Friday, February 7th 2025

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

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.
Sources: @jaykihn0 on X, via Tom's Hardware
Add your own comment

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

#26
john_
dyonoctisZen 6 will have 12 cores CCD apparently, so they need to fight back the incoming 24 cores 48HT mainstream Ryzen CPU coming up. Nobody is going to pay a premium over AMD for a 24P cores from Intel, with all the expensive TSMC die real estate their P cores are using :D
AMD's Zen was successful because it was offering more cores than Intel with a good enough IPC. Intel managed to turn the tables with it's Hybrid CPUs and started taking back market share because it was offering more cores than AMD(most people wouldn't care to differentiate E cores from P cores, I can see myself looking at my phone specs and convincing me that it has an 8 core SOC when in fact only 4 cores are P cores). In the near future AMD will have to either change it's marketing and start advertising threads, or find a way to match Intel's core counts, because it will be difficult to convince the uneducated consumer than a 24 core CPU is as good as a 48 core CPU. I remember my self having problems to convince people that a mid range GPU with 2GB VRAM is better than a low end GPU with 4 GB slow VRAM, because people where thinking that double VRAM equals double performance.
Of course people with some basic knowledge will be looking at IPC, number of P cores and the implementation of X3D cache on the CPUs they are buying. But big OEMs will be advertising "32/40/48/52 cores CPUs".
Posted on Reply
#27
kondamin
3valatzyYes, but for Apple. For Intel no.
Apple is sticking to 3nm this(m5) generation
which is supposed to launch h2 2025
maybe m6 will be 2nm but that's not going to launch until 2026

I don't know who's going to be taking up the first generation of 2nm but it's not going to be Apple this time.
Probably still reeling from n3b like intel is.
Posted on Reply
#28
_roman_
john_most people wouldn't care to differentiate E cores from P cores
If you stick to windows only - it will most likely not matter.
I could not find any specs for those E cores for those intel processors on intel homepage. I could not find any cpuinfo for that. I doubt my box will boot or can execute the code i compiled on that intel cpu. Maybe when i optimise the code, downgrade compability for i686, for a ~15 year old cpu generation, which I do not want for a new platform.
john_I remember my self having problems to convince people that a mid range GPU with 2GB VRAM is better than a low end GPU with 4 GB slow VRAM
The vram "fairy tale"
2GB VRAM is enough
8GB VRAM is enough
john_Of course people with some basic knowledge will be looking at IPC, number of P cores and the implementation of X3D cache on the CPUs they are buying.
I want a cpu with big cache. That never changed.
john_But big OEMs will be advertising "32/40/48/52 cores CPUs".
How is the operating system support? Does the customer have to pay for every single extra cpu core in software ?
Posted on Reply
#29
Arco
NostrasI 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.
Yeah but for game server hosting I need every single full core I can get. I don't wanna pay TR prices. This was the main reason I got a 7950X instead of the 13900K when it launched.
Posted on Reply
#30
R0H1T
Water, water everywhere but not a drop to drink ~ Intel right now :laugh:
Posted on Reply
#31
dyonoctis
_roman_If you stick to windows only - it will most likely not matter.
I could not find any specs for those E cores for those intel processors on intel homepage. I could not find any cpuinfo for that. I doubt my box will boot or can execute the code i compiled on that intel cpu. Maybe when i optimise the code, downgrade compability for i686, for a ~15 year old cpu generation, which I do not want for a new platform.
Intel is selling E cores Xeon, (that have benchmarked on Linux ) so I doubt that it require such exotic requirements. Those Xeon support hundred of gigabytes of ram, so e cores are definitely not a 32bits arch.
www.techpowerup.com/cpu-specs/xeon-6780e.c3664
Posted on Reply
#32
N/A
AquinusFirst 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.
On toms the source, Hopefully 144MB can provide enough buffered memory. But it could be done with just 96MB if the IMC was on the CPU die where it would only take up 0.5mm of space and make the CPU run at 50% free performance.
Posted on Reply
#33
Nhonho
Another Intel CPU that will heat up like an electric shower, with high latencies between the core sets and also with the RAM memory and another sales failure...
john_Intel will sell cores the same way Nvidia sells frames.
The best comment I've ever seen here on TPU.
memory latency
If AMD put the memory controller on the same die as the x86 cores, there would be a direct performance gain of at least ~15% in the IPC of its CPUs. But AMD is an expert at hesitating, selling CPUs with 16MB of L3 to lose many sales opportunities...
Posted on Reply
Add your own comment
Mar 12th, 2025 06:34 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts