Tuesday, August 18th 2026

Intel "Razor Lake" to Use TSMC's N2X Node, Brings bLLC to Laptop SKUs

Intel's "Razor Lake" is a next-generation processor lineup set to launch after the upcoming "Nova Lake" Core Ultra 400 series, scheduled for early 2027. Today, we learn that Intel is planning some "Razor Lake" production to be done on TSMC's N2X node, which is also scheduled for 2027. According to a well-known Intel leaker, Jaykihn, Intel's node of choice for "Razor Lake" will be the enhanced 2 nm design from TSMC. The N2X node is part of the N2 node family but is designed to maintain high frequencies at the cost of some power efficiency compared to the standard N2. This suggests that Intel is preparing to boost frequencies for its next-generation designs, likely to compete with AMD and introduce a performance-leading design.

However, N2X might not be the only node used for production, as Intel is known for dual-sourcing compute tiles, similar to the approach with the upcoming "Nova Lake." Last month, rumors emerged that Intel initially planned to outsource the main compute tile production for "Nova Lake" to TSMC, utilizing its 2 nm N2 node. The majority of the production, about 60-70%, was to be handled by TSMC, with the remainder produced in-house by Intel Foundry on the 18A node. This meant Intel would dual-source production for its next-generation chip. However, the latest supply chain information suggests that the company now intends to bring the majority of "Nova Lake" compute tile production in-house, with about 80-90% of the tiles being manufactured by Intel Foundry. The same could happen with "Razor Lake," depending on the performance of the 18A-P node. Image below is just an illustration.
Additionally, the leaker claims that "Razor Lake" will also introduce Intel's big Last Level Cache (bLLC) technology to laptop designs, which will boost gaming performance on mobile platforms. This also indicates that the upcoming "Nova Lake" laptop SKUs will not feature bLLC technology, which will only be present on the desktop SKUs that Intel will offer.
Source: Jaykihn on X
Add your own comment

18 Comments on Intel "Razor Lake" to Use TSMC's N2X Node, Brings bLLC to Laptop SKUs

#1
ncrs
[...] as Intel is known for dual-sourcing compute tiles [...]
Are they? As far as I know they didn't dual-source compute tiles yet. Meteor Lake - Intel 4, Lunar Lake - TSMC N3B, Arrow Lake - TSMC N3B, Panther Lake - Intel 18A.
Panther Lake does dual-source the GPU tile between TSMC N3E and Intel 3, but they are different designs - 12 and 4 Xe-cores respectively.
Posted on Reply
#2
Darmok N Jalad
I wonder if this means AMD might bring 3DV to mobile ahead of this. Considering how much cache helps both CPUs and GPUs, it would be interesting to see the results.
Posted on Reply
#3
Space Lynx
Astronaut
I really hope we get a Nova Lake or Razor Lake that has no igpu, and instead focuses more on cpu performance or more "x3d" cache

I just have no need for an igpu with two dedicated graphics cards, one main and one backup
Posted on Reply
#4
wNotyarD
Darmok N JaladI wonder if this means AMD might bring 3DV to mobile ahead of this. Considering how much cache helps both CPUs and GPUs, it would be interesting to see the results.
They already do have 3DV$ on mobile, albeit only two models: 7945HX3D and 9955HX3D. Sure could do with expanding their range.
Posted on Reply
#5
Space Lynx
Astronaut
wNotyarDThey already do have 3DV$ on mobile, albeit only two models: 7945HX3D and 9955HX3D. Sure could do with expanding their range.
What surprises me most is that Razor Lake is expected to get some of the latest node time... and I was under impression all the latest node was already sold out to ai until like 2029... I find it all very confusing sometimes...
Posted on Reply
#6
mikesg
AMD doens't have a strong market segment for laptops with 3D-vcache. Their best laptop & mini-pc product is Strix Halo/Gorgan Halo/Medusa Halo with the huge iGPU integrated. Up to 96MB L3 with the top next-gen Medusa Halo. 32-64MB for others.

In the $10k workstation laptop segment and lesser gaming segment, they are basically all Intel/Nvidia combos.
Posted on Reply
#7
windwhirl
Space LynxI really hope we get a Nova Lake or Razor Lake that has no igpu, and instead focuses more on cpu performance or more "x3d" cache

I just have no need for an igpu with two dedicated graphics cards, one main and one backup
Both Intel and AMD ship such "no IGP" CPUs already, doubt they will stop
Posted on Reply
#8
ixi
Stop with blahblah and deliver :D
Posted on Reply
#9
Bobaganoosh
Space LynxWhat surprises me most is that Razor Lake is expected to get some of the latest node time... and I was under impression all the latest node was already sold out to ai until like 2029... I find it all very confusing sometimes...
There was a rumor a few weeks ago (maybe a couple months, I lose track) that Intel was going to pull Nova Lake's compute tiles fully internal, which was contrary to previous rumors about those being at TSMC. There was some conjecture about what that meant for their TSMC allocations they had definitely already paid for. This might be the answer to that.
Posted on Reply
#10
Space Lynx
Astronaut
windwhirlBoth Intel and AMD ship such "no IGP" CPUs already, doubt they will stop
I don't think you can get an AM5 x3d cpu without an igpu
Posted on Reply
#11
Geddagod
Space LynxI really hope we get a Nova Lake or Razor Lake that has no igpu, and instead focuses more on cpu performance or more "x3d" cache
It's not as if the -KF skus from Intel have had way more CPU perf than the regular -K skus, right?
wNotyarDThey already do have 3DV$ on mobile, albeit only two models: 7945HX3D and 9955HX3D. Sure could do with expanding their range.
I'm not so sure if there's that much of a point of a lower core count X3D model that uses the same desktop chiplet architecture that the listed skus above use.
And I think hoping that AMD will add TSVs to their monolithic APUs is a pipedream with how stingy they are with area on those chips. Though honestly that might end up being a great APU knowing how much power can be saved by utilizing the 3D V-cache's gaming perf/clock improvements to compensate for lowering the boost clocks a bunch to claw back a bunch of power.
Space LynxWhat surprises me most is that Razor Lake is expected to get some of the latest node time... and I was under impression all the latest node was already sold out to ai until like 2029... I find it all very confusing sometimes...
Intel's a pretty large customer of TSMC, and a massive chunk of the AI demand comes from Nvidia which won't move to a N2 family node till late 2028 (a year after RZL).
Don't forget Mediatek is launching a mobile SOC this year on N2P, an Qcomm is expected to as well. Do you expect those companies to stop producing N2 chips in 2027 and 2028, or have to move their flagships back to N3, because apparently all N2 capacity is sold out till 2029?
And it's not as if Intel is asking for N2 capacity on the fly here. RZL was likely planned to be N2 for a while (for development purposes, it takes years to design these chips), and whether it uses N2X or N2P shouldn't change that, they should run on the same lines.
BobaganooshThere was a rumor a few weeks ago (maybe a couple months, I lose track) that Intel was going to pull Nova Lake's compute tiles fully internal, which was contrary to previous rumors about those being at TSMC. There was some conjecture about what that meant for their TSMC allocations they had definitely already paid for. This might be the answer to that.
I think at most it was a month (maybe 2) ago lol. But I'm very interested in how the analysts (I believe it was UBS?) that reported that expect NVL to change manufacturers on a whim only months before launch. At best it could be explained that since the 4+8 tile was planned to be internal for a while, that the tile was dual tracked and so a switch is very possible. But they also specifically claimed DT chips, which was always rumored to be N2.
Posted on Reply
#12
shuffy_losys
Space LynxI really hope we get a Nova Lake or Razor Lake that has no igpu, and instead focuses more on cpu performance or more "x3d" cache

I just have no need for an igpu with two dedicated graphics cards, one main and one backup
Thankfully it's easy for Intel to not include an iGPU since it's its own tile.

Unfortunately, that likely won't mean you'll get a better performing CPU. Just a probably cheaper version of an existing product.
Posted on Reply
#13
windwhirl
Space LynxI don't think you can get an AM5 x3d cpu without an igpu
Like it hardly matters when you can shut it off and have it dead to the world, not a single watt wasted in the IGP. It matters even less when it's in a separate piece of silicon away from the CCDs.

You could ask for a better interconnect between the IOD and the CCDs tho, that I'd totally understand. Or a revamped design altogether that gets rid of the issues of the current one. I think Zen6 is meant to change it significantly, tho it remains to be seen how much.
Geddagodbecause apparently all N2 capacity is sold out till 2029?
TSMC is unlikely to change reserved allocations, I think? Whatever capacity exists, if it's sold out it's because it's already reserved and no company is fancying that they can get to bypass the line. I don't think even Nvidia or Apple could get TSMC to tell another customer to go fuck themselves just because either of those two companies presented a bigger cash bag at the reception desk... which is why they just reserve the expected capacity they will need well in advance.
Posted on Reply
#14
LittleBro
It's pretty clear Intel Foundry is unable so satisfy even Intel's own demand, yet they want to make chips for Nvidia, Apple, Amazon, etc.
When experts said (month ago) that Apple's allocation of Intel 18A production quotas will cause harm to TSMC only in cosmetic way, they were entirely true.
Intel Foundries will be mainly used for packaging chips, not for production of computing tiles for Apple, Nvidia, ...
Intel just can't satisfy the demand. New TSMC's production sites (4 in Taiwan) are being built while I'm writing this post.
Posted on Reply
#15
Space Lynx
Astronaut
windwhirlLike it hardly matters when you can shut it off and have it dead to the world, not a single watt wasted in the IGP. It matters even less when it's in a separate piece of silicon away from the CCDs.

You could ask for a better interconnect between the IOD and the CCDs tho, that I'd totally understand. Or a revamped design altogether that gets rid of the issues of the current one. I think Zen6 is meant to change it significantly, tho it remains to be seen how much.


TSMC is unlikely to change reserved allocations, I think? Whatever capacity exists, if it's sold out it's because it's already reserved and no company is fancying that they can get to bypass the line. I don't think even Nvidia or Apple could get TSMC to tell another customer to go fuck themselves just because either of those two companies presented a bigger cash bag at the reception desk... which is why they just reserve the expected capacity they will need well in advance.
i think you misunderstand, i mean i wish they would design a chip to be better from the ground up to focus fully on cpu and gaming performance, no space allocated to igpu in previous design etc
Posted on Reply
#16
shuffy_losys
Space Lynxi think you misunderstand, i mean i wish they would design a chip to be better from the ground up to focus fully on cpu and gaming performance, no space allocated to igpu in previous design etc
Then basically you want an X3D version of Threadripper.
Posted on Reply
#17
Jaykihn
I'm late, but I wanted to issue a correction: Razor Lake uses N2P V2, not N2X.
Posted on Reply
#18
kazuviking
GeddagodIntel's a pretty large customer of TSMC
You mean they bought the bare minimum wafers for both arl and novalake. Its not a secret that intel bought the bare minimum wafers required.
Posted on Reply
Add your own comment
Nov 12th, 2026 04:40 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts