Intel's "Software Defined Silicon" Linux Support Moving Along

[phoronix]

Phoronix: Intel's "Software Defined Silicon" Linux Support Moving Along

While Intel has not publicly announced their plans around Software Defined Silicon (SDSi), the Linux kernel patches allowing activation of licensed CPU features is continuing to move forward toward mainline integration...

https://www.phoronix.com/scan.php?page=news_item&px=Intel-SDSi-Linux-2022

[uid313]

With people sitting on perfectly good computers that they throw away for no good reason and the environmental impact this has, I hope that Intel gives away free updates after some years for old hardware.

[Papicha]

Software-brickable Silicon would be a better name.

[Anux]

I hope that Intel gives away free updates after some years for old hardware.

Uh, any action of Intel in the past that ignites your hopes? I think thats highly unlikley.

Imagin this patch and something like Meltdown leading to millions of low priced full featured CPUs and Intel not getting a penny of it. That gloat ...

[skeevy420]

With people sitting on perfectly good computers that they throw away for no good reason and the environmental impact this has, I hope that Intel gives away free updates after some years for old hardware.

A lot of times the "good reason" is because they're reached the limits of their hardware and can't upgrade it any further. It sucks when your ram and hdd are soldered to the board. It doesn't matter how perfectly good a computer is if it doesn't do what the user wants it to do and the user is unable to upgrade it to be capable for their needs.

Just look at how many perfectly good Skylake PCs are going to be trashed just because of Windows 11 and TPM. This shit is infuriating

[M@GOid]

I'm hopping so bad this thing backfires and gets cracked fast. Just to make clear this is not the path the industry should take.

[skeevy420]

I'm hopping so bad this thing backfires and gets cracked fast. Just to make clear this is not the path the industry should take.

What I imagined when I read your post:


Last week's BoBF was one of the best Mando episodes ever
Kind of a crappy Boba episode

[ermo]

This feels like something you'd use in a leasing scenario as a cloud provider? Buy a bunch of CPUs and then pass on the unlocking cost to the consumers wanting dedicated hardware with feature X enabled?

[Solid State Brain]

On a superficial look this feature looks like a very bad deal for end-users, but it's not that product segmentation with modern CPUs is much more ethical. CPU features or also cores for lower models are often "fused out" (i.e. disabled) at the hardware level from higher models, but their transistors still physically exist in the silicon and therefore in theory could still work.

[skeevy420]

On a superficial look this feature looks like a very bad deal for end-users, but it's not that product segmentation with modern CPUs is much more ethical. CPU features or also cores for lower models are often "fused out" (i.e. disabled) at the hardware level from higher models, but the transistors allowing them to work still physically exist in the silicon and therefore in theory could still work.

I've heard a lot of reasoning behind that is because it's easier to create the high core count model and then disable any cores that have failures, don't pass QA tests, and call it a lower spec model. While it'd be nice to enable a feature to roll the dice on how well the disabled cores work, because we'd be rolling the dice is precisely why they don't want it to happen and won't offer it unless they're pressured to do so by governments.

That said, if they sold CPUs with the QA results, somehow embedded QA results into the CPU itself, we'd at least know if the cores are disabled because they can't run faster than 2.2GHz or because certain instructions don't work and then tweak our schedulers and software accordingly after enabling what was disabled for stability reasons.