Microsoft Shows Love for Intel and AMD But Chooses Qualcomm For Copilot+ PCs

To deliver the high demand for HPC, Microsoft announced its intention to provide its cloud computing clients with AI chips developed by AMD.

Additionally, Microsoft will unveil a preview of its new Cobalt 100 custom processors at the upcoming conference. These chips, first announced in November, reportedly deliver a 40% performance improvement over other ARM-designed competitors. Companies such as Adobe and Snowflake are already utilising these processors.

During a keynote speech, Microsoft introduced the Copilot+ PC, their brand of AI PCs. This sets specific hardware requirements for systems to qualify as ‘AI-first’ PCs. These PCs will have the capability to manage certain AI-accelerated tasks, such as chatbots and image generation locally, rather than depending on cloud services.

However, new hardware will be necessary to execute these tasks swiftly and efficiently. For this, Microsoft has announced a partnership with Intel, AMD, and Qualcomm to build neural processing units (NPUs), allowing these AI models to run locally on PCs.

Minimum requirements built for Qualcomm

At a minimum, systems will need 16GB of RAM and 256GB of SSD to meet the memory and on-disk storage requirements for LLMs such as Microsoft’s Phi-3. Microsoft has announced that all PCs powered by the Snapdragon X Plus and X Elite chips will have Copilot+ features pre-installed. According to Microsoft, they will start shipping on June 18th.

During the keynote speech, Microsoft said that the Copilot+ PC runs 58% faster than the MacBook Air M3 in sustained multi-threaded performance on Snapdragon X Elite.

All this sounds fine for now for Intel and AMD.

However, the most significant new requirement, which will pose a challenge for almost all existing Windows PCs, is the need for an integrated NPU. Microsoft mandates an NPU with a performance rating of 40 trillion operations per second (TOPS), a benchmark used by Microsoft, Qualcomm, Apple, and others in comparing NPU capabilities. 

The challenge for Intel and AMD is that none of their current-generation chips meet this requirement, even those equipped with NPUs. Intel's Meteor Lake-based Core Ultra NPUs peak at 10 TOPS, while some of AMD's Ryzen 7000 and Ryzen 8000 desktop and laptop processors with NPUs range between 12 and 16 TOPS.

Currently, this requirement is met by only one chip in the Windows PC ecosystem: Qualcomm's Snapdragon X Elite and X Plus. These chips, set to launch in the new Surface PCs and offerings from Dell, Lenovo, HP, Asus, Acer, and other major OEMs in the coming months, feature NPUs capable of 45 TOPS, slightly exceeding Microsoft's minimum requirement.

In comparison, Apple’s latest M4 processor is capable of 38 TOPS of NPU performance. 

While both Intel and AMD have future products in their roadmap aimed at enhancing NPU performance to meet Microsoft's Copilot+ PC requirements, none of the existing hundreds of millions of x86-based Windows PCs, including current models available for purchase, will qualify for the Copilot+ label.

Intel and AMD's CPUs and GPUs can add to the total TOPS a system can achieve. However, NPUs are specialised hardware designed to execute AI workloads more efficiently and with lower power consumption.

Intel and AMD Catching up

As mentioned before, both Intel and AMD have plans in the pipeline to meet these requirements. Intel has revealed that starting from the third quarter of 2024, its highly anticipated client processors, codenamed Lunar Lake, are slated to power over 80 fresh laptop designs across more than 20 OEMs.

Lunar Lake is boasting over three times the AI performance of its predecessors. With an impressive 40+ NPU TOPS, Intel’s next-gen processors are poised to deliver the capabilities required for the upcoming Copilot+ experiences. Moreover, Lunar Lake will feature over 60 GPU TOPS, amounting to more than 100 platform TOPS in total.

Similarly, AMD is also coming up with its new APU, the Ryzen 8050, featuring a Zen 5 CPU and an XDNA 2 NPU architecture for AI PC workloads. This will boast a performance of around 50 TOPS, more than ideal for running Microsoft’s AI PC goals. These have been dubbed the Strix Point APUs. AMD is expected to launch these by the second half of 2024.

Once released, the two will try to outcompete each other as to who can offer better AI PC goals. Meanwhile, Qualcomm, with its Snapdragon X Plus and X Elite chips, will start shipping by the end of next month.

This is all while NVIDIA criticises Microsoft’s AI PC conversations. The company said that its GPUs provide better performance than any NPUs in the market, calling all other NPUs only good enough for ‘basic’ AI tasks. It also claims that its RTX GPUs can achieve between 100 to 1,300 TOPS of power, depending on the GPU.