Panther Lake notebook processor: Better e-cores, new GPU, big question marks
Intel's Core Ultra 300, made with 18A process, aims to merge Lunar Lake and Arrow Lake series. Key parameters are still missing.
Intel is proud of its own production technology 18A and shows the mobile processor Panther Lake in its own semiconductor plant.
(Image: Intel)
At this year's Tech Tour event, Intel revealed the technical details of the Panther Lake processor generation, which will be launched in January for the 2026 model year – In all likelihood as the Core Ultra 300. The CPUs will only be available for notebooks and mini PCs that use soldered mobile chips. Intel is not planning a desktop version – Users will have to make do with an Arrow Lake refresh.
Panther Lake is intended to end the two-track nature of the 2025 processors. The Core Ultra 200 generation consists of not one, but two series that have little in common technically. Core Ultra 200U/H (Arrow Lake) are the pin-compatible successors to Core Ultra 100 (Meteor Lake), which combine up to six P-cores (performance) with up to eight E-cores (efficiency), garnished with two LP-E-cores (low-power conversion of the E-cores). The more efficient specialty Core Ultra 200V (Lunar Lake) only has four P and LP-E cores each, but a more powerful graphics unit. It is the only Intel series to date with a Neural Processing Unit (NPU), which is suitable for Microsoft's Copilot+ AI logo with over 40 TOPS.
(Image:Â Mark Mantel / heise medien)
Panther Lake consists of several chiplets, called tiles by Intel. The partitioning has changed again in this generation: Intel has conceptually combined the separate chiplets for system-on-chip (SoC) functions and CPU cores in Meteor Lake and Arrow Lake into one. There are two expansion stages: a weaker one with four P and LP-E cores each (i.e. similar to Lunar Lake) and a stronger one with four P, eight E and four LP-E cores. Intel does not envisage even larger variants. Compared to Meteor and Arrow Lake, two P cores are "missing", which the additional LP-E cores do not compensate for. The latter have the same architecture as the normal E cores, but are electrically connected to a different power island on the chip. In addition, they do not have direct access to the fast ring bus and the large shared caches that all other cores share.
(Image:Â Intel)
No information on frequencies
Considering the data sheet, it is questionable whether Panther Lake will really be at least as fast on the CPU side as all previous Core Ultra 200s, as Intel promises. Both core architectures, i.e. Cougar Cove (P cores) and Darkmont (E cores), roughly correspond to their predecessors, Lion Cove (P) and Skymont (E), as used in Lunar Lake and Arrow Lake. Apart from changes such as 18 instead of 12 MByte level 3 cache for the P cores and further minimal fine-tuning, it appears to be simply a matter of porting from the previous production process (TSMC N3) to a new one (Intel 18A).
Videos by heise
Intel has not yet revealed a direct performance comparison. At the Tech Tour, the manufacturer did show impressive curve diagrams showing how the cores perform in terms of performance per watt with a higher clock rate and energy per clock cycle. However, the axes were unlabeled – which robs the diagram of any resilience. Higher absolute performance per core inevitably requires higher clock frequencies with an almost identical core architecture, but Intel has so far been completely silent on specific values.
(Image:Â Intel)
Without these, it is not possible at present to assess what the brand new 18A manufacturing process, in which the Panther Lake chiplet with the CPU cores rolls off the production line, is capable of. Intel boldly describes 18A as the world's most advanced manufacturing process and the first in the 2-nanometĂŞr class to be ready for series production. However, this statement in itself means little.
Whatever happens, the benchmark that the Panther Lake P-cores manufactured in 18A have to meet is 5.1 GHz. This is how high the P-cores in the Core Ultra 9 288V, the fastest expansion stage of Lunar Lake, clock, which are produced by chip contract manufacturer TSMC using N3P technology. And the 5.1 GHz is not enough to reach the single-threaded performance regions in industry-standard benchmarks such as Cinebench or Geekbench, which Apple's M4 has been achieving for a year and which Qualcomm's Snapdragon X2 Elite will probably also come close to from 2026. Incidentally, both are also 3-nanometer chips from TSMC.
