Intel "Nova Lake-S" Core Ultra 3, Ultra 5, Ultra 7, and Ultra 9 Core Configurations Surface

[btarunr]

Intel will be significantly increasing CPU core counts with its next-generation "Nova Lake" client processor microarchitecture. For the desktop variant, the "Nova Lake-S," this marks a maximum CPU core configuration of 16P+32E+4LPE for an impressive 52 cores per socket. That's 16 "Coyote Cove" P-cores, 32 "Arctic Wolf" E-cores spread across eight E-core clusters; and 4 additional low-power island E-cores based on the same "Arctic Wolf" core design. While the P-cores and E-core clusters share an L3 cache, the LPE cores are not part of the CPU complex, and are located in a low-power island that's part of the SoC region of the chip, an arrangement similar to "Meteor Lake." The top Core Ultra 9 is expected to come with a massive L3 cache of 144 MB, which should benefit gaming workloads.

Intel will give only its top Core Ultra 9 SKUs the maxed out configuration of 16P+32E+4LPE. This chip is expected to come with a processor base power value of 150 W for its unlocked K or KF SKUs. We are now learning what the core configurations of the other brand extensions, 7, 5, and 3, could look like. For the Core Ultra 7 SKUs, Intel is expected to opt for a 14P+24E+4LPE configuration for a total of 42 cores. To do this, the company will disable two P-cores, and two E-core clusters, while also reducing the L3 cache. The top K- and KF-SKUs of Core Ultra 7 are expected to come with the same 150 W processor base power values as the Core Ultra 9 K/KF SKUs.



Things get very interesting with the Core Ultra 5 "Nova Lake-S" series. Intel is dialing up the core-counts it's held onto for three generations, and giving the top Core Ultra 5 K/KF series chips an 8P+16E+4LPE core configuration, for a 28-core per socket product. There will also be a non-K/KF SKU with 8P+12E+4LPE configuration, and a lower-end Core Ultra 5 SKU with 6P+8E+4LPE. We're not sure if this is based on physically the same chip as the Core Ultra 9, or if Intel is developing a physically smaller Compute tile. The SoC tile with 4 LPE cores are expected to remain the same. The Core Ultra 5 series will hence come with three core configurations depending on the processor model—8P+16E+4LPE for the top K/KF SKUs, 8P+12E+4LPE for the non-K/KF top SKUs, and 6P+8E+4LPE for the lower-end SKU.

We now move into the value segment, and Intel is planning to address this with two Core Ultra 3 processor types. The upper SKU comes with 4P+8E+4LPE configuration (16 cores) with 65 W processor base power. At the very entry level, is a SKU with 4P+4E+4LPE (12 cores), with 65 W base power.

As for platform I/O, Intel tends to keep its connectivity consistent across processor brand extensions, differentiated only by the motherboard chipset models. With its top chipset model, the "Nova Lake-S" platform is expected to offer a total of 32 PCI-Express 5.0 lanes, and a total of 16 PCI-Express 4.0 lanes. It is likely that Intel updates the chipset bus with newer DMI 5.0 for bandwidth resembling PCI-Express 5.0 x8 for the top chipset model, and PCI-Express 5.0 x4 for the mid-tier chipset model.

The iGPU of "Nova Lake-S" is expected to be based on the Xe3 "Celestial" graphics architecture for generational performance gains, although this will be a smaller iGPU than the one expected to feature in the mobile H-segment and U-segment variants of "Nova Lake." All processor models are expected to come with an NPU that meets Microsoft Copilot+ local acceleration requirements.

Intel is expected to debut "Nova Lake-S" in 2026.

View at TechPowerUp Main Site | Source

 

[GerKNG]

that's a lot of cinebench accelerators! they will probably inflate the prices by a lot.
16P with HT 0 - 0 please.

 

[Eva01Master]

I really hope these deliver, a CPU war would drive the platform prices down, they've been creeping up for a while.

 

[BArms]

More cores do not necessarily a better CPU make.

Guessing AMD will keep gaining market share. It seems to me, and I could be wrong, that Intel is going down another dead road ala NetBurst where it chased higher MHz at the cost of all else. Now they're cramming in as many cores as they can for apps that mostly can't use them.

 

[ZoneDymo]

More cores do not necessarily a better CPU make.

Guessing AMD will keep gaining market share. It seems to me, and I could be wrong, that Intel is going down another dead road ala NetBurst where it chased higher MHz at the cost of all else. Now they're cramming in as many cores as they can for apps that mostly can't use them.

I feel Intel has no focus, it just switches constantly, I also feel getting rid of Gelzinger was a mistake....but time will tell

 

[rattlehead99]

that's a lot of cinebench accelerators! they will probably inflate the prices by a lot.
16P with HT 0 - 0 please.

Hyper-threading is a crutch for bottlenecked architectures. You want your core to be 90-95% well fed, not 70-75%. Meteor Lake already proved that Hyper-Threading is almost worthless when the core is not bottlenecked hard as it gained between 7-15% more performance from it and Lunar Lake/Arrow Lake is even less so, so I imagine NovaLake won't be bottlenecked either. That said with the new CEO I expect smaller cores that are bottlenecked to return to cut costs and maximize profits.

 

[Chaitanya]

I feel Intel has no focus, it just switches constantly, I also feel getting rid of Gelzinger was a mistake....but time will tell

No, getting rid of Jim Keller and throwing out his work along with him was a blunder of epic proportions.

 

[rattlehead99]

No, getting rid of Jim Keller and throwing out his work along with him was a blunder of epic proportions.

True that

 

[N/A]

The Core Ultra 5 K 8P+16E+4LPE +144MB is probably similar to the 9950X3D, but without the drawbacks.

 

[efikkan]

More cores do not necessarily a better CPU make.

Yes, there will always be diminishing returns.
The usefulness beyond 8 cores in interactive applications is slim.

Hyper-threading is a crutch for bottlenecked architectures. You want your core to be 90-95% well fed, not 70-75%. Meteor Lake already proved that Hyper-Threading is almost worthless when the core is not bottlenecked hard as it gained between 7-15% more performance from it and Lunar Lake/Arrow Lake is even less so, so I imagine NovaLake won't be bottlenecked either. That said with the new CEO I expect smaller cores that are bottlenecked to return to cut costs and maximize profits.

SMT has always been about utilizing idle cycles for other work, in order to get closer to a fully saturated core. But as we know, the cost and complexity of SMT is increasing, and the benefits diminishing as the CPU front-end gets better. This engineering effort and transistor budget is much better spent elsewhere.

SMT is only going to get even more irrelevant when APX is added, as it will greatly reduce the amount of pipeline stalls. We know it's coming with Diamond Rapids which have a similar launch window as Nova Lake, but there is no confirmation about Nova Lake yet.

 

[JustBenching]

Yes, there will always be diminishing returns.
The usefulness beyond 8 cores in interactive applications is slim.

8 cores are super restrictive for actual pc use case. Its not a console man.

 

[ZoneDymo]

No, getting rid of Jim Keller and throwing out his work along with him was a blunder of epic proportions.

wait, did Gelzinger do that?

 

[efikkan]

8 cores are super restrictive for actual pc use case. Its not a console man.

This is the problem with people who base their understanding of computing on synthetic benchmarks or batch loads, rather than having a deep understanding of how software works.

Faster cores are always going to be better than more cores for user-interactive workloads.

 

[Rus4kova]

What happened to keeping it simple ?
This looks overly complex simply to hide the fact that the have no idea how to move forward.

 

[JustBenching]

This is the problem with people who base their understanding of computing on synthetic benchmarks or batch loads, rather than having a deep understanding of how software works.

Faster cores are always going to be better than more cores for user-interactive workloads.

What more understanding do i need other than actually trying both 8 core and 16 core options and figuring out that there are things that the 8 cores simply cant do properly.

Less cores doesnt mean faster cores anyways. A 9700x has more cores than a 9600x but each core is equally fast.

Stop the passive aggressive thing

 

[sarthurn]

And another new socket type...sheesh. New mobo, new brackets for cooling, etc etc.

 

[dyonoctis]

More cores do not necessarily a better CPU make.

Guessing AMD will keep gaining market share. It seems to me, and I could be wrong, that Intel is going down another dead road ala NetBurst where it chased higher MHz at the cost of all else. Now they're cramming in as many cores as they can for apps that mostly can't use them.

Because you think that AMD will stay with a 16 cores/32 threads max config for zen 6 even though they have 12/16 cores CCD now? Intel put more cores because the competion is also about to put more cores. It's really that simple, don't read too much into it

This is the problem with people who base their understanding of computing on synthetic benchmarks or batch loads, rather than having a deep understanding of how software works.

Faster cores are always going to be better than more cores for user-interactive workloads.

But we are also in an era where the Highest core count SKU on the mainstream platform is also the fastest in ST. I don't understand why I'm still seeing so many people acting as if we are still in the X99/X299 era were clock speed tanked hard starting from six cores...But I also don't understand why the top SKU having more cores than most people need seems to be a bother. Just buy a core i5 with 8 cores for cheaper. Or is there some kind of social status stigma about not having a core i9?

 

[Tpanon]

8 cores are super restrictive for actual pc use case. Its not a console man.

8 cores are restrictive for workstation use cases.
Most pc users would never use more than 8 effectively.

 

[JustBenching]

8 cores are restrictive for workstation use cases.
Most pc users would never use more than 8 effectively.

Sure, but then most people shouldn't be looking at the latest and greatest super high end i9s and ryzen 9s anyways.

 

[Hecate91]

Because you think that AMD will stay with a 16 cores/32 threads max config

No it's because of the diminishing returns of just adding more cores without additional bandwidth, especially with the marketing tactic of throwing more E cores on, it seems like more of a stopgap than any real performance increase. Intel getting rid of Jim Keller and the Royal Core project was a terrible mistake.

9950X3D, but without the drawbacks.

I would expect Intel to charge a hefty premium for 16 P cores and 144mb cache, and no upgrade path to the platform would be a significant drawback.

 

[Chaitanya]

wait, did Gelzinger do that?

He was in charge while the internal politics of Intel pushed out Keller and then to throw out his work. Worst of it came when Intel started using TSMC fabs(while talking down them publicly) which was apparently one of the friction points with Keller who wanted to use TSMC fabs a lot sooner than it happened. Gelsinger's post Intel activities/statements do raise some concerns about him in general and his leadership.

 

[JustBenching]

No it's because of the diminishing returns of just adding more cores without additional bandwidth, especially with the marketing tactic of throwing more E cores on, it seems like more of a stopgap than any real performance increase.

So novalake wont have any performance increases over arrowlake. Im willing to bet everything ive got that you are wrong. Lets wait for reviews and ill quote you.

 

[Sol_Badguy]

Scratching my head at the "i9" designation going on in this here thread.

I see some serious difference regarding the PCI-E lanes compared to the previous platforms:

Z790 vs Z890: What's the difference?

So the Z890 chipset will have 24 lanes Gen 4.0 and no PCIe 3.0, whereas the Z790 had 20 lanes Gen 4.0 and 8 lanes 3.0.

www.corsair.com

And another new socket type...sheesh. New mobo, new brackets for cooling, etc etc.

You mean the cooler mounts or the contact frames?
The holes might stay the same, but the contact frames will inevitably require new models, as it should be.

Because you think that AMD will stay with a 16 cores/32 threads max config for zen 6 even though they have 12/16 cores CCD now? Intel put more cores because the competion is also about to put more cores. It's really that simple, don't read too much into it

It's very convenient for some to miss the positive aspect of doubling the P-Cores (and quadrupling the cache), and instead focus on the "negative" aspect of doubling the E-Cores.
And AMD is increasing the core count by 50%, yet it's the people's champion.

The Core Ultra 5 K 8P+16E+4LPE +144MB is probably similar to the 9950X3D, but without the drawbacks.

The one in the table has 72MB, half of the flagship.

I would expect Intel to charge a hefty premium for 16 P cores and 144mb cache, and no upgrade path to the platform would be a significant drawback.

That would mean AMD does the same. So if the 10950X will not have a $650 MSRP like the 9950X, then what will it be? $700 like the 9950X3D?
I would say so, I don't see them going towards $1000 like the old Intel HEDT.
So if the 10950X gets $700 so will the 485K.
It's nasty but if both deliver then at least it's palatable.

^Please don't consider that I'm advocating/defending increased MSRPs, I would love the flagships to be $500 but can we realistically expect that? Even with performance parity?
People keep saying that real competition will bring down prices, but will it really? Both companies can just slash $50 off the flagships, and then say: phuck y'all, if you still can't afford it buy the next one down in the stack.

 

[igormp]

All I hope for is for them to bring back AVX512 across their entire stack.

 

[docnorth]

This is the problem with people who base their understanding of computing on synthetic benchmarks or batch loads, rather than having a deep understanding of how software works.

Faster cores are always going to be better than more cores for user-interactive workloads.

Not always, but mostly yes.
Anyway the most important leak might be the big last-level cache SKUs, if true, already mentioned by some members.