Intel Core Ultra 300 Series "Panther Lake-H" Leaks: 18 CPU Cores, 12 Xe3 GPU Cores, and up to 45 Watt TDP

[AleksandarK]

Details have emerged about Intel's upcoming "Panther Lake" processors, set to be the third generation of Core Ultra mobile chips. Called the Core Ultra 300 series, these CPUs are expected to succeed "Lunar Lake". According to recent leaks, Panther Lake-H will be manufactured using Intel's cutting-edge 18A process node. The chips are said to feature a combination of Cougar Cove P-Cores, Skymont E-Cores, and Xe3 (Celestial) integrated graphics. This architecture builds upon Intel's hybrid core design, refining it for even better performance on mobile devices. The leaked information suggests a range of configurations for Panther Lake-H, the high-perfomance variant of the lineup. These include models with varying core counts and power envelopes, from efficient 25 W parts to more interesting 45 W options. Notably, some SKUs reportedly feature up to 18 cores in total, combining P-cores, E-cores, and LP E-cores in a five-tile package. This is an increase from previously believed 16 cores.




Here are the leaked SKUs:

PTL-H 4+8+4+4Xe 25 W
PTL-H 4+8+4+12Xe 25 W
PTL-H 4+0+4+4Xe 25 W
PTL-H 6+8+4+4Xe 45 W
PTL-H 6+8+4+12Xe 28 W

One of the most intriguing aspects of the leak is the inclusion of powerful integrated graphics in some models. Certain SKUs are said to boast 12 Xe3 GPU cores, even in relatively low-power 25 W and 28 W configurations. This could significantly boost graphics performance in thin and light laptops. Some high-wattage SKUs, like the 45 W model, should be configured with higher clocks. While these specifications are still subject to change before the official release, they provide an exciting glimpse into Intel's future mobile offerings. The Panther Lake series is expected to debut in the second half of 2025 if everything goes according to plan at Intel HQ.

View at TechPowerUp Main Site | Source

 

[kondamin]

So TSMC late Q4 2025 tiny trickle on to the market after CES.

anyone know if there are plans for an n100 successor?

 

[john_]

4 P cores. Intel back in 2012.

OK...OK..... there are 6P cores models too. How exciting!!!
That 6P 12Xe model looks promising for gaming without discrete graphics card.
That 4P 12Xe model goes directly to handhelds.

 

[LittleBro]

Do they long for another scheduler issues or what?

Having 3 different types of core clusters in one CPU is crazy. AFAIK, they still haven't 100% solved issues relating to putting heavy load on e-cores on existing hybrid CPUs.

What's the point of LP-e-cores? Weren't regular e-cores meant to be P-cores with lower power and higher efficiency? They overcomplicate things. This might strike back on them badly.

And talking about Xe3 Celestial GPU units in CPUs in 1st half of 2025? After that voving from Xe Arc to Xe2 Battlemage took them almost 2 years?

 

[usiname]

Do they long for another scheduler issues or what?

Having 3 different types of core clusters in one CPU is crazy. AFAIK, they still haven't 100% solved issues relating to putting heavy load on e-cores on existing hybrid CPUs.

What's the point of LP-e-cores? Weren't regular e-cores meant to be P-cores with lower power and higher efficiency? They overcomplicate things. This might strike back on them badly.

And talking about Xe3 Celestial GPU units in CPUs in 1st half of 2025? After that voving from Xe Arc to Xe2 Battlemage took them almost 2 years?

By the time they release Battle Mage, they will mark 3 years since the first released Alchemist

 

[Chaitanya]

4 P cores. Intel back in 2012.

OK...OK..... there are 6P cores models too. How exciting!!!
That 6P 12Xe model looks promising for gaming without discrete graphics card.
That 4P 12Xe model goes directly to handhelds.

These are laptop cpus targetting ultra thins configs and there are a lot of CPUs right now in that segment with 2P cores.

 

[ncrs]

Do they long for another scheduler issues or what?

Having 3 different types of core clusters in one CPU is crazy. AFAIK, they still haven't 100% solved issues relating to putting heavy load on e-cores on existing hybrid CPUs.

What's the point of LP-e-cores? Weren't regular e-cores meant to be P-cores with lower power and higher efficiency? They overcomplicate things. This might strike back on them badly.

Meteor Lake has 2 LPE-cores that live outside of the main core cluster so that P- and E-cores can be wholly powered down, including the uncore (ring bus). The downside of this approach is increased latency for cores living outside of the ring bus - ~200ns vs. ~42ns.
Lunar Lake removes LPE cores, but moves E-cores outside of the ring bus, which is now only for P-cores. The latency penalty will most likely remain similar, but its impact might be higher due to basically half the cores being outside vs. just 2 in MTL.
Panther Lake seems to reintroduce dedicated LPE cores. The question is whether E-cores remain outside of the ring bus or are they moved back to the main core cluster like in Meteor Lake?
In any case, scheduling problems will be present in all of mentioned generations albeit having different characteristics which makes it harder to support properly.

 

[Melvis]

Panther Lake? what a dumb ass name! literally makes no sense.....

 

[ncrs]

Panther Lake? what a dumb ass name! literally makes no sense.....

Intel usually uses real geographical locations for its codenames.

 

[Wirko]

Panther Lake? what a dumb ass name! literally makes no sense.....

The first one was the only one that did. Sky Lake = cloud.

 

[Dristun]

Intel usually uses real geographical locations for its codenames.

And the funny part is there are usually so many! I hope the Lakes never end.

 

[john_]

These are laptop cpus targetting ultra thins configs and there are a lot of CPUs right now in that segment with 2P cores.

You mean "a lot of INTEL CPUs". Yeah, Intel plays a lot with marketing here and this is the reason why these hybrid CPUs where created in the first place. Intel sells "Cores" and consumers get 2P cores and a tone of E cores, in the near future also Extra Energy efficient E cores that their only purpose is to help Intel win battery efficiency benchmarks.

 

[Darmok N Jalad]

These are laptop cpus targetting ultra thins configs and there are a lot of CPUs right now in that segment with 2P cores.

I have one of those 2P core CPUs in my work laptop, and I can't believe they call it an i7. It's painfully slow at relatively simple tasks sometimes.

You mean "a lot of INTEL CPUs". Yeah, Intel plays a lot with marketing here and this is the reason why these hybrid CPUs where created in the first place. Intel sells "Cores" and consumers get 2P cores and a tone of E cores, in the near future also Extra Energy efficient E cores that their only purpose is to help Intel win battery efficiency benchmarks.

Yes, it feels a bit like a marketing scam. Honestly, the 4P+4E or Meteor Lake is at least an improvement over the 2P+8E it's replacing.

 

[TheinsanegamerN]

These are laptop cpus targetting ultra thins configs and there are a lot of CPUs right now in that segment with 2P cores.

Ultra thins are not using 25W CPUs. Check those numbers again.

By the time they release Battle Mage, they will mark 3 years since the first released Alchemist

And Alchemist was already over a year late.

4 P cores. Intel back in 2012.

OK...OK..... there are 6P cores models too. How exciting!!!
That 6P 12Xe model looks promising for gaming without discrete graphics card.
That 4P 12Xe model goes directly to handhelds.

The Xe cores are the only interesting thing about this gen. Everything else looks like major suckage.

 

[AusWolf]

Even two different core types are too much in a single CPU, and now we have three? Thanks, but no, thanks. Give me Bartlett Lake instead.

 

[Minus Infinity]

4 P cores. Intel back in 2012.

OK...OK..... there are 6P cores models too. How exciting!!!
That 6P 12Xe model looks promising for gaming without discrete graphics card.
That 4P 12Xe model goes directly to handhelds.

Don't forget Panther Lake gets Celestial (Xe3), so should be even much stronger than Battlemage (Xe2) in Lunar Lake, which is showing to be 30-50% faster than Alchemist and faster than 890M (according to Intel).

 

[Prima.Vera]

Totally confused about the new naming scheme...

 

[Minus Infinity]

Totally confused about the new naming scheme...

100 = Mediocre Lake
200 = Arrow Lake
300 = Panther Lake
400/500 = Arrow Lake refresh (probably) (not sure if Bartlett Lake still is coming and gets same naming convention so it could be 400)
500/600 = Nova Lake

Basically starting new gen of architecture at 100 and increments 100 for each new gen. You can still recognise i5, i7 and i9 equivalents via Ultra 5, 7 or 9 tag.

 

[john_]

Don't forget Panther Lake gets Celestial (Xe3), so should be even much stronger than Battlemage (Xe2) in Lunar Lake, which is showing to be 30-50% faster than Alchemist and faster than 890M (according to Intel).

According to Intel it is probably faster than 4090. And also unfortunately seeing AMD's official benchmarks about Zen 5 and what in the end we got, AMD is slipping in marketing lies together with Intel.

That being said, I do expect Intel to win the next round of iGPUs, maybe not in stability, power efficiency and drivers, but in performance. I think MSI has seen what Xe2 could do and the first Claw was just a test drive, while also a way to please Intel. It wouldn't be the first time AMD is sleeping while being ahead of the competition, only to end up behind chasing. AMD needs to integrate Infinity Cache in it's iGPUs or do something similar to the good old Sideport Memory in their future designs, because I think they hit a ceiling with their architecture and the bandwidth they can get from DDR5.

 

[MaMoo]

4 P cores. Intel back in 2012.

OK...OK..... there are 6P cores models too. How exciting!!!
That 6P 12Xe model looks promising for gaming without discrete graphics card.
That 4P 12Xe model goes directly to handhelds.

Frankly, that decade of quad cores was when Intel ruled the roost. Ever since core counts went up, Intel has lost its edge.

 

[Darmok N Jalad]

According to Intel it is probably faster than 4090. And also unfortunately seeing AMD's official benchmarks about Zen 5 and what in the end we got, AMD is slipping in marketing lies together with Intel.

That being said, I do expect Intel to win the next round of iGPUs, maybe not in stability, power efficiency and drivers, but in performance. I think MSI has seen what Xe2 could do and the first Claw was just a test drive, while also a way to please Intel. It wouldn't be the first time AMD is sleeping while being ahead of the competition, only to end up behind chasing. AMD needs to integrate Infinity Cache in it's iGPUs or do something similar to the good old Sideport Memory in their future designs, because I think they hit a ceiling with their architecture and the bandwidth they can get from DDR5.

I don't know how you can call it a "win" if stability, efficiency, and drivers are all issues. Gaming is about the experience, and winning some benchmarks only gets you so far. It will get you a sale and then a return. OEMs will love that. The issue with any IGP products is bandwidth and cost. Once you start adding that much complexity to the chip, you drive up die size and cost, only to still lose out to a mobile dGPU anyway. I guess we'll see with Strix Halo, which forgoes that convention, but it's not going to be cheap as an "iGPU solution." It's more like a console solution on a desktop.

 

[john_]

I don't know how you can call it a "win" if stability, efficiency, and drivers are all issues. Gaming is about the experience, and winning some benchmarks only gets you so far. It will get you a sale and then a return. OEMs will love that.

When talking about stability and drivers, I didn't meant that half stuff will be crushing. I mean that there might be some games that don't work as expected, probably older ones that few care about. I bet Intel is doing nicely with their drivers lately, because I haven't seen articles about angry ARC users. As for efficiency, the next Claw will have a bigger battery, so for most people this will be an OK fix.

The issue with any IGP products is bandwidth and cost. Once you start adding that much complexity to the chip, you drive up die size and cost, only to still lose out to a mobile dGPU anyway. I guess we'll see with Strix Halo, which forgoes that convention, but it's not going to be cheap as an "iGPU solution." It's more like a console solution on a desktop.

Yeah and AMD is sleeping. I don't know what they wait first to happen, to start improving the only stuff that they have and it sells by itself. Their APUs. Adding more CUs is the easy fix, how about improving the architecture or the platform? I have said it before and I think I will keep repeating it that, I believe they finally put some effort on RT performance in RDNA4, only to not lose SONY. But in laptops and handhelds they either wait for Steam to ask them for a new APU, or wait to lose the first place to Intel to start throwing money on R&D. Adding some Infinity cache or implementing something like the Sideport Memory I mentioned, something that could be done easily 15 years ago and even supported on cheap motherboards I doubt it will drive costs up or add complexity. Infinity cache is already something done on their GPUs, so they don't really have to figure out something new, Sideport Memory is just a VRAM chip on the motherboard. It was done 15 years ago, it can be done easier today.

 

[Solaris17]

I think MSI has seen what Xe2 could do and the first Claw was just a test drive

Incidentally I think it was IGN that MSI told they would be remaking the 7 with lunar lake

 

[remixedcat]

Do they long for another scheduler issues or what?

Having 3 different types of core clusters in one CPU is crazy. AFAIK, they still haven't 100% solved issues relating to putting heavy load on e-cores on existing hybrid CPUs.

What's the point of LP-e-cores? Weren't regular e-cores meant to be P-cores with lower power and higher efficiency? They overcomplicate things. This might strike back on them badly.

And talking about Xe3 Celestial GPU units in CPUs in 1st half of 2025? After that voving from Xe Arc to Xe2 Battlemage took them almost 2 years?

and digital audio workstations are still having issues w the current big.LITTLE and P and E core architecture! let alone introducing this new thing as well... bitwig just got stuff stable w big.LITTLE and ableton and FL studio still struggle and have crashes and DSP issues! This will just make it that much worse!

 

[Wirko]

100 = Mediocre Lake
200 = Arrow Lake
300 = Panther Lake
400/500 = Arrow Lake refresh (probably) (not sure if Bartlett Lake still is coming and gets same naming convention so it could be 400)
500/600 = Nova Lake

Basically starting new gen of architecture at 100 and increments 100 for each new gen. You can still recognise i5, i7 and i9 equivalents via Ultra 5, 7 or 9 tag.

So you basically multiply by 10, add 14000, take away the "Ultra" and put back the "i", and here we are, the world is a familiar place once again! It's almost as simple as the conversion from marketing nanometers to real nanometers.