Track Hyper | In addition to XPU: Lenovo's new AI PC sets a new standard for memory

Author: Zhou Yuan / Wall Street News

Lenovo Group is the most active in promoting the popularization of AI PCs. Of course, this is easy to understand, as Lenovo Group is the world's largest PC shipper.

Among the various AI PCs launched by Lenovo Group, the Lenovo ThinkPad P1 Gen 7 (personal mobile workstation) is quite special, or more appropriately called "relatively typical".

Although it does not yet have the "Copilot" dedicated key that Microsoft hopes for, the ThinkPad P1 Gen 7 has two main features at the hardware level. One of them is that the memory model carried by this model is LPDDR5x-7467MT/s LPCAMM2 replaceable memory, which is also the industry's first new generation memory standard "CAMM2" LPDDR5x that meets the DDR6 standard.

CAMM2 is expected to become the mainstream memory standard for AI PCs (especially notebooks) in the future. However, LPDDR5x-7467MT/s LPCAMM2 is not LPDDR6 specifically designed for the CAMM2 standard, but a module made with LPDDR5x. In terms of physical structure, it is already prepared for the future.

"This is a product that balances performance, power consumption, and maintenance convenience." An SK Hynix technical engineer told Wall Street News, "LPDDR6 that meets the CAMM2 standard is likely to be launched in 2026, while DDR6 will be later; DDR6 for AI servers will be even later."

Nevertheless, the Lenovo ThinkPad P1 Gen 7 remains the industry's first AI PC to adopt the new generation memory standard "CAMM2".

AI Heterogeneous Computing Chip Intel Core Ultra

What is an AI PC? There is no unified definition in the industry yet.

According to the definition of the "AI PC Industry (China) White Paper" jointly released by Lenovo and IDC on December 7, 2023, an AI PC is a hybrid that integrates AI large models, applications, and hardware devices.

The "White Paper" believes that AI PCs have five main features: an intelligent body embedded with personal large models and user natural interaction, personal knowledge base, local heterogeneous AI computing power (XPU), an open artificial intelligence application ecosystem, and personal data and privacy protection.

Among them, XPU is a heterogeneous computing structure composed of CPU, GPU, and NPU. Currently, upstream chip designers such as Intel's Core Ultra platform, AMD's Ryzen 8040 series platform, and Qualcomm's Snapdragon X Elite platform all use XPU architecture.

The "White Paper" states that processors using XPU architecture are mainly aimed at providing strong parallel computing capabilities for PCs through the coordination of heterogeneous computing power.

Heterogeneous hybrid computing uses computing units with different instruction sets and system architectures to form a local computing system, matching corresponding computing units to different AI workloads The leap in parallel computing power enables PCs to easily perform complex AI model inference tasks, demonstrating higher intelligence.

The Intel Core Ultra 9185H processor installed in the Lenovo ThinkPad P1 Gen 7 is the flagship model of the "Intel Core Ultra" platform. Core Ultra is an AI PC heterogeneous computing processor platform launched by Intel in September 2023, with a computing power 8 times that of the previous generation (Core Ultra 7155H), adapted to over 10 domestic AI large models, while reducing power consumption by 36%.

Core Ultra has revolutionized the processor architecture that has been in use for over 40 years, enabling the "XPU" (CPU+GPU+NPU) to support LLM with a scale of 20 billion parameters for on-device operation.

The Core Ultra 9185H processor uses Intel's 4th generation process technology (similar to TSMC's 7nm process) and also incorporates Foveros 3D packaging technology, making it the first PC processor in the industry to use a modular architecture.

In terms of structure, it has transitioned from the previous "P-Core" (performance core) and "E-Core" (efficiency core) to adding an "LP-E-Core" (low-power efficiency core): 6 performance cores, 8 efficiency cores, and 2 low-power cores, totaling 16 cores and 22 threads; with a maximum frequency of 5.1GHz, 24MB of L3 cache, and a base TDP (Thermal Design Power) of 45W.

Base TDP refers to the maximum heat dissipation of the processor chip under normal operation, commonly known as "power consumption."

As a heterogeneous AI computing processor, the Core Ultra 9185H is the first to integrate an NPU computing unit for AI acceleration. Currently, the NPU module in PC heterogeneous computing chips is mainly used for low-power AI tasks (such as noise reduction in meetings, image enhancement, background blur in video conferences), with the primary function being energy efficiency.

The NPU computing unit integrated in the Core Ultra 9185H is not sufficient to handle heavy AI inference applications. In this case, the NVIDIA RTX Ada workstation-specific GPU card installed in the ThinkPad P1 Gen 7 comes into play.

With the combined effect of Intel's AI chip and NVIDIA's dedicated AI acceleration card, according to technical data disclosed by Intel, AI PCs equipped with Intel Core Ultra processors can achieve a maximum 2.2x improvement in AI performance compared to traditional PCs, a 47% increase in productivity, and a remarkable 12x performance improvement in professional applications

Debut of Detachable CAMM2 LPDDR5x Memory

In addition to being equipped with Intel's AI chip and NVIDIA's dedicated AI acceleration card, the Lenovo ThinkPad P1 Gen 7 also features a detachable memory module that meets the future technology and application direction of AI PCs: the Micron LPDDR5x-7467MT/s LPCAMM2, which adopts the new CAMM2 memory standard, allows for interchangeable memory (Crucial LPDDR5x modules for LPCAMM2 memory).

Apart from technical features (for more information on the technical advantages of CAMM2 LPDDR5x, please refer to the related content "New Standard DDR6 Strikes: Boosting AI PC Performance"), the Micron detachable memory in the Lenovo ThinkPad P1 Gen 7 directly impacts C-end users' experience in two main aspects: reducing the overall weight of the device and enhancing overall performance. Compared to models using the mainstream SO-DIMM memory standard and traditional CPU and GPU, it can more easily handle AI inference tasks.

In comparison to the Lenovo ThinkPad P16v i Gen 2, both models have identical chips and graphics cards. The latter model uses traditional SO-DIMM standard LPDDR5 memory (with dual slots for replacement), with overall dimensions of 365 x 262 x 24.66mm and a starting weight of 2.22kg. This is 7.61mm thicker and 400g heavier than the Lenovo ThinkPad P1 Gen 7.

In terms of physical structure, the LPCAMM2 LPDDR5x can be securely fastened to the motherboard with 3 cross screws, allowing for upgrades. Unlike traditional LPDDR (5/5x) memory, which is soldered directly to the motherboard and cannot be upgraded in the future (e.g., adding or changing memory specifications).

Furthermore, each LPCAMM2 memory module defaults to dual-channel, reducing power consumption by 58% and saving 64% of space compared to standard SO-DIMM memory.

Due to the CAMM2 memory standard being a new technological product, the entire supply chain needs to be adjusted accordingly, resulting in higher overall costs. The Micron 64GB LPDDR5X-7500 used in the Lenovo ThinkPad P1 Gen 7 has a high cost price of 2300 RMB, leading to a higher retail price for the whole device: starting at $2619 in North America (approximately 18988 RMB); the Lenovo ThinkPad P16v i Gen 2 has a North American retail price of $2279 (approximately 16523 RMB), making it 2465 RMB cheaper than the ThinkPad P1 Gen 7.

In terms of other configurations, the only standout feature is a 5 million IR camera, with the rest being relatively standard. On the software side, they both come with standardized suites. As a personal workstation for AI PCs, in addition to hardware features that align with AI characteristics, there is also a software system that meets the 5 major features outlined in the "White Paper." Overall, in terms of hardware performance, the Lenovo ThinkPad P1 Gen 7 can be considered as a top-tier device currently, but it comes with a high price tag, so the shipment volume may not be very high