Wednesday, February 11th 2026

Intel "Nova Lake" Compute Tile Die-sizes Surface

Intel is preparing to launch its next generation Core Ultra Series 4 "Nova Lake" processors in the second half of 2026. We have some hint on what the sizes of the Compute tiles of these chips could measure. "Nova Lake" remains a tile-based disaggregated chip, just like the current "Arrow Lake," with the most advanced foundry node being allocated only to the components that can benefit the most from it, the CPU cores. The CPU complex of the processor is located in the Compute tile, while low-power island E-cores are located on the SoC tile that's built on a slightly older foundry node.

Intel is building the Compute tile of "Nova Lake" on the TSMC N2 (2 nm nanosheet) foundry node. There will be different types of Compute tiles based on the CPU core count and cache sizes. The two most popular ones for the desktop platform will be the mainstream tile with 8P+16E core configuration and a standard L3 cache size shared among the eight P-cores and four E-core clusters. The second, more premium Compute tile type will be 8P+16E with bLLC, or big last-level cache. bLLC is an enlarged L3 cache that's estimated to be 3-4 times the size of the regular L3 cache. This is Intel's answer to AMD's 3D V-Cache. Estimates put the regular Compute tile to measure around 110 mm², while the premium variant with bLLC is estimated to measure over 150 mm². This 36% increase in die-area is entirely from the enlarged last-level cache.
This isn't the entirety of the "Nova Lake" chip, there will be other components, such as the SoC tile that contains the low-power island E-cores, the DDR5 memory controllers, the 74 TOPS NPU, and the PCI-Express Gen 5 root complex, besides the System Agent, Management Engine, and security processor. Besides the SoC tile, are the Graphics tile, which is likely to be built on a more advanced node than the SoC, but less advanced than the Compute tile. Intel has plenty of options for the Graphics tile, including its own Intel 4 foundry node.
Source: HXL (Twitter)
Add your own comment

55 Comments on Intel "Nova Lake" Compute Tile Die-sizes Surface

#1
ThomasK
Funny how Intel tounted 18A as the next big thing, yet it got relegated only to mobile parts. Guess TSMC will be running to the bank with the checks for fabbing both Intel's CPUs and GPUs after all.
Posted on Reply
#2
ncrs
ThomasKFunny how Intel tounted 18A as the next big thing, yet it got relegated only to mobile parts. Guess TSMC will be running to the bank with the checks for fabbing both Intel's CPUs and GPUs after all.
They might do a hybrid rollout like with Panther Lake's iGPU which builds low end on Intel 3 and high end on TSMC.
Since Nova Lake is supposed to span most desktop markets creating lower end SKUs with Intel 18A CPU tile might make financial sense where TSMC N2 wouldn't have.
The flexibility of tile/chiplet designs at work ;)
Posted on Reply
#3
BlueScreenJunky
ThomasKFunny how Intel tounted 18A as the next big thing, yet it got relegated only to mobile parts. Guess TSMC will be running to the bank with the checks for fabbing both Intel's CPUs and GPUs after all.
I suspect laptops and servers are the only important markets now, and Intel is using their own fab for both (although Xeon 600 is using Intel 3 and not 18A).
Posted on Reply
#4
ThomasK
ncrsThey might
Lip Butt Tan might also start telling shareholders that they are gonna have to wait for 14A instead.
BlueScreenJunkyI suspect laptops and servers are the only important markets now, and Intel is using their own fab for both (although Xeon 600 is using Intel 3 and not 18A).
18A still has poor yields, so using it only for mobile parts, which are generally smaller and easier to fab, makes sense.
Posted on Reply
#5
rattlehead99
ThomasKFunny how Intel tounted 18A as the next big thing, yet it got relegated only to mobile parts. Guess TSMC will be running to the bank with the checks for fabbing both Intel's CPUs and GPUs after all.
When you have a contract with a fab you can't just ask them to do one generation, that's absolutely not worth it for them. Intel is very likely to be contractually obligated to use TSMC for a while.
It's the same reason AMD went with Global Foundry's vastly inferior 14nm process node instead of using TSMC's 16nm. They weren't making almost any CPUs at the time, but had a contract with GloFo and had to adhere to it.
Posted on Reply
#6
igormp
btarunrEstimates put the regular Compute tile to measure around 110 mm², while the premium variant with bLLC is estimated to measure over 150 mm². This 36% increase in die-area is entirely from the enlarged last-level cache.
So the bLLC is actually part of the compute tile, instead of being another tile or stacked in some way?
That's disappointing in a way. This variant should be really expensive considering that SRAM scales awfully with node size and represents a lot of die area in such cases.

Even worse considering that there is that rumored dual compute tile config, so that means 300m²+ for those tiles alone. I wonder how big the actual chip and its socket will be.
btarunrthe DDR5 memory controllers
I thought that one of the complaints that folks had with ARL was that the memory controller was not on the compute tile, and thus gaming performance suffered a bit, and due to that Intel would move the memory controllers back into the compute tile. Seems like I did the wrong assumption, but having this in the SoC tile makes way more sense anyway.
ThomasKFunny how Intel tounted 18A as the next big thing, yet it got relegated only to mobile parts. Guess TSMC will be running to the bank with the checks for fabbing both Intel's CPUs and GPUs after all.
Being the devil's advocate here, I assume that 18A is pretty efficient, but doesn't scale that well with power/clocks, and thus it should not be that great for desktop and other higher-clocking parts. If it still manages to be really efficient with a competitive pricing, it might be a great option for mobile options (not only laptops, but phones as well), as well as dense server options, which is arguably a way larger market.
PTL will be where Intel shows if the above is true, given that it parted away with lots of decisions that made LNL hella efficient, but still touting the same efficiency at ARL levels of performance. If that pans out, I can see other players going for 14A.

Rumors point that Apple will use 18A for lower-end M-chips, and a more recent one says that MediaTek will be using 14A, those are really big wins and I can see other customers such as Qcom and AMD (for their dense/laptop offerings) following suit if TSMC doesn't reduce their pricing.
ncrsThey might do a hybrid rollout like with Panther Lake's iGPU which builds low end on Intel 3 and high end on TSMC.
This is also a great marketing strategy for their EMIB offerings, going all "hey, we can package all kinds of stuff without issues" lol
ThomasK18A still has poor yields, so using it only for mobile parts, which are generally smaller and easier to fab, makes sense.
AFAIK it's doing better than Samsung's current node, but I'm not sure if that says much lol
Agreed that mobile parts make more sense, adding they are not only cheaper to produce but are higher volume as well.
Posted on Reply
#7
Ed_1
So if this is on TSMC node, it won't have backside power delivery which was one of key new features.
Posted on Reply
#9
GenericUsername2001
Its interesting that Nova Lake & Zen 6 are going to be (partially) fabbed on the same TSMC N2 process. Will allow some nice apples to apples comparisons. The most recent rumors have Zen 6 using a 76 mm2 chiplet with 12 cores & 48 MB of L3 cache on board, for comparisons sake.
Posted on Reply
#10
TristanX
ThomasKFunny how Intel tounted 18A as the next big thing, yet it got relegated only to mobile parts. Guess TSMC will be running to the bank with the checks for fabbing both Intel's CPUs and GPUs after all.
Mobile parts are 'only' 80% of produced chips
Posted on Reply
#11
Evrsr
ThomasKFunny how Intel tounted 18A as the next big thing, yet it got relegated only to mobile parts. Guess TSMC will be running to the bank with the checks for fabbing both Intel's CPUs and GPUs after all.
This has been common for Intel process to be adopted by mobile first. It was the case with Broadwell (14nm) and they also tried but failed with Cannon Lake on 10nm.
Either due to power or yield issues that only make it feasible for small dies.

18A also has the issue that unlike the development that was flipped chip designs, we have gone 50% back where there are now signal layers between the transistors and heatsink. This negates some of the benefit of lower power due to BSPDN.
Modern designs that are predicated in very high clocks, voltages and have a big number of hotspots will take some work to move to 18A or the upcoming BSPDN of TSMC.

But it is a path worth taking, as it also allows redesign of the SRAM cells akin to allow for a very big increase in SRAM density that has been faltering for many years.
Posted on Reply
#13
JohH
It seems much larger than expected. I posited ~95mm2 for normal tile and ~120mm2 for bLLC tile.
Posted on Reply
#14
LastDudeALive
If the bLLC performs well, I think I'll upgrade from my 5800X3D.
Posted on Reply
#15
Tomorrow
Get ready for high prices. Panther Lake (the 12c iGPU variant) is already almost as expensive as Strix Halo.
I doubt Nova Lake will be cheap either with these die sizes.
Posted on Reply
#16
igormp
ThomasKMobile ≠ notebooks, and even combined, they are not anywhere near 80% of all chips produced.

Mobile alone is well under 30%.

www.mordorintelligence.com/industry-reports/semiconductor-industry-landscape
Where did you get this number from from that source of yours? I could not easily get their mentioned detailed PDF.
From the report page it lists the entirety of the CONSUMER industry under those 30%, including mobile phones, laptops and desktop devices.
The remaining of those 70% include tons of things that do not use the latest and greatest nodes, with stuff like MCUs and power regulators, which wouldn't use a modern node by any means, so you can't compare it to what Intel is targeting.
Posted on Reply
#18
XeonsNOpterons
This is interesting news, though I am pretty sure Nova Lake is supposed to be 18A as well, just like Panther Lake, Clearwater Forest and Diamond Rapids mid this year. Die sizes make sense considering Intel P-cores are larger and more powerful than Zen 5 (and Coyote Cove will likely maintain that lead against Zen 6) and that Intel is packing 16 E-cores in the compute tile. I expect the E-cores to be pretty powerful, even more so than Arrow Lake's. Excited to see Nova Lake and Zen 6's launches, will likely be a tight competition. I predict this time around Intel will take the lead in both compute/productivity and gaming. Especially that 52-core chip, easily should destroy the 24-core Zen 6 flagship in compute/productivity.
Posted on Reply
#19
Tomorrow
XeonsNOpteronsI predict this time around Intel will take the lead in both compute/productivity and gaming.
Productivity - likely yes.
Gaming - unlikely considering the kind of deficit they'll have to claw back to first match Zen 5 X3D and then Zen 6 and then Zen 6 X3D on top of that.
If they were equal today then yes they would have a good chance, but they're not. They're far behind now. Double digit percentages. I forgot the exact number, but it is bad.
XeonsNOpteronsEspecially that 52-core chip, easily should destroy the 24-core Zen 6 flagship in compute/productivity.
The gap will likely not be as big as 24 vs 52 would suggest considering more than half of these cores are far weaker than Zen 6 cores.
Also if AMD really wanted they could make a Zen 6c (dense) core available on AM5 by cutting some of the cache.
This will likely get close to the 52 core count of Intel.

Whether they'll do it or not remains to be seen as so far dense cores have only been used in server primarily and mobile occasionally.
Posted on Reply
#20
Wirko
Evrsr18A also has the issue that unlike the development that was flipped chip designs, we have gone 50% back where there are now signal layers between the transistors and heatsink. This negates some of the benefit of lower power due to BSPDN.
No, not really. In the usual flip-chip design, there's the entire substrate between the transistors and the heatsink. It can be thinned down but I don't know if manufacturers ever do that. However, they do that when stacking dies, which includes building the BSPDN structure.

Here's an illustration from Cadence. First is flip-chip, second is a chip with BSPDN (BSM = backside metal), third is a stack of both.
Posted on Reply
#21
XeonsNOpterons
TomorrowThe gap will likely not be as big as 24 vs 52 would suggest considering more than half of these cores are far weaker than Zen 6 cores.
Also if AMD really wanted they could make a Zen 6c (dense) core available on AM5 by cutting some of the cache.
This will likely get close to the 52 core count of Intel.

Whether they'll do it or not remains to be seen as so far dense cores have only been used in server primarily and mobile occasionally.
Because Intel will not have SMT just like Arrow Lake, it is better to compare thread counts. Intel's 285K beats/matches the AMD 9950X in compute/productivity overall. That is 24 Intel threads vs 32 AMD threads. So despite a thread count advantage, AMD barely matches. The 485K (presumably its name) will have 52 threads vs AMD's 10950X's 48 threads. Considering I expect at minimum a IPC boost that is similar between both companies (around 15%), then the 485K should win well. But the rumors are likely to give Intel a much larger IPC gain from Arrow Lake than Zen 6 vs Zen 5, so that gap will likely be even bigger. I just wish both teams offered 4-channel on the mainstream already.

Do not underestimate the E-cores, especially the ones which come in Nova Lake. My current estimates based on aggregate FP data from PassMark has Arrow Lake E-cores around 80% as good as Zen 5 cores in IPC. Nova Lake E-cores are going to have easily double digit IPC gains according to rumors, likely to facilitate reunification of P and E cores with Titan Lake.
Posted on Reply
#22
AntiX3DGlazer
ThomasKFunny how Intel tounted 18A as the next big thing, yet it got relegated only to mobile parts. Guess TSMC will be running to the bank with the checks for fabbing both Intel's CPUs and GPUs after all.
2 of the tiles found on Nova Lake are built on 18A (SoC and Graphics iirc).
ThomasK

semiwiki.com/wikis/industry-wikis/tsmc-n2-process-technology-wiki/
N2 does not have BSPDN. that is incorrect. BSPDN is being introduced in A16, which includes what TSMC calls Super Power Rail. www.tsmc.com/english/dedicatedFoundry/technology/logic#l_A16
Posted on Reply
#23
LastDudeALive
TomorrowProductivity - likely yes.
Gaming - unlikely considering the kind of deficit they'll have to claw back to first match Zen 5 X3D and then Zen 6 and then Zen 6 X3D on top of that.
If they were equal today then yes they would have a good chance, but they're not. They're far behind now. Double digit percentages. I forgot the exact number, but it is bad.

The gap will likely not be as big as 24 vs 52 would suggest considering more than half of these cores are far weaker than Zen 6 cores.
Also if AMD really wanted they could make a Zen 6c (dense) core available on AM5 by cutting some of the cache.
This will likely get close to the 52 core count of Intel.

Whether they'll do it or not remains to be seen as so far dense cores have only been used in server primarily and mobile occasionally.
Arrow Lake already caught up to Zen 5 in gaming, the "9" and "7" tiers are practically within the margin of error for the CPU-bottlenecked 720p benchmarks (the 245K is probably a % point or two below, but it wasn't included in the 9850X3D review, possibly hasn't been comprehensively re-tested yet). Zen 5 was barely even an improvement over Zen 4. At 1080p or above, and/or with anything other than a 4090/5090 (i.e. the world) you would not be able to tell the difference between a 245K and a 9950X in any game. Only the 3D cache provides any practical improvement, which is precisely why Intel is introducing bLLC.



And you're completely wrong about the productivity benchmarks, the e cores provide far more benefit for the space they take up than a fat Zen c core.

The 285K essentially matches the 9950X in the entire application test suite. Doing so with 8 less threads (and no AVX-512) shows that e-cores are a much better way to get multi-threading power. If an 8P+16e CPU is matching a 16C/32t CPU, a 16P+36e processor will absolutely wipe the floor with a 24C/48t CPU (unless Zen 6 brings a miraculous IPC improvement).
Posted on Reply
#24
AntiX3DGlazer
LastDudeALiveThe 285K essentially matches the 9950X in the entire application test suite. Doing so with 8 less threads (and no AVX-512) shows that e-cores are a much better way to get multi-threading power. If an 8P+16e CPU is matching a 16C/32t CPU, a 16P+36e processor will absolutely wipe the floor with a 24C/48t CPU (unless Zen 6 brings a miraculous IPC improvement).
That speaks to the power of the CPU cores in ARL, given the otherwise pretty piss-poor hand that the memory subsystem has to play.
To be clear, Zen cores are nearly half the size of comparable Coves, and the Monts are slower while taking up around the same die area (comparing half a Zen 5 8-core CCX with cache to an E-core cluster in Arrow Lake desktop). Intel is paying a price for its performance, and that's cost.
Posted on Reply
#25
LastDudeALive
AntiX3DGlazerThat speaks to the power of the CPU cores in ARL, given the otherwise pretty piss-poor hand that the memory subsystem has to play.
To be clear, Zen cores are nearly half the size of comparable Coves, and the Monts are slower while taking up around the same die area (comparing half a Zen 5 8-core CCX with cache to an E-core cluster in Arrow Lake desktop). Intel is paying a price for its performance, and that's cost.
Right now the 245K and 265K are $100 cheaper than the Zen 5 CPUs they match in productivity (9700X and 9900X, respectively). Maybe Intel is sacrificing its margins, but the 245K and 265K are the best options for multi-threaded work from a price/performance standpoint rn. And ultimately, that's what matters. Maybe that math will change with Zen 6 & Nova Lake, maybe it won't, we'll see.
Posted on Reply
Add your own comment
Nov 17th, 2026 13:57 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts