Intel "Nova Lake" to Appear with up to 52 Cores: 16P+32E+4LPE Configuration

[AleksandarK]

Intel's upcoming "Nova Lake" desktop processors are taking shape slowly, featuring a three-tier core design that could reach 52 total cores. Set for 2026, the flagship SKU combines 16 "Coyote Cove" P-cores with 32 "Arctic Wolf" E-cores, supplemented by 4 LPE-cores for background task management. Intel is reportedly also considering 28-core (8P + 16E + 4LPE), and 16-core (4P + 8E + 4LPE) SKUs too. The architectural design choice centers on Intel's hybrid manufacturing approach, leveraging both its internal 14A node and TSMC's 2 nm process technology. This strategic decision addresses supply chain resilience while potentially enabling higher yields for critical compute tiles. Intel's interim co-CEO Michelle Johnston Holthaus noted that Intel Foundry will need to earn Intel Product's trust with each new node, so if a node is not the best for their in-house IP, Intel will move to TSMC for production.

Initial engineering samples are already circulating among developers, according to shipping documentation from NBD, suggesting the validation phase is proceeding on schedule. Some specifications point to significant cache improvements, with documentation suggesting a 144 MB L3 cache implementation. However, the cache topology—whether unified or segmented—remains unspecified. The platform is expected to support PCIe Gen 6.0, though Intel has yet to confirm socket compatibility or memory specifications. However, we need to hold our expectations low. Previously unrealized configurations in Intel's roadmaps, like 40-core "Arrow Lake," never materialized, and instead, we got an eight-P-core version with 16 E-cores, totaling 24 cores. Final specifications may evolve as the platform progresses through development phases.



View at TechPowerUp Main Site | Source

 

[john_]

Intel will sell cores the same way Nvidia sells frames.

 

[DemonicRyzen666]

Intel will sell cores the same way Nvidia sells frames.

Well, you're not wrong there it is a type of "Performance" it's "Throughput" performance though

 

[N/A]

eight-P-core version with 16 E-cores, totaling 32 cores

There must be a mistake. Fixed

off-die memory controller,

Please No.

 

[Arco]

There must be a mistake.

Introducing our new multi-core generation tech using state-of-the-art AI to make stunning new fake cores.

 

[TumbleGeorge]

Don't be mistaken. AMD also will increase core count. I hope that next series continuing with SMT.

 

[Wirko]

Intel's interim co-Ceo Michael J. Holantus

Who?

 

[Bobaganoosh]

I don't know that there'd be a big reason to complain about all those e-cores, but I just feel like they're totally unnecessary for most people. I'd rather see something like 12-16 P-cores and 4-8 e-cores. Also, with improvements to the memory latency re: Arrow Lake issues.

 

[_roman_]

I wonder how the software will deal with that. Good vs crap vs crappy cores in regards of power consumption, cpu features, speed.

High benchmark numbers in nonsense benchmarks which can utilise all cores vs real life scenario which needs performance cores with all cpu instructions with high ipc.

 

[AleksandarK]

Who?

My bad! Fixed!

 

[watzupken]

While it will be nice to see Intel go all in on the core count war, I do think it may not be feasible to produce a retail chip with this many cores. For 1, the power requirement will be very high and likely very difficult to cool. To offset the power requirements, clock speed will likely be impacted, which will somewhat dull the single core performance advantage. And to be honest, do away with that many redundant e-cores. The "E" meaning has changed from Efficient, to Excessive.

 

[kondamin]

So both Intel and tsmc have a running 14a/2nm line up and running?

that high a number of e cores is silly, 8 is plenty
4 is good 16 p cores and full avx please

 

[TumbleGeorge]

So both Intel and tsmc have a running 14a/2nm line up and running?

that high a number of e cores is silly, 8 is plenty
4 is good 16 p cores and full avx please

Do you have enough dual channel memory bandwidth to set 16p cores on full throttle?

 

[Prima.Vera]

Too many E cores.... Just make 16P, 12P and 8P Core CPUs, with maximum 8E Cores. Nobody cares about shitty E-Cores, which creates more trouble than they are worth it.

 

[kondamin]

Do you have enough dual channel memory bandwidth to set 16p cores on full throttle?

It’s no issue on zen 5 which is about as wide as a p core
so that shouldn’t be a problem.
and since it if it ever were to come to market is going to be 2027 product memory clocks are going to be nearer the 10k mark

 

[Nostras]

Too many E cores.... Just make 16P, 12P and 8P Core CPUs, with maximum 8E Cores. Nobody cares about shitty E-Cores, which creates more trouble than they are worth it.

I think what Intel is doing is actually right. In cases where single-core performance is really important nothing really scales beyond 8 cores. Inverse applies as well.
8+32 would make for an excellent productivity GPU with very competent gaming. Assuming the new gen at least beats Raptor Lake in games.

 

[agent_x007]

I say it's plain dumb, because Windows already has issues managing 8P+16E we already have on Arrow Lake, and you want to throw 16P+32E configuration on top of that ?
Oof... someone at M$ team will try to pull his hair out making this work like it should.

Just throw away E-cores, and do P-core only die, please.
It shouldn't be that much harder than what Intel's currently doing.

 

[TumbleGeorge]

I say it's plain dumb, because Windows already has issues managing 8P+16E we already have on Arrow Lake, and you want to throw 16P+32E configuration on top of that ?
Oof... someone at M$ team will try to pull his hair out making this work like it should.

Just throw away E-cores, and do P-core only die, please.
It shouldn't be that much harder than what Intel's currently doing.

Windows runs professional processes with many more cores and even optional modules. I think that problem isn't from windows side. And yes i know that Linux is better and cheaper for supercomputers because its clusters scale architecture is specific for any facility.

 

[agent_x007]

All of "Windows version [XYZ], and/or patch [ABC] required for best performance", notes from Intel that were released up to this point, don't provide enough evidence to support :

Windows runs professional processes with many more cores and even optional modules.

^this (at least for current gen stuff).

I agree that it may support those, but only when CPUs are of the "usual" architecture [ie. non-hybrid].
Reason : Those CPUs had simply build on top of code that previous years of work laid (squishing most of the bugs inside kernel earlier).

 

[_roman_]

Who needs all those low value - - can not execute all instruction - cores?

28-core (8P + 16E + 4LPE) - I agree with these 8 Performance Cores, but who needs those 20 low value trash cores?

I think android based smartphones and tablets have 4 + 4 ARM cores without hyperthreading for a long time.

I do not see any drawback with a Ryzen 7600X. For a low end entry point gaming box or desktop system.

The linux kernel has no support for those intel special reduced cores and windows also does not have it.
I talk about full 100% support in the year 2023, month january.
I do check regularly the changes for the linux kernel - there is so much done in regards of those mixxed core stuff and scheduler. The work has not really started.

If you disagree - please run your windows installation with the first windows 10 iso - first release version. Not windows 11 or later. Same with all those chipset driver and other nonsense. Use the first ever released version - the stock universal one - from the first windows 10 iso.

In 2023 there were several months or years already those mixed core processors widely available to the end consumer.

 

[Aquinus]

Please No.

First of all, who are you quoting? Second of all, AMD put the memory controller on the IO die instead of on the CCDs and it seems to work pretty well for them. Assuming Intel is going with the tile-based approach like they did with Arrow Lake, it would make sense to put IO on its own tile.

 

[freeagent]

I am interested in its thermal characteristics and clock speeds.

Intel right now is like moar coars is bedder. But is it?

How about 12 or 16 cores at 6GHz plus boost with competition crushing IPC?

No? Thought not..

 

[TumbleGeorge]

at 6GHz

Enough is 1-2 cores on 10GHz or more. Was in plans from before 20 years. That not happened.

 

[3valatzy]

So both Intel and tsmc have a running 14a/2nm line up and running?

Yes, but for Apple. For Intel no.

Intel right now is like moar coars is bedder. But is it?

The concept in general is good, if they find a way to put all those cores in 95W TDP or max 125W, it will be game over for AMD's Ryzen, which for some weird reason still stays on the ancient core count of only 16.
I bet 2026 is not the year, but maybe in 2032, definitely.

 

[igormp]

The linux kernel has no support for those intel special reduced cores and windows also does not have it.
I talk about full 100% support in the year 2023, month january.
I do check regularly the changes for the linux kernel - there is so much done in regards of those mixxed core stuff and scheduler. The work has not really started.

linux already has support for it for over a year now, you may be reading the wrong places.
Scheduler changes are normal and an ever going with, but the LPE cores have been supported for a really long time, even before meteor lake.

First of all, who are you quoting? Second of all, AMD put the memory controller on the IO die instead of on the CCDs and it seems to work pretty well for them. Assuming Intel is going with the tile-based approach like they did with Arrow Lake, it would make sense to put IO on its own tile.

AMD's efficient, mobile designs are monolithic tho.