces that list with a series of license codes to turn things on and off, much easier for everybody involved.
I would have never thought about this aspect, makes me think if done correctly(I know I'm dreaming) this could actually be not bad
I always knew Intel sku was convoluted but never realized just how convoluted it is. I can definitely see your point from a procurement perspective that pay for feature proposition is better as there is less headache for everyone involved trying to figure out which sku to get.
This isn't any different than now when we're paying for chips that literally have good silicon lasered off to hit a price/performance mark. Imagine being able to buy a Radeon 7xxx and in the future being able to upgrade to a 7xxx XT with a software code.


I have to agree with the whole mega corps are bad mantra, but unfortunately it takes thousands of people to efficiently build something as complex as a CPU. I just want fast good cheap shit. If something like this helps, and it might, I'm all for it.
 
Yeah but no, AMD offered a simple yet superior platform for the sub $50,000 market for the past few years. The AMD Epyc platform is the perfect combo of simple and effective especially if you are looking to host 4+ VM’s on a single host. If you are looking to host VM’s AMD is your platform of choice, and that gained them market share.
They did way more than that. Take SGX for instance which isolates all of those programs on those Intel VMs. In order to pull that off you have recompile all of your software to use it. AMD includes it for free and it's hardware based. Intel also used to stratify core counts / memory / socket availability which didn't end until Ice Lake. Meaning you want high core count? More memory? Then that's a new socket. AMD didn't do any of that.

I mean you even highlighted that.
I mean here is a list of the current Xeon suffix list that requires thought and mixing and matching and they are all different SKUs for Intel so they have to estimate demand and plan their production schedules accordingly.

Remember Optane? Where Intel would force you into Optane if you wanted more than 1.5TB per socket? All of that is before we even get to TSX, Intel's own personal nightmare. I'm sure all of those Optane users are much happier now.

AMD is a business to be sure and will do what businesses do. AMD offered a simple platform to understand, build for, and support. While Intel did not. But Intel also made their situation waaay worse by doing the aforementioned (and I'm pretty sure I didn't catch all of it either).
 
Last edited:
  • Like
Reactions: erek
like this
That's probably the case, but one of the main reasons AMD has been picking up server market share from Intel is by not following them.
Threadripper Pro would like a word.
 
  • Like
Reactions: erek
like this
Threadripper Pro would like a word.
You want to compare that with workstation xeon have at it. What's the core count those workstation Xeons max out at? That's also not the server market. Even still AMD gave the same core counts in Threadripper as it did in its server parts. Intel did not.
 
Last edited:
You want to compare that with workstation xeon have at it. What's the core count those workstation Xeons max out at? That's also not the server market. Even still AMD gave the same core counts in Threadripper as it did in its server parts. Intel did not.
Re-reading the post I'd replied to a bit flippantly, you're 100% right. My mind was elsewhere when I wrote it - namely the idea that "AMD would not follow Intel". Because what immediately came to mind was how they changed after the success of Zen2 Threadripper, and for Zen3 TR Pro abandoned HEDT/WS/Enthusiast and with really no competition in that market began to act in many ways exactly like Intel would, with the timed-exclusivity lockins (and lockouts), i.e. Lenovo, and leaving AIB partners to bleed out in the snow.

And I don't fault AMD for acting like a business is supposed to - no kidding you sell your silicon for $110 per core to server/datacenter instead of to the market that expects it for $35 per core for the same piece of silicon, even though the latter market helped get them to the favorable market position they found themselves in. It's just that the narrative about AMD being the good guys that would never act like Intel is becoming a little worn, if not detached from reality.

AMD will act every bit like Intel - and Nvidia for that matter - the moment they're in the market position to be. IMO.
 
Last edited:
Re-reading the post I'd replied to a bit flippantly, you're 100% right. My mind was elsewhere when I wrote it - namely the idea that "AMD would not follow Intel". Because what immediately came to mind was how they changed after the success of Zen2 Threadripper, and for Zen3 TR Pro abandoned HEDT/WS/Enthusiast and with really no competition in that market began to act in many ways exactly like Intel would, with the timed-exclusivity lockins (and lockouts), i.e. Lenovo, and leaving AIB partners to bleed out in the snow.

And I don't fault AMD for acting like a business is supposed to - no kidding you sell your silicon for $110 per core to server/datacenter instead of to the market that expects it for $35 per core for the same piece of silicon, even though the latter market helped get them to the favorable market position they found themselves in. It's just that the narrative about AMD being the good guys that would never act like Intel is becoming a little worn, if not detached from reality.

AMD will act every bit like Intel - and Nvidia for that matter - the moment they're in the market position to be. IMO.
How quickly did AMD stop providing the Wraith Prism (good Quality) stock cooler in the box with their high-end processors? About as soon as they had the performance crown.

AMD only plays nice when they are the underdog. Once they're the top-dog, its a dog-eat-dog world.

dogs.
 
Back
Top