Micron has also unveiled a fresh modular memory system, LPCAMM2, that is directly targeting the new generation of AI-enabled laptops. The module is an enhancement to the capabilities of the laptop memory, by integrating the benefits of the LPDDR5X with a small, upgradeable form factor - and the implications of future AI laptops may be huge.
New thing: velocity, shape, versatility.
The LPCAMM2 modules of Crucial brand in Micron operate at a speed of 8,533 MT/s and have densities reaching 64 GB. The modules functionally match LPDDR5X in terms of low power consumption, as it is claimed to consume considerably less active and standby power than a typical DDR5 SODIMM, but less than half the size of typical size laptop memory modules. The reduced size also provides designers with a wider range of options in the form of thinner devices or larger batteries. In addition to this, LPCAMM2 is modular, unlike most LPDDR modules which are soldered in to the motherboard, which is to say memory upgrades or replacements become possible.
When Micron launched LPCAMM2 in early 2024, it said it had up to 61 percent lower power consumption and up to 71 percent higher performance in PCMark tests than DDR5 SODIMMs, and was 64 percent smaller. In more current sources, Micron indicates that the new modules at present can perform 1.5 times as well as normal DDR5 5,600 MT/s.
Effects on the future AI laptops and benchmarks.
Inference, local model execution, data preprocessing, and so on AI workloads tend to push the memory subsystem. LPCAMM2 should also be able to support AI laptops with more heavy models (e.g. local LLMs, vision tasks), because of its increased bandwidth and reduced energy consumption.
Since the module is comparatively smaller and uses less power, OEMs can use the internal space to fit bigger batteries or add other parts or just demand even narrower chassis designs without compromising performance.
Memory is also soldered, which is one of the largest weaknesses of many AI laptops. The modular design of LPCAMM2 implied that users would be able to upgrade memory in the future when models require more memory powerfully, which would enhance the product life cycle and decrease e-waste.
Generative AI benchmarks (e.g. inference latency), memory-sensitive benchmarks (e.g. ML inference suites), productivity benchmarks (e.g. PCMark, content creation tasks) may mostly clearly show an increase when machines are using LPCAMM2 over traditional SODIMM-based laptops. In Micron early comparisons give examples of 7 percent in digital content creation and 15 percent in productivity workloads.
Naturally, it will be adopted with OEM support. Up to this point, the adoption can be seen in some AI workstations of Lenovo and Dell. However in the case of wider consumer-based AI laptops, compatibility of the motherboard and chipsets (i.e. CAMM2 standard) will be important. The new standard is compatible with the CAMM2 specification of JEDEC that can assist in leaping forward by other memory vendors and OEMs.
Altogether, LPCAMM2 may be a breakthrough: it is the potential to take high-bandwidth, low-power, modular memory to AI laptops. To the users it equates to more competent and durable machines. To the industry, it is a step towards a new level beyond the soldered memory. It will not take long before benchmarks and real world AI tasks react once again when devices start shipping with LPCAMM2.
إرسال تعليق