Intel Software Defined silicon coming soon to a Linux Distro near you!

L

Lakados

[H]F Junkie
Joined
Feb 3, 2014
Messages
10,361
https://www.phoronix.com/scan.php?page=news_item&px=Intel-SDSi-Linux-5.18

Intel Software Defined Silicon Planned For Integration In Linux 5.18

For those who don't follow this the Intel Software Defined Silicon Program is designed to let them reduce the number of Xeon SKU's that Intel offers by instead offering a base model, then you can get modules installed on it that then give you the desired functions.

So for say their Gold lineup of Xeons they no longer have to offer the L, M, N, S, T, V, and Y variants of each chip and instead use some sort of FPGA built into the new sapphire rapids cores add on the desired features and provide updates to them through some sort of subscription service.

A neat idea, it makes me feel a smidge weird, but thinking about it the stuff that they could do that makes me weirded out would get them sued so fast that it's not a real concern.
ex. They pull a feature you've been using for years and paying the sub for and rely on, but really the people buying Xeon Golds have the cash to make Intel pay hard for that sort of transgression.
 
I'm excited to see fpgas on server silicon. Although I think intel will abuse this from a listening perspective the ability to develop software to use the programmable reasorses would be very powerful.

I feel amd will be taking a similar party with their recent accusation of xilinx, and honestly I have more hope the amd solution will be open enough to dictate the use of the fpga however you want.
 
cdabc123 said:
I'm excited to see fpgas on server silicon. Although I think intel will abuse this from a listening perspective the ability to develop software to use the programmable reasorses would be very powerful.

I feel amd will be taking a similar party with their recent accusation of xilinx, and honestly I have more hope the amd solution will be open enough to dictate the use of the fpga however you want.
Click to expand...
I'm hoping it leads to a release of a PCIe5 add-on card running a new class of FPGA on their 7nm process, that would be way more fun to play with and less dangerous than working on a production CPU, would hate to be the guy responsible for bricking one of those with a bad logic patch.
 
Format _C: said:
So this is like a CPU with a built in conditional access modules so you can buy more features I say LoL to all of this!
Click to expand...
Well, up there Intel operates something like 30 SKU's in various configurations of their feature sets.
  • L- Large DDR Memory Support (up to 4.5TB)
  • M- Medium DDR Memory Support (up to 2TB)
  • N- Networking/Network Function Virtualization
  • S- Search
  • T- Thermal
  • V- VM Density Value
  • Y- Intel Speed Select Technology
Then they offer those variations for each core count and CPU speed selection, it makes their SKU list huge and adds to cost, so if this lets them simplify the stack and go from 30 down to 3 where then people can pay an extra few hundred for the specific features you need then it should end up cheaper for everybody. And if that includes added support as memory controller instructions get better of improved search algorithms are released it ends up as kind of a cool thing. I mean when you are dropping $15,000 per CPU and sticking as many as 8 of them into one box, then buying god only knows how many of those boxes, that adds up to some pretty big savings for everybody involved.

Is it a weird concept for hardware, I mean sort of, we've been fine buying feature-locked hardware for decades having the ability to remove unneeded features to add more relevant ones or improve the ones we are using for me is sort of a welcome upgrade.
 
Lakados said:
I'm hoping it leads to a release of a PCIe5 add-on card running a new class of FPGA on their 7nm process, that would be way more fun to play with and less dangerous than working on a production CPU, would hate to be the guy responsible for bricking one of those with a bad logic patch.
Click to expand...
I would be more excited about a large one on chip. A pcie card is nice due to large tdp limits and normally quick memory (ddr4 most of the time) however a on chip one would already have access to a large memory pool as well as fast interconnects to all perpetuals. In addition it would have extremely quick access to a beefy x86 CPU and a cache pool. Both of which is an advantage from any fpga on the market. (Most use small arm cores)

This would also force intels hand to push the fpgas on the most modern lithography. Unless they go with a weird soc design
 
Lakados said:
it should end up cheaper for everybody.
Click to expand...
1644541108396.jpeg
 
cdabc123 said:
I would be more excited about a large one on chip. A pcie card is nice due to large tdp limits and normally quick memory (ddr4 most of the time) however a on chip one would already have access to a large memory pool as well as fast interconnects to all perpetuals. In addition it would have extremely quick access to a beefy x86 CPU and a cache pool. Both of which is an advantage from any fpga on the market. (Most use small arm cores)

This would also force intels hand to push the fpgas on the most modern lithography. Unless they go with a weird soc design
Click to expand...
Yeah the DE10 and the Cyclone V stuff is oooold.
 
Lakados said:
Yeah the DE10 and the Cyclone V stuff is oooold.
Click to expand...
Yes but old fpgas will always be the butter of the enthusiast market. Intel is the only one capable of pushing consumer fpgas to a more advanced node but honestly they shouldn't. With the demand of modern silicon it wouldn't be to profitable to push the small fpgas onto these desirable nodes.

Maybe pushing fpgas to cpus will allow them to make them relatively affordable (enterprise CPU pricing that may trickle to the used market eventually)

A modern lithography fpga on a pcie card will always be several grand.
 
cdabc123 said:
Yes but old fpgas will always be the butter of the enthusiast market. Intel is the only one capable of pushing consumer fpgas to a more advanced node but honestly they shouldn't. With the demand of modern silicon it wouldn't be to profitable to push the small fpgas onto these desirable nodes.

Maybe pushing fpgas to cpus will allow them to make them relatively affordable (enterprise CPU pricing that may trickle to the used market eventually)

A modern lithography fpga on a pcie card will always be several grand.
Click to expand...
Maybe??? But Intel has a lot of space on 14nm, their Cyclone FPGA’s are still on 28nm. Even moving from 28 to 14 would improve yields, use fewer materials, and improve power/performance blah blah blah..
 
Lakados said:
Maybe??? But Intel has a lot of space on 14nm, their Cyclone FPGA’s are still on 28nm. Even moving from 28 to 14 would improve yields, use fewer materials, and improve power/performance blah blah blah..
Click to expand...
Maybe, only because this will be a enterprise product with little relevant competition. Intel wont make it cheap they will be gold/platinum xeons with a premium tacked on. If amd competes in this space successfully, then we may see some that are obtainable to most intrested consumers.

I'm not sure how much it would take to throw them on 14nm, Intel seams more interested in keeping the fpgas on the easily customizable nodes. Personally I think intel has been mismanaging altera as there is a huge potential if they properly integrate fpgas with their available nodes and engineering resources.

Amd cannot afford to do the same with xilinx so hopefully they release some interesting and properly integrated products (although they arnt in the same fab situation as Intel)
 
I think these programmable circuits are going to be the future. And when you factor in things like software defined networking .... dang ..

Remember in TNG when they are rerouting power and building new connections just from their console ... that's this.
 
cdabc123 said:
Maybe, only because this will be a enterprise product with little relevant competition. Intel wont make it cheap they will be gold/platinum xeons with a premium tacked on. If amd competes in this space successfully, then we may see some that are obtainable to most intrested consumers.

I'm not sure how much it would take to throw them on 14nm, Intel seams more interested in keeping the fpgas on the easily customizable nodes. Personally I think intel has been mismanaging altera as there is a huge potential if they properly integrate fpgas with their available nodes and engineering resources.

Amd cannot afford to do the same with xilinx so hopefully they release some interesting and properly integrated products (although they arnt in the same fab situation as Intel)
Click to expand...
I would love to see Intel or AMD one day release an FPGA console. Imagine where developers could customize the hardware their games were running on… it would be the future of Arcades past.
 
Lakados said:
And if that includes added support as memory controller instructions get better of improved search algorithms are released it ends up as kind of a cool thing.
Click to expand...
When was the last time you got a new microcoded instruction after the CPU shipped? One that wasn't just oh crap, this doesn't work so we're going to turn it off? I highly doubt this is going to deliver features developed after the die was finalized. It's very likely to be used to turn on and off features that are already on the die, so they don't have to blow fuses at the factory.
 
toast0 said:
When was the last time you got a new microcoded instruction after the CPU shipped? One that wasn't just oh crap, this doesn't work so we're going to turn it off? I highly doubt this is going to deliver features developed after the die was finalized. It's very likely to be used to turn on and off features that are already on the die, so they don't have to blow fuses at the factory.
Click to expand...
It depends. Imagine a CPU with accessible fpga resources (abit different then what was proposed in this article) that would absolutely be optimized with the software as time progresses.
 
toast0 said:
When was the last time you got a new microcoded instruction after the CPU shipped? One that wasn't just oh crap, this doesn't work so we're going to turn it off? I highly doubt this is going to deliver features developed after the die was finalized. It's very likely to be used to turn on and off features that are already on the die, so they don't have to blow fuses at the factory.
Click to expand...
That’s where the stuff they posted to GitHub gets me, yes it certainly uses license keys to either turn on or off features. But it also looks like it can be used to deliver new software packages.

The SDSi looks to be able to activate code with a license key, deliver new code, and read the status of the existing code.

https://github.com/intel/intel-sdsi
 
Maybe someone will figure out a way to hack the shit out of those and force uploads of task-specific microcode?
Which might end up in Xeons being used in mining and as a result costing as much as a decent car?
 
michalrz said:
Maybe someone will figure out a way to hack the shit out of those and force uploads of task-specific microcode?
Which might end up in Xeons being used in mining and as a result costing as much as a decent car?
Click to expand...
A fpga on a CPU would be good for mining :p even better if the include some on chip memory (like the intel phi processors)
 
michalrz said:
Maybe someone will figure out a way to hack the shit out of those and force uploads of task-specific microcode?
Which might end up in Xeons being used in mining and as a result costing as much as a decent car?
Click to expand...
Based on the pricing of their predecessors they already do.
 
This is so many stupids I can't even imagine how this will turn out. This isn't different from Tesla locking away hardware features on hardware you own. Better yet, this isn't different from when ATI locked away pipelines on their Radeon 9700 cards or even Nvidia doing the same on their Geforce 6800 cards. You know why they stopped doing this nearly two decades ago? Because they software locked it and as we all know software locking works, amiright?

Sounds like they're keeping this to Xeons for the moment. Probably because nobody who is going to spend thousands on server hardware is going to download some firmware to flash to unlock all the features. When they do bring this to end user hardware then that's when you'll see people finding ways to hack it and unlock all the features. Never mind that Intel is doing this because they'll claim copyright on the software and prevent people legally from messing around with the hardware they own. Because the copyright system is absolutely broken and auto manufactures as well as Intel are now taking advantage of it. All because Intel doesn't want to make one single product because they want to sell you tiers of hardware to maintain price control. It is literally cheaper for them to make one hardware than to make multiple hardware. Savings they could pass to you but won't.
 
DukenukemX said:
This is so many stupids I can't even imagine how this will turn out. This isn't different from Tesla locking away hardware features on hardware you own. Better yet, this isn't different from when ATI locked away pipelines on their Radeon 9700 cards or even Nvidia doing the same on their Geforce 6800 cards. You know why they stopped doing this nearly two decades ago? Because they software locked it and as we all know software locking works, amiright?

Sounds like they're keeping this to Xeons for the moment. Probably because nobody who is going to spend thousands on server hardware is going to download some firmware to flash to unlock all the features. When they do bring this to end user hardware then that's when you'll see people finding ways to hack it and unlock all the features. Never mind that Intel is doing this because they'll claim copyright on the software and prevent people legally from messing around with the hardware they own. Because the copyright system is absolutely broken and auto manufactures as well as Intel are now taking advantage of it. All because Intel doesn't want to make one single product because they want to sell you tiers of hardware to maintain price control. It is literally cheaper for them to make one hardware than to make multiple hardware. Savings they could pass to you but won't.
Click to expand...
How much time have you spent working with large virtualized Xeon servers?

This is going to simplify things a fair bit take this example.

You upgrade your Sharepoint and Exchange CU’s it gets you all the needed updated security but now search performance sucks. Looking over the logs your performance in most of your metrics is fine so you decide you can swap out one of the N series Xeons for an S series.

$30,000 and a few hours of downtime later the server is back up and running and everybody is happy.

With the new Xeons it’s a phone call, an email, a $2000 license and a reboot.

I mean yeah it would be great if they included all the features on all the chips but they don’t and so far they haven’t. I’m hoping AMD can challenge them enough that they have to start but sadly it seems to be the opposite with AMD starting to segment their offerings for the larger EPYC systems too.
 
I wonder how much they would charge to unlock the multiplier on the 56-core? I'll bet someone out there would pay a ridiculous figure for that.
 
JSHamlet234 said:
I wonder how much they would charge to unlock the multiplier on the 56-core? I'll bet someone out there would pay a ridiculous figure for that.
Click to expand...
Well it’s only available in the Platinum and Gold series chips which are expensive to begin with but probably a few hundred extra for sure.
 
DukenukemX said:
This is so many stupids I can't even imagine how this will turn out. This isn't different from Tesla locking away hardware features on hardware you own. Better yet, this isn't different from when ATI locked away pipelines on their Radeon 9700 cards or even Nvidia doing the same on their Geforce 6800 cards. You know why they stopped doing this nearly two decades ago? Because they software locked it and as we all know software locking works, amiright?

Sounds like they're keeping this to Xeons for the moment. Probably because nobody who is going to spend thousands on server hardware is going to download some firmware to flash to unlock all the features. When they do bring this to end user hardware then that's when you'll see people finding ways to hack it and unlock all the features. Never mind that Intel is doing this because they'll claim copyright on the software and prevent people legally from messing around with the hardware they own. Because the copyright system is absolutely broken and auto manufactures as well as Intel are now taking advantage of it. All because Intel doesn't want to make one single product because they want to sell you tiers of hardware to maintain price control. It is literally cheaper for them to make one hardware than to make multiple hardware. Savings they could pass to you but won't.
Click to expand...
Those last two sentences are all this is about, and is very much the bottom line.
This is good for Intel, who will save massively on production costs, and will now charge more and more down the road for feature-set licensing to their customers, vendor-locked or not.

More anti-customer licensing and Corporatist propaganda from Intel.
Anyone who says this will be good for customers needs to look up the definition of "useful idiot".
 
Lakados said:
$30,000 and a few hours of downtime later the server is back up and running and everybody is happy.

With the new Xeons it’s a phone call, an email, a $2000 license and a reboot.
Click to expand...
If this were literally all the farther Intel took this new licensing, and it were still 200X, it would be good feature - assuming it is optional, which it won't be.
However, this is Intel we are talking about in 202X, and they will eventually push this garbage licensing down to the lowest tablet SoC, and it isn't an 'if', it's a 'when'.

The only thing this benefits is Intel; prepare for vendor lock-in in ways no one could have foreseen. :borg:
 
Red Falcon said:
If this were literally all the farther Intel took this new licensing, and it were still 200X, it would be good feature - assuming it is optional, which it won't be.
However, this is Intel we are talking about in 202X, and they will eventually push this garbage licensing down to the lowest tablet SoC, and it isn't an 'if', it's a 'when'.

The only thing this benefits is Intel; prepare for vendor lock-in in ways no one could have foreseen. :borg:
Click to expand...
So far all they’ve really mentioned is the ability to unlock the variants. If they added that ability on the lower variants that would be nice for smaller single or dual socket VM hosts (which tend to not have any of those options). It’s going to depend on how AMD positions the new EPYCs and how they perform, though their performance and feature set means very little if they are essentially unavailable. I have a first gen 7551P that’s nearing retirement and I don’t want to be given another 8 month delivery time when I finally get funding approval to do so because it screws up my departments numbers until it arrives and is online.

Though it would be super awesome if they allowed that to be user addressable like their other FPGA devices, even if not ultra powerful having the ability to deploy custom solutions on there would be dope like custom encryption modules, or job specific accelerators.
 
The actual issue leads to an interesting but very long discussion, and one that is not new. The usual arguments in support of market segmentation and so forth are expected. Perspectives vary, but predictably reveal the flavor of propaganda that one has been exposed to, and how much of it has been swallowed.

From the perspective of an individual with an interest in technology? Fuck DRM in all its forms.
 
ElementDave said:
The actual issue leads to an interesting but very long discussion, and one that is not new. The usual arguments in support of market segmentation and so forth are expected. Perspectives vary, but predictably reveal the flavor of propaganda that one has been exposed to, and how much of it has been swallowed.

From the perspective of an individual with an interest in technology? Fuck DRM in all its forms.
Click to expand...

From the perspective of an individual with an interest in technology? Im happy to see effective utilation of programable silicon on a cpu. Of course, intel will take plenty of profit in the enterprise segment first however they will eventually face pressure from amd for user accessable fgpa resources on chip. My guess is it will work its way into the user adressable range in <5years. Intel has already released fpgas on intel gold cpus this is just the marketing department having abit of fun with the ability to do so. Abit of competition will push products into the more rational use of this concept.
 
Lakados said:
How much time have you spent working with large virtualized Xeon servers?

This is going to simplify things a fair bit take this example.

You upgrade your Sharepoint and Exchange CU’s it gets you all the needed updated security but now search performance sucks. Looking over the logs your performance in most of your metrics is fine so you decide you can swap out one of the N series Xeons for an S series.

$30,000 and a few hours of downtime later the server is back up and running and everybody is happy.

With the new Xeons it’s a phone call, an email, a $2000 license and a reboot.

I mean yeah it would be great if they included all the features on all the chips but they don’t and so far they haven’t. I’m hoping AMD can challenge them enough that they have to start but sadly it seems to be the opposite with AMD starting to segment their offerings for the larger EPYC systems too.
Click to expand...
You're obviously happy for paying for features you already have. Like I said, for Xeons it makes sense for Intel to get away with this because as you pointed out you're happy to pay $2000 for what could be a $30k loss in downtime. Ain't nobody who's buying a Xeon is going to use some Russian firmware that hacks and unlocks all the features. Intel could just make one chip and just sell it to you cheaper because that's what they're doing anyway, but now features are hidden behind micro-transactions. My point is that eventually Intel will bring this to their end user products and expect nobody to download that Russian firmware. But they will, because nobody cares about security when they can run games with extra frame rates and pay less for a CPU. It's been done before with Intel and AMD. There's a reason why they laser cut chips and not limit the features with software anymore.
 
DukenukemX said:
You're obviously happy for paying for features you already have. Like I said, for Xeons it makes sense for Intel to get away with this because as you pointed out you're happy to pay $2000 for what could be a $30k loss in downtime. Ain't nobody who's buying a Xeon is going to use some Russian firmware that hacks and unlocks all the features. Intel could just make one chip and just sell it to you cheaper because that's what they're doing anyway, but now features are hidden behind micro-transactions. My point is that eventually Intel will bring this to their end user products and expect nobody to download that Russian firmware. But they will, because nobody cares about security when they can run games with extra frame rates and pay less for a CPU. It's been done before with Intel and AMD. There's a reason why they laser cut chips and not limit the features with software anymore.
Click to expand...
If intel brings that shit to the consumer level. I will be the first to start throwing hands, that is an enterprise need not a consumer one, though now that you say it I can totally see them using it to make it so any chip for an extra $50 can be a K series and that made me throw up a little in my mouth.

But sweet Jesus the chaos it could become if it was released at the consumer level, I mean the "Microsoft" support calls are bad enough when they were just installing the shit they were already installing, can you imagine if they figured out how to install their own custom packages to the FPGA that would be a nightmare.
 
Lakados said:
How much time have you spent working with large virtualized Xeon servers?

This is going to simplify things a fair bit take this example.

You upgrade your Sharepoint and Exchange CU’s it gets you all the needed updated security but now search performance sucks. Looking over the logs your performance in most of your metrics is fine so you decide you can swap out one of the N series Xeons for an S series.

$30,000 and a few hours of downtime later the server is back up and running and everybody is happy.

With the new Xeons it’s a phone call, an email, a $2000 license and a reboot.

I mean yeah it would be great if they included all the features on all the chips but they don’t and so far they haven’t. I’m hoping AMD can challenge them enough that they have to start but sadly it seems to be the opposite with AMD starting to segment their offerings for the larger EPYC systems too.
Click to expand...
Or, more likely, you will overpay for the features you truly need for the never needed option of unlocking features you don't.

I've already puked enough when Intel sold its servers with built in IPMI features disabled with the option to pay to unlock them. Something my puny Supermicro and ASRock Rack boards have by default. You can bet Intel didn't pass that saving to the customer.
 
Last edited:
Lakados said:
If intel brings that shit to the consumer level. I will be the first to start throwing hands, that is an enterprise need not a consumer one, though now that you say it I can totally see them using it to make it so any chip for an extra $50 can be a K series and that made me throw up a little in my mouth.

But sweet Jesus the chaos it could become if it was released at the consumer level, I mean the "Microsoft" support calls are bad enough when they were just installing the shit they were already installing, can you imagine if they figured out how to install their own custom packages to the FPGA that would be a nightmare.
Click to expand...
One thing worth mentioning... what happens when Intel decides to put a kill-date on the licensing?
Those Xeons with all of the features are going to work great until the license date ends - perhaps this is fine for enterprise, but then lets say said enterprise decides to sell the servers or donate them, and the Intel CPU license kill date is soon after they sell them.

I guess those CPUs and server platforms will be bricked, or have next to no value with minimal functionality.
That's not a problem for the new buyer or donation receptor of the used hardware, they can just buy a new license for those CPUs - oh wait, Intel doesn't support licenses for those CPUs any longer, and they are truly bricked and/or removed of all features and value.

Guess those perfectly good used servers are now going in the garbage and will become more e-waste - but Intel is green and totally cares about the environment over profits, riiiight.
This is pure Corporatism by Intel and screws everyone involved, and as you said, just imagine the chaos if this gets down to consumer-level products...

Suddenly ARM, RISC-V, and every other pre-licensed-enabled x86-64 CPU is going to become extremely attractive and valuable, not to mention rare, like no other.
This isn't Cyberpunk 2020, this is Cyberpunk RED 20 years ahead of schedule. :dead:
 
You must log in or register to reply here.
Back
Top