Intel’s absence of a desktop CPU with AI capabilities — until later this year — has prompted PC manufacturers to consider chip startups as alternatives. The Lenovo ThinkCentre Neo Ultra, expected to hit the market in June for around $1,000, may be equipped with AI cards from MemryX and Kinara, according to Lenovo product manager Bryan Lin at CES 2024.
AMD, Intel, and Qualcomm have all demonstrated mobile processors with integrated AI NPUs. AMD is the only one to have announced a desktop Ryzen processor with an APU inside. Intel, the dominant player in PC processors, will have to hold back until the launch of Arrow Lake before they can make an NPU available for desktop PC makers. As a result, more PC makers are acknowledging that a “AI PC” can be developed using just a CPU and a GPU, though NPUs offer more power-efficient AI.
“What we’re seeing now is that the discrete graphics card is too hungry in terms of form factor and power, thermal design, et cetera,” Lin said. “So an NPU card drawing about 5 to 10 watts can give us a certain level of AI capabilities.”
The ThinkCentre Neo Ultra will incorporate up to an Intel Core i9 vPro processor and a creator-class Nvidia GeForce RTX 4060 GPU. In addition to this, it will have up to 64GB of DDR5-5200 memory, up to 4TB of SSD storage, and a 350W internal power supply. So as to optimize the workload, Lenovo has included an AI engine.
Meet MemryX, one of the first AI accelerators
MemryX produces the MX3 Edge AI Accelerator, which integrates four MX3 chips mounted on an M.2 PCI Express card (Gen3). Each MX3 chip can deliver 10 TFLOPs. According to Roger Peene, MemryX’s vice president of product and business development, this solution serves as an attractive alternative to discrete solutions until Intel or AMD integrates it. Peene noted that Intel is behind and is not content that Lenovo would opt for a startup to run AI in a PC.
As for Kinara, the company launched its Ara-2 Edge AI processor last fall. They claim that Ara-2 can process Generative AI models at a speed of 10 seconds per image for Stable Diffusion and produce tens of tokens/sec for LLaMA-7B, but we have not been able to confirm this yet.
Both MemryX and Kinara AI chips are initially targeting AI for image recognition. MemryX demonstrated its ability to recognize whether construction workers were wearing the right protective gear. There are, however, many other potential uses for AI, including games, avatars, local language models/chatbots, and more. Moreover, the emergence of AI accelerator cards may challenge established players like Nvidia and others in the content-creation and gaming industry.