Intel 15th-Generation Arrow Lake-S Could Abandon Hyper-Threading Technology

[AleksandarK]

A leaked Intel documentation we reported on a few days ago covered the Arrow Lake-S platform and some implementation details. However, there was an interesting catch in the file. The leaked document indicates that the upcoming 15th-Generation Arrow Lake desktop CPUs could lack Hyper-Threading (HT) support. The technical memo lists Arrow Lake's expected eight performance cores without any threads enabled via SMT. This aligns with previous rumors of Hyper-Threading removal. Losing Hyper-Threading could significantly impact Arrow Lake's multi-threaded application performance versus its Raptor Lake predecessors. Estimates suggest HT provides a 10-15% speedup across heavily-threaded workloads by enabling logical cores. However, for gaming, disabling HT has negligible impact and can even boost FPS in some titles. So Arrow Lake may still hit Intel's rumored 30% gaming performance targets through architectural improvements alone.

However, a replacement for the traditional HT is likely to come in the form of Rentable Units. This new approach is a response to the adoption of a hybrid core architecture, which has seen an increase in applications leveraging low-power E-cores for enhanced performance and efficiency. Rentable Units are a more efficient pseudo-multi-threaded solution that splits the first thread of incoming instructions into two partitions, assigning them to different cores based on complexity. Rentable Units will use timers and counters to measure P/E core utilization and send parts of the thread to each core for processing. This inherently requires larger cache sizes, where Arrow Lake is rumored to have 3 MB of L2 cache per core. Arrow Lake is also noted to support faster DDR5-6400 memory. But between higher clocks, more E-cores, and various core architecture updates, raw throughput metrics may not change much without Hyper-Threading.



View at TechPowerUp Main Site | Source

 

[Onasi]

Interesting. I assume the Rentable Units approach, if implemented, will require yet another overhaul to Windows thread scheduling to work properly? Seems we’ve been getting those rather frequently lately.
Honestly, getting rid of HT seems weird. It doesn’t really have any major downsides and leaving any performance on the table is unlike Intel. Unless they are THAT sure that the new approach will compensate and more?

 

[R-T-B]

It doesn’t really have any major downsides

From a hardware security perspective it's sort of been a nightmare for them.

 

[Onasi]

From a hardware security perspective it's sort of been a nightmare for them.

That is true, actually. Wasn’t that one of the benefits of AMDs SMT implementation, less potential holes? Still, I assume this will be something for the consumer space specifically and the architectures will diverge even further with regards to the server/workstation platforms. I mean… they can’t really reasonably abandon HT there, not with AMD currently trouncing them on thread count, surely?

 

[R-T-B]

Wasn’t that one of the benefits of AMDs SMT implementation, less potential holes?

It's not been perfectly kind to amd either. Sharing a core is just prone to data leakage, architecturally speaking.

 

[R0H1T]

I mean… they can’t really reasonably abandon HT there, not with AMD currently trouncing them on thread count, surely?

No this is exactly for that reason, security on (shared) or private servers is a major issue. No one gives two hoots about HT on your gaming rig

 

[Onasi]

It's not been perfectly kind to amd either. Sharing a core is just prone to data leakage, architecturally speaking.

Out of interest, since this is a bit out of my field, was this also the case with the “separate cores sharing some HW” approach the like of which Bulldozer used?

No this is exactly for that reason, security on (shared) or private servers is a major issue. No one gives two hoots about HT on your gaming rig

True. The servers do give many hoots about cramming as much threads into as little physical space as possible though. Rented Units, as described above, don’t SEEM to be doing the same thing as HT did, but I guess we’ll see. Obviously, so far this is just speculation based on a leak, so who knows what we will end up with.

 

[R-T-B]

Out of interest, since this is a bit out of my field, was this also the case with the “separate cores sharing some HW” approach the like of which Bulldozer used?

I'd think so but I'm not sure how many researchers really bothered to poke at old bulldozer era chips.

 

[TumbleGeorge]

The main design of Arrow Lake is complete, so without HT maybe with a next architecture.

 

[Vayra86]

Interesting. Intel innovation firing on all 2 cylinders here
Ahem

 

[Bwaze]

Interesting. I assume the Rentable Units approach, if implemented, will require yet another overhaul to Windows thread scheduling to work properly? Seems we’ve been getting those rather frequently lately.
Honestly, getting rid of HT seems weird. It doesn’t really have any major downsides and leaving any performance on the table is unlike Intel. Unless they are THAT sure that the new approach will compensate and more?

That might be the sole purpose of this exercise - Intel has much larger partnership with Microsoft, and by forcing a redesign of Microsoft thread scheduling they might hope to break or at least slow down multithreading on AMD processors - both consumer and server. And it might even come as a feature - "hyperthreading is inherently unsafe"!

 

[Crackong]

I am curious.
What kind of hardware issue would lead to disabling of HT ?
Intel already had weak server CPU product line right now and they are disabling HT in their next architecture?

 

[ratirt]

Does HT impact power consumption in any meaningful way? If it does, maybe that is Intel's goal. Substitute HT with ecores to reduce power consumption

 

[usiname]

It could be paid feature like in Coffee Lake refresh (9th gen) where i7 was 8/8 and i9 was 8/16

 

[R0H1T]

I am curious.
What kind of hardware issue would lead to disabling of HT ?
Intel already had weak server CPU product line right now and they are disabling HT in their next architecture?

If this is from the same source that our not so favorite YTer recycles I bet we'll get 10 other possibilities before its release, that or WTFtech

 

[ncrs]

I am curious.
What kind of hardware issue would lead to disabling of HT ?
Intel already had weak server CPU product line right now and they are disabling HT in their next architecture?

Several past Intel CPU vulnerabilities can not be worked around fully without disabling HT (SMT):

• L1TF - To prevent the SMT issues of L1TF it might be necessary to disable SMT completely.
• MDS - All MDS variants except MSBDS can be attacked cross Hyper-Threads. That means on CPUs which are affected by MFBDS or MLPDS it is necessary to disable SMT for full protection.
• TAA - tsx_async_abort=full,nosmt - The same as tsx_async_abort=full, with SMT disabled on vulnerable CPUs that have TSX enabled. This is the complete mitigation.
• INTEL-SA-00330 - The mechanism does not mitigate L1D data leaks between tasks belonging to different processes which are concurrently executing on sibling threads of a physical CPU core when SMT is enabled on the system. This can be addressed by controlled placement of processes on physical CPU cores or by disabling SMT.
• MMIO Stale Data - mmio_stale_data=full,nosmt - Same as full, with SMT disabled on vulnerable CPUs. This is the complete mitigation.

 

[Melvis]

Well to be fair HT was always pretty shit, it would give you at most 25% extra performance per core and that was with a steep down hill with a strong breeeze behind it.

 

[ncrs]

Well to be fair HT was always pretty shit, it would give you at most 25% extra performance per core and that was with a steep down hill with a strong breeeze behind it.

While it's true that HT is heavily workload dependent, and for most software its benefit is not that high, there are ones that do scale well. Try running a 7-zip benchmark on a modern (Zen/Alder Lake and newer) CPU. My 5800X is able to achieve almost 100% scaling in decompression.

 

[Rus4kova]

I don't read this a security related at all. I see this as Intel's attempt to mitigate their vastly "superior" power usage compared to AMD.
They know they can't compete in efficiency with the new process optimized 4c cores so they "invent" this instead to try and derail the AMD CPU train.

 

[Redwoodz]

Soon to be announced the new I914900KFH- with HT enabled for only $100 more

 

[Bwaze]

Nah, disabling of HT will be paraded as the great security upgrade, and AMD cited as the unsafe company that still uses the inherently unsafe tech (although many of the security risks were Intel specific).

 

[FoulOnWhite]

Had a read of the RU link, interesting, and may be the end of hyper threading.

 

[TumbleGeorge]

What kind of hardware issue would lead to disabling of HT ?

Tiles, chiplets?

 

[Daven]

Intel is making the same mistake AMD did with Bulldozer’s CMT architecture. Trying to use bunches of less functional cores to try and offset lower IPC.

Low IPC E cores
Constant refreshes
Lower IPC on MTL
No HT
Low core counts on Xeon
Glacial GPU development

I’m not sure Intel is really trying anymore with its chip designs.

 

[Assimilator]

Can y'all please read the article linked in the post instead of speculating pointlessly on why Intel is introducing this? It very clearly explains that the problem is that scheduling tasks on hybrid architectures can result in a task being split over a P- and E-core, and the P-core finishes its work first then has to wait for the E-core to do its part, which wastes valuable P-core time.

That said, I'm not sure why this requires HT to go away; it feels like something that should be fixed at the scheduling level, not by completely reworking some very fundamental ways in which cores have been designed for over a decade. But considering how much pain scheduling over P- and E-cores has given Intel so far (WRT having both the CPU and OS having to be aware of which cores to schedule tasks to), they may have determined that this rather drastic approach - which seems to imply moving scheduling out of the operating system and fully back onto the CPU - is worth it.

Finally, it is important to remember that this is just one of hundreds of patents that Intel files every year - it absolutely does not mean this is the path their future CPUs will take.