With so many AI CPUs on the market, which one should you buy?
AI CPUs are specialized computer processors that integrate neural processing units (NPUs). AI processors are designed to help you perform AI tasks on local devices. They are increasingly appearing in more devices and are essential for running AI assistants such as Copilot and Apple Intelligence.
So, with so many AI CPUs on the market, which one should you buy?
AI CPU Comparison
Intel, AMD, Apple, and Qualcomm have announced the adoption of new SoC (System on Chip) designs for their latest mobile processors. These new processors combine the CPU, GPU, and NPU into a single chip to provide efficient AI computing capabilities. Although some of these new SoCs are still awaiting release in 2024, official announcements, design specifications, as well as self-reported and independent benchmark tests can help us determine whether these upcoming processors are worth the wait, or if you should purchase an AI laptop immediately.
To help you decide which AI processor to buy, here are the latest developments in AI processors from Intel, AMD, Apple, and Qualcomm.Intel Core Super 200V (Lunar Lake)
Intel unveiled its new Lunar Lake processors at the 2024 Computex in Taiwan, China. This new mobile processor features several improvements over the previous generation, focusing mainly on heat dissipation, energy efficiency, better GPU, and AI computing capabilities, while still using the x86 architecture. Notable SoC design features include:
Unified Memory Architecture: The Intel Lunar Lake processor now integrates LPDDR5 RAM into its SoC design. This allows for higher bandwidth and lower power consumption when transferring data between RAM and the processor.
3nm Process: With the 3nm process, Intel has packed more transistors into Lunar Lake, enhancing its performance and power efficiency.
Integrated NPU: The Lunar Lake SoC features six NPU computing engines, delivering up to 40 TOPS (trillion operations per second) of AI computing power with INT8 precision.
Hyper-Threading Disabled: Hyper-threading is disabled on all eight cores (four performance cores and four efficiency cores) to improve battery life rather than performance.
With this new SoC design, Intel's Lunar Lake processors are expected to offer three times the AI performance, 1.5 times faster graphics processing, and approximately 40% better energy efficiency compared to the previous generation Meteor Lake processors.
AMD Ryzen AI 300 (Strix Point)
In contrast to Intel's energy-saving approach to handling x86, AMD places a greater emphasis on performance, albeit at the cost of higher power consumption. Here are some features that make these processors powerful:Zen 5 Microarchitecture: Delivers significant improvements in IPC (Instructions Per Clock) and overall performance.
Integrated RDNA 3.5 Graphics: Enhances the previous RDNA architecture with notable performance boosts in graphics and AI-related tasks.
XDNA2 NPU: The highest-performing NPU on the SoC. Capable of up to 50 TOPS at INT8 precision, suitable for Copilot+ requiring 40 TOPS.
Block FP16: Achieves higher precision for AI tasks with minimal impact on performance.
This makes AMD's Ryzen AI 300 series processors a powerful choice for demanding AI and computational tasks, leveraging advanced graphics and AI processing capabilities.
Apple M4
The Apple M4 uses similar technologies to the M3, such as a 3nm process node, chip-integrated memory, chiplet design, and a hybrid architecture. The M4 has been integrated into the latest iPad Pro, offering 9 or 10 CPU cores (3 or 4 performance cores and 6 efficiency cores), a 16-core NPU capable of reaching 35 TOPS, and a 10-core GPU that is four times faster than the M2. The design changes are not as drastic as Intel's Lunar Lake, mainly because the M-series chips have been well-optimized by this point, and ARM devices are more energy-efficient than x86 devices.
Qualcomm Snapdragon X Elite
Qualcomm is now producing powerful ARM processors for Windows machines! The Snapdragon X Elite processor operates on RISC (Reduced Instruction Set Computing) instead of CISC (Complex Instruction Set Computing), which is commonly found on most Windows computers. Qualcomm states that the X Elite SoC features a 12-core ARM v8 Oryon CPU, Adreno X1 GPU, and Hexagon NPU, capable of reaching 45 TOPS at INT8 precision, making it a powerful processor for Windows Copilot Plus. The combination of RISC and a powerful SoC makes Qualcomm's Snapdragon X Elite a strong competitor to Apple's M-series chips, which are also high-performance RISC processors.Intel, AMD, Apple, and Qualcomm: AI Processor Comparison
Below is a comparison table for Intel Lunar Lake, AMD Ryzen AI 300, Apple M4, and Qualcomm Snapdragon X Elite:
According to the table above, we have two x86 (Lunar Lake and Ryzen AI 300) and two ARM (M4 and Snapdragon X Elite) AI processors. It is well known that ARM processors have better energy efficiency, while x86 offers higher performance. However, as the M4 and X Elite become more powerful and Lunar Lake and Ryzen AI 300 become more energy-efficient, the gap between performance and energy efficiency seems to be narrowing.
In terms of energy efficiency for x86 processors, Intel performs better with its 3nm process node, on-chip memory, disabled hyper-threading, and a lower number of CPU cores. Meanwhile, AMD's Ryzen AI SoC offers better performance with 24 threads (slightly higher CPU clock speed), a more powerful GPU, and an NPU with block FP16 functionality.
As for ARM AI processors, Apple's M4 outperforms the X Elite in terms of thermals, CPU, and even GPU due to its hardware-accelerated tracking features and native support for macOS applications. However, it is worth noting that despite emulation and other software issues, the X Elite chip remains a powerful ARM-based processor that can compete with Apple's M3, Intel's Meteor Lake, and AMD's Ryzen 7000 processors.
Which AI CPU should you buy?
Laptop manufacturers typically offer options with different hardware specifications, including processors. So, with the new AI-supported SoCs hitting the market this year, which AI CPU should you buy?Apple M4 (Donan): Best for macOS users. Designed and optimized for macOS, offering competitive performance and longer battery life.
AMD Ryzen AI 300 (Strix Point): Ideal for gamers. Its high-performance multi-threaded CPU paired with a powerful integrated GPU makes it the ideal choice for gaming and other intensive tasks.
Intel Core Ultrabook 200V (Lunar Lake): Balanced performance. It achieves a good balance between performance and battery efficiency. Suitable for gaming (especially e-sports), productivity tasks, media consumption, and general web browsing.
Qualcomm Snapdragon X Elite: The most power-efficient Windows AI processor currently available. It is the first processor to natively support Windows Co-Pilot Plus. Great for general work, web browsing, and media consumption.
Although all these processors have AI capabilities through their integrated NPU, it may take some time before we can fully benefit from them. Developers need more time to create software that fully utilizes the NPU.
While it may be tempting to buy a new laptop now, the AI features on these new SoCs are significantly better than those released in 2023. Therefore, if AI features are important to you, either purchase the Snapdragon X Elite PC now or wait for the upcoming M4, Core Ultra 200V, or Ryzen AI laptops to be released later this year.
Comment