Support NeoGAF

[winjer]

Intel 15th-Generation Arrow Lake-S: A Shift Away from Hyper-Threading Technology?

Recent leaks have revealed insights into Intel's 15th-Generation Arrow Lake-S platform, particularly focusing on its CPU architecture and potential shift in multi-threading approach.

www.guru3d.com

Recent leaks have revealed insights into Intel's 15th-Generation Arrow Lake-S platform, particularly focusing on its CPU architecture and potential shift in multi-threading approach. The disclosed information suggests that the Arrow Lake desktop CPUs might not support Hyper-Threading (HT), a notable departure from previous generations.

The technical documentation indicates that Arrow Lake will likely feature eight performance cores without Simultaneous Multi-Threading (SMT) capabilities. This development aligns with prior speculation about Hyper-Threading's exclusion in the Arrow Lake series. The removal of HT could impact the platform's performance in multi-threaded applications, especially compared to the Raptor Lake predecessors. Hyper-Threading generally offers a 10-15% boost in performance for thread-intensive tasks by enabling additional logical cores. However, its absence may not significantly affect gaming performance and could potentially improve frames per second (FPS) in certain games. Despite this, Arrow Lake is expected to aim for a 30% enhancement in gaming performance through architectural improvements.

The rumour of moving away from Hyper-Threading has been circulating at Intel for some time. Intel had considered introducing "Rentable Units" as an alternative, which would provide similar functionality within the CPU but would not be accessible to the operating system. It remains unclear whether Arrow Lake will incorporate these units, as Hyper-Threading was initially planned and reportedly removed at a later stage. The effectiveness of Hyper-Threading has diminished with the advent of many-core processors, yet it still contributes to performance gains in certain benchmarks. In heavily multi-threaded benchmarks, HT can offer a 20-30% performance increase. Conversely, in single-thread dominant tasks, such as many Adobe benchmarks, HT may have minimal or even negative impact, averaging a performance effect of around 10-15%.

In the absence of Hyper-Threading, Arrow Lake's application performance may not significantly surpass that of its predecessor, Raptor Lake. Even with architectural improvements, the lack of HT could negate potential performance gains. Without a suitable alternative like Rentable Units, this could result in a stagnation of application performance. Intel appears to be considering "Rentable Units" as an alternative to traditional Hyper-Threading. This approach, responding to the trend towards hybrid core architectures, involves splitting instruction threads and distributing them across cores based on complexity. Arrow Lake is rumored to feature 3 MB of L2 cache per core and support DDR5-6400 memory, enhancing both performance and efficiency. The below chart is courtesy of 3dcenter.

If this is confirmed, it will be a major shift for Intel CPUs, since most of them have used HT, for the last 20 years.
Why would Intel do this. Maybe they consider that their branch predictor is so good now, that there are not enough unused units left to justify generating a new thread.
Or maybe Intel is banking on the e-cores to have extra threads, while avoiding potential thread contention in the p-cores.

 

[Erebus]

Doesn't HT negatively affect some games or at least is not beneficial at all?

 

[winjer]

Doesn't HT negatively affect some games or at least is not beneficial at all?

No.
It's a mixed bag, with some games improving with it and others losing.

 

[Skifi28]

Next they should start cutting the number of cores, let's go back to two beefy one.

 

[KuraiShidosha]

Next they should start cutting the number of cores, let's go back to two beefy one.

I'd unironically prefer 12 performance cores and 0 junk cores over some 8 core + 24 junk core chip. Waste of silicon.

 

[Bernoulli]

Introducing the new Hyper-Heating inside

 

[Bojji]

Introducing the new Hyper-Heating inside

 

[T4keD0wN]

Or maybe Intel is banking on the e-cores to have extra threads, while avoiding potential thread contention in the p-cores.

Performance cores without SMT, but E cores with it sound ideal to me as long as theres at least 6 P cores or better yet 8 of them. Could be good for productivity efficiency, but it will definitely be better at least for gaming since windows scheduler is too stupid.

 

[foamdino]

Intel 15th-Generation Arrow Lake-S: A Shift Away from Hyper-Threading Technology?

Recent leaks have revealed insights into Intel's 15th-Generation Arrow Lake-S platform, particularly focusing on its CPU architecture and potential shift in multi-threading approach.

www.guru3d.com

If this is confirmed, it will be a major shift for Intel CPUs, since most of them have used HT, for the last 20 years.
Why would Intel do this. Maybe they consider that their branch predictor is so good now, that there are not enough unused units left to justify generating a new thread.
Or maybe Intel is banking on the e-cores to have extra threads, while avoiding potential thread contention in the p-cores.

Thread title says the opposite ;-)

 

[Unknown Soldier]

Thread title says the opposite ;-)

OP was in such a hurry to post that he didn't bother proof reading his thread title

 

[winjer]

OP was in such a hurry to post that he didn't bother proof reading his thread title

Yes, i mistyped the title.
Maybe a moderator can fix it to: "might not have Hyper-Threading"

 

[DaGwaphics]

Interesting. HT is the most useful when you are short on threads, maybe they figure they have so many threads now that it isn't useful.

 

[OverHeat]

I got 16 physical core…fuck HT

 

[Magic Carpet]

Got a 13700F, 8E+8P cores. My first PC to have dedicated potato cores. Whats the best way to know whats going on and what programs are using what, when?

 

[Dr.D00p]

Time to finally announce that Pentium 5, with a single core at 12GHZ!

 

[Nvzman]

I'd unironically prefer 12 performance cores and 0 junk cores over some 8 core + 24 junk core chip. Waste of silicon.

E-Cores are genuinely useful for multitasking, I don't understand why people have this impression that they are "junk" or "fake" because they don't always improve gaming performance.
For things like Discord streaming or even when downloading games on Steam while playing (which can absolutely devour CPU utilization with fast internet) the split setup makes them massively better than having a unified traditional setup because Windows does intelligently allocate all the background work to the E-Cores.
I noticed absolutely massive gains going from a 10900k to a 13700k when doing multitasking work while gaming or rendering videos, I know it's a three gen jump but the fact that the E-Cores are doing 100% of the background work while the P-cores are handling whatever I'm playing makes it essentially butter smooth and stutter-free, which was never the case in my experience prior.

 

[ReBurn]

Intel 15th-Generation Arrow Lake-S: A Shift Away from Hyper-Threading Technology?

Recent leaks have revealed insights into Intel's 15th-Generation Arrow Lake-S platform, particularly focusing on its CPU architecture and potential shift in multi-threading approach.

www.guru3d.com

If this is confirmed, it will be a major shift for Intel CPUs, since most of them have used HT, for the last 20 years.
Why would Intel do this. Maybe they consider that their branch predictor is so good now, that there are not enough unused units left to justify generating a new thread.
Or maybe Intel is banking on the e-cores to have extra threads, while avoiding potential thread contention in the p-cores.

I think you answered your own question as to why Intel may change the threading model. After 20 years it's not unreasonable to expect that progression of CPU technology would require rethinking how performance is managed. The current hyperthreading could be holding the tech back.

I would be surprised if they completely abandon the current model as software will need time to adapt. But there has to come a time when fiddling with cores just produces diminishing returns.

 

[Kenpachii]

E-Cores are genuinely useful for multitasking, I don't understand why people have this impression that they are "junk" or "fake" because they don't always improve gaming performance.
For things like Discord streaming or even when downloading games on Steam while playing (which can absolutely devour CPU utilization with fast internet) the split setup makes them massively better than having a unified traditional setup because Windows does intelligently allocate all the background work to the E-Cores.
I noticed absolutely massive gains going from a 10900k to a 13700k when doing multitasking work while gaming or rendering videos, I know it's a three gen jump but the fact that the E-Cores are doing 100% of the background work while the P-cores are handling whatever I'm playing makes it essentially butter smooth and stutter-free, which was never the case in my experience prior.

Went from 9900k towards a 12900 and i noticed massive gains also because all the windows junk and stuff u do on the background gets offloaded to non gaming cores. It's great.

 

[winjer]

I think you answered your own question as to why Intel may change the threading model. After 20 years it's not unreasonable to expect that progression of CPU technology would require rethinking how performance is managed. The current hyperthreading could be holding the tech back.

I would be surprised if they completely abandon the current model as software will need time to adapt. But there has to come a time when fiddling with cores just produces diminishing returns.

The debate between using SMT technologies vs better front-ends, as been going on for a while.
Branch predictors have become very good. At the same time, the width of CPU pipelines has not increased that much.
So SMT has fewer opportunities to enter and increase performance.
It would be interesting to see CPUs with wider and wider pipelines, but I guess that means too much trouble for little gains.

 

[Skifi28]

Time to finally announce that Pentium 5, with a single core at 12GHZ!

 

[MayauMiao]

If it helps with smooth performance I'm all for it.

But let's wait what the benchmark will say.

 

[violence]

Keep your lousy string theory out of my silicon.

 

[LordOfChaos]

Might to have

Does it have, or does it haven't?

Jk. This seems to be preparing for Jim Keller's Royal Core, with a next gen method named rentable units that should be much better than hyperthreading. 15th gen seems to be the first glimpses of that coming down the pipeline, removing the legacy Hyperthreading to be ready for Rentable Units in a next core.