OnProcessor codenamed Lunar LakeAfter successfully demonstrating the performance and energy efficiency of the x86 architecture, Intel plans to launchPanther Lake computing platformNot only does it maintain this advantage, but it also draws on previous design experience to further create a more flexible platform design, allowing OEM manufacturers to create diverse products to meet different market demands. It also opens up new possibilities for the development of the x86 architecture in the AI PC market.
Flexible chiplet design to meet diverse market demands
Departing from the previous Lunar Lake design experience, which packaged memory directly on the chip, the Panther Lake platform adopts a more flexible design. Not only does it eliminate pre-packaged memory on the chip, but it also provides OEMs with the flexibility to choose between LPDDR5x or DDR5 memory in specific versions. It also officially supports LPCAMM modules and allows customers to choose the appropriate PMIC for different device types.
The Panther Lake platform also utilizes a chiplet design, incorporating Intel's process and packaging technology innovations. The CPU tile will be manufactured using Intel's 18A process featuring a RibbonFET structure and PowerVIA backside power delivery, while also utilizing Foveros 2.5D and 3D packaging technologies. Furthermore, the Scalable Fabric Gen 2 architecture, which supports multiple IPs, enhances die combination flexibility.
Compared to Lunar Lake, which only offers a single design, Panther Lake will be available in three CPU and GPU combinations. This includes a top-of-the-line 16-core CPU (4P+8E+4LP-E) paired with a 12-core Xe GPU, a 16-core CPU (4P+8E+4LP-E) paired with a quad-core Xe GPU that can be paired with a discrete graphics card, and a mainstream 8-core CPU (4P+4LP-E) paired with a quad-core Xe GPU.
In the version that can be paired with an additional independent graphics card, 8 PCIe Gen 4 and 12 PCIe Gen 5 lanes will be provided through the IO Tile, allowing OEM manufacturers to allocate 8 PCIe Gen 5 lanes to pair with mid-to-high-performance independent graphics cards, while at the same time balancing performance with the built-in Xe GPU in daily use.
New CPU core design further improves energy efficiency
The Panther Lake platform uses the Cougar Cove architecture for its P Core, while the Darkmount architecture for its E Core and LP-E Core. Because the platform design focuses on energy efficiency, the P Core configuration has a maximum of 8 cores. However, the LP-E Core is adjusted based on the design experience of the Lunar Lake platform, and all versions use a 4-core configuration to meet the needs of most daily workloads.
Whether it's Cougar Cove or Darkmount, Intel has enhanced operations like memory disambiguation, branch prediction, and TLB, while also leveraging artificial intelligence to enhance energy management. This allows the Panther Lake platform to achieve superior energy efficiency over Lunar Lake and surpass the multi-core performance of Arrow Lake-H with the same core count.
In addition, Intel has further enhanced the CPU memory subsystem. Cougar Cove maintains an independent L2 cache per core and shares an 18MB L3 cache with the E Core. Darkmount continues the quad-core shared L2 design, so the 16-core version will consist of two sets of quad-core E Cores sharing L2 to form an 8-core E Core. As for the LP-E Core, although it does not share L3 with the P Core and E Core, it can reduce external DRAM access by sharing 8MB of Memory Side-Cache with the I/O and media engines.
The Darkmount core design achieves peak performance nearly identical to Raptor Cove at extremely low power consumption. Consequently, Intel has reordered the core workflow on the Panther Lake platform, prioritizing tasks with the lowest possible power consumption by moving from LP-E Core to E Core and finally to P Core. Enhanced Thread Direct technology also disables the LP-E Core in scenarios requiring CPU and GPU performance balance (such as gaming), freeing up the GPU for increased power and overall performance.
Xe3 GPU and NPU 5 design, corresponding to higher AI computing power
The Panther Lake platform is the first product to introduce the Xe3 GPU architecture, but Intel still calls it the Arc B series and has not changed the name to the Arc C series. It is available in two versions: 12-core Xe Core and 4-core Xe Core. Both support Intel XeSS technology, including the newly announced XeSS multi-frame generation function. The 12-core Xe Core version will provide a sufficient gaming experience while also bringing excellent GPU heterogeneous acceleration effects, providing up to 120 TOPS computing power performance.
As for the NPU, the NPU 5 has been introduced. Compared with the NPU 4 in the Lunar Lake platform, the performance improvement is not significant, only reaching 50 AI TOPS, but the focus is on reducing the unit area of the architecture and improving energy efficiency. At the same time, support for INT8 precision has been added. However, Intel also admitted that the NPU 5's performance is only reduced to 9% in Softmax operations compared to the NPU 4, but it can improve other mainstream AI inference applications by at least 1.1 times.
Intel did not elaborate on why the NPU performance was not significantly improved, but the explanation was that since the NPU usage rate in the Copilot+ PC ecosystem depends on Microsoft, Intel believes that overall heterogeneous computing performance is more important than focusing on the performance of a single NPU.
Wi-Fi 7 and Bluetooth 6.0 connectivity, corresponding to higher transmission efficiency
In terms of connectivity, the Panther Lake platform will integrate an Intel wireless network chip corresponding to the Wi-Fi 7 R2 specification and support the 6GHz frequency band and 320MHz channel bandwidth. Compared to Wi-Fi 6, it will bring higher transmission throughput. It also supports technologies such as multi-link operation, allowing devices to connect to different frequency band networks simultaneously, thereby achieving higher transmission efficiency and connection stability.
The Bluetooth connection part complies with the Bluetooth 6.0 specification and supports LE Audio and Auracast broadcasting technologies, allowing users to share audio content with more people or connect multiple audio devices. It also supports channel detection function, allowing devices to accurately determine relative positions and even correspond to applications such as digital keys.
IPU 7.5 image processor enhances the video camera application experience
The DSP corresponding to the video camera is the IPU 7.5, which has artificial intelligence-based image noise reduction, Local Tone Mapping, and hardware-accelerated HDR. It also supports three cameras and 16MP 120fps slow-motion shooting, with a power consumption of only 1.5W. The multimedia engine Xe Media Engine now supports AVC, AV1 and Sony's XAVC encoding.
Design differences from Lunar Lake
Compared to the Lunar Lake platform design, the Panther Lake platform clearly offers significantly greater design flexibility. For example, the design eliminates the need to package memory directly on the chip, allowing OEMs to flexibly choose memory specifications and capacities. It also offers three different CPU and GPU combinations, and can even be paired with a discrete graphics card, allowing OEMs to create appropriate specifications for different product positioning.
The CPU core configuration is also different from the Lunar Lake design, which only uses 4 P Cores and 4 LP-E Cores. The Panther Lake platform can support up to 4 P Cores, 8 E Cores and 4 LP-E Cores. Through a larger L3 cache and memory subsystem optimization, overall performance is improved, while also strengthening heterogeneous computing scheduling capabilities.
Comparison with Qualcomm and Apple processors
Judging from the design of the Panther Lake platform, Intel clearly aims to compete with Arm-based processors in terms of energy efficiency, such as Qualcomm's Snapdragon X Elite series and Apple's M4 series processors. Although the NPU computing power of these processors is advertised as up to 50 TOPS, compared to the Snapdragon X Elite's advertised NPU computing power of 45 TOPS and the Apple M4 processor's advertised NPU computing power of 38 TOPS, there is actually no significant difference.
In terms of overall platform computing power, the Panther Lake platform boasts a maximum of 180 TOPS through the collaboration of the CPU, GPU, and NPU. This is significantly higher than the 75 TOPS of the Snapdragon X Elite and the overall computing power of the Apple M4 processor. However, whether actual applications can fully utilize this computing power will likely depend on software optimization.
In terms of connectivity, the Panther Lake platform's integrated Wi-Fi 7 and Bluetooth 6.0 features are essentially the same as those found in the Snapdragon X Elite processor. However, Apple has yet to integrate Wi-Fi 7 into its processors, giving the Panther Lake platform a clear advantage in this area.
However, regarding power consumption, while Intel emphasizes that the Panther Lake platform will be manufactured using the 18A process and touts its improved energy efficiency, compared to Qualcomm and Apple's long-term commitment to Arm-based processor design, whether it can actually deliver higher computing performance with the same power consumption will likely require further testing after the actual product launch.
The author's opinion
The Panther Lake platform design demonstrates Intel's commitment to leveraging its Lunar Lake platform experience. Not only does it continue to improve energy efficiency, but it also significantly enhances design flexibility, allowing OEMs to tailor specifications to meet diverse product needs. For example, the platform offers the option of pairing with a discrete graphics card or flexibly selecting memory specifications and PMICs. This is expected to enable product designs to cater to a wider range of markets.
Intel's emphasis on having a Windows PPM package ready for the Panther Lake platform before its launch suggests it's trying to avoid a repeat of the Arrow Lake platform's initial performance disappointment. It's also trying to convince the market that its x86 architecture can still compete with Arm-based processors in terms of energy efficiency and performance.
However, the actual performance of the Panther Lake platform will still depend on the test results after the actual products are released. However, the fact that Intel has announced its design details in advance and emphasized its flexibility and performance shows that Intel is very confident in this platform. At the same time, it also plans to use it to compete with Arm architecture processors in the thin and light notebook market.
