Author Topic: Intel: software-defined supercores  (Read 1579 times)

0 Members and 2 Guests are viewing this topic.

Offline madiresTopic starter

  • Super Contributor
  • ***
  • Posts: 9270
  • Country: de
  • A qualified hobbyist ;)
Intel: software-defined supercores
« on: September 01, 2025, 11:24:37 am »
Media:
- Intel plans virtual "super processor cores" - https://www.heise.de/en/news/Intel-plans-virtual-super-processor-cores-10626874.html
- Intel Patents Software-Defined Super Cores to Maximize Single-Core Performance - https://www.techpowerup.com/340508/intel-patents-software-defined-super-cores-to-maximize-single-core-performance

Sounds like Intel wants to drop the P cores and go for supercore capable E cores only, at least for some CPU models. Maybe for mobile CPUs and inexpensive models for office PCs? One benefit is reduced power consumption. But what happens when you run a singlethreaded application without the additional 'hints' for super cores? And multithreaded applications can use multiple cores anyway. E cores need less die area than P cores. So Intel could replace one P core with possibly four supercore capable E cores. Based on the applications you run there might be a minor benefit or even a drawback. Therefore I'm sceptical. Something like a 32 core ARM comes to mind. What do you think?
 

Offline SiliconWizard

  • Super Contributor
  • ***
  • Posts: 17840
  • Country: fr
Re: Intel: software-defined supercores
« Reply #1 on: September 01, 2025, 02:26:38 pm »
I don't know the details of implementation, but using several cores to emulate one more powerful core is never going to give you the same amount of performance, you'll still have the bottleneck issues of spreading the workload across two cores. Maybe they have devised a way of distributing instructions across several cores, like a big out-of-order controller that feeds several cores instead of several units inside a single core. I don't know. If not, it sounds like just multi-core and I wouldn't know what is worth patenting. Is the patent available yet?
 

Offline ejeffrey

  • Super Contributor
  • ***
  • Posts: 4865
  • Country: us
Re: Intel: software-defined supercores
« Reply #2 on: September 01, 2025, 04:11:01 pm »
It's always worth reading patents skeptically.  This could be something that Intel is already building into a product, or it could be a neat idea some engineer had.

It sounds to me a bit like inverse hyper threading / SMT.  Instead of making a big core that can act as two virtual cores, you have mostly independent cores that can share a few backend execution units.

The article specifically calls out SIMD instructions.  I could see this making sense: maybe you have a two cores each with a single 256 but SIMD ALU, but you can reconfigure that to give one core a wider 512 bit or a pair of 256 bit ALUs.  That could boost single thread performance at least for certain workloads. 

I'm not sure how the hints would work, or how big the benefit would be.  I think most applications that benefit from super wide SIMD can also be made multi threaded.


« Last Edit: September 01, 2025, 04:12:32 pm by ejeffrey »
 

Offline Someone

  • Super Contributor
  • ***
  • Posts: 6124
  • Country: au
    • send complaints here
Re: Intel: software-defined supercores
« Reply #3 on: September 02, 2025, 12:04:11 am »
It's always worth reading patents skeptically.  This could be something that Intel is already building into a product, or it could be a neat idea some engineer had.

It sounds to me a bit like inverse hyper threading / SMT.  Instead of making a big core that can act as two virtual cores, you have mostly independent cores that can share a few backend execution units.

The article specifically calls out SIMD instructions.  I could see this making sense: maybe you have a two cores each with a single 256 but SIMD ALU, but you can reconfigure that to give one core a wider 512 bit or a pair of 256 bit ALUs.  That could boost single thread performance at least for certain workloads. 

I'm not sure how the hints would work, or how big the benefit would be.  I think most applications that benefit from super wide SIMD can also be made multi threaded.
Back to the future, like early multicore AMD designs.
 

Online magic

  • Super Contributor
  • ***
  • Posts: 8072
  • Country: pl
Re: Intel: software-defined supercores
« Reply #4 on: September 02, 2025, 08:32:06 am »
More like Bulldozer modules, AMD's earlier CPUs had straightforward independent cores.
 
The following users thanked this post: Someone

Offline tszaboo

  • Super Contributor
  • ***
  • Posts: 9917
  • Country: nl
  • Current job: ATEX product design
Re: Intel: software-defined supercores
« Reply #5 on: September 02, 2025, 11:58:58 am »
I think it's a great idea, and it can really be a game changer. The E cores today have the same performance as processors a decade ago, like an E core only N150 will have similar performance as a i5 4570 from 2013.
Only the N15 does it from like 5W versus 50W.
So if they can run these 1-2W consuming cores, and run for example 4 of them to add even just 60% single threaded performance, would be a very good tradeoff.
Measure power rails with 2GHz bandwidt: https://eleshop.eu/prp1-power-rail-probe.html
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf

 

- - -
-->