Bits With Brains
Curated AI News for Decision-Makers
What Every Senior Decision-Maker Needs to Understand About AI and its Impact
Intel's AI Chip Onslaught: Resistance is Futile?
12/24/23
Editorial team at Bits with Brains
Intel has unveiled a long-awaited series of powerful new chips aimed at capturing more market share across key high-growth segments - data centers, PCs and edge devices - by delivering specialized AI acceleration, according to CEO Pat Gelsinger.
Intel has unveiled a long-awaited series of powerful new chips aimed at capturing more market share across key high-growth segments - data centers, PCs and edge devices - by delivering specialized AI acceleration, according to CEO Pat Gelsinger.
The star announcement was Gaudi3, Intel's next-generation AI accelerator for deep learning and large language models. Launching in 2024, Gaudi3 will offer quadruple the AI throughput versus its predecessor Gaudi2, which Intel says is already beating Nvidia's flagship H100 chip on some workloads.
With AI model sizes exploding, Intel is betting Gaudi3's combination of high performance and competitive pricing can continue taking share from Nvidia, which currently dominates the training accelerator market. Gaudi3 will pack more on-the-chip memory, bandwidth, and networking capability to handle the most demanding AI workloads.
It also unveiled Core Ultra, the first client processor with a dedicated AI accelerator built into every chip. This Neural Processing Unit (NPU) specialized for AI tasks will enable a new generation of AI applications on laptops and desktops.
Over 100 independent software vendors are already working on hundreds of apps leveraging the on-chip NPU in Core Ultra. By offloading AI processing onto the NPU, these apps can run complex neural networks much more efficiently than relying on the CPU and GPU alone.
Intel predicts that thanks to this innovation, AI-enhanced PCs will reach 80% of the laptop market by 2028. The company is aggressively courting developers to build this new AI app ecosystem and lock in customer demand for Intel's chips.
Meanwhile in the data center, Intel took the wraps off its 5th Gen Xeon server CPUs, now with AI acceleration baked into every processor core. They claim up to 42% higher inferencing throughput on large language models compared to CPUs without embedded AI acceleration.
Intel is delivering specialized on-chip AI to data centers for the first time, allowing customers to run AI workloads without adding discrete accelerators like GPUs. This should improve performance and reduce costs for many leading-edge workloads involving recommendations, search, logistics and more.
With these chips spanning massive data centers down to the PC on your desk, Intel is executing an "AI Everywhere" strategy chasing the booming demand for AI computing power across the entire technology stack.
And Intel is already claiming some big design wins. Major partners Microsoft, Oracle and Alibaba announced plans to use the latest Xeon chips, while PC makers like Dell, HP and Lenovo previewed upcoming Core Ultra laptops.
With its rivals also gearing up new AI offerings, Intel is moving aggressively to compete for its share of the accelerating AI chip market. Analysts will be watching closely to see if Intel's combination of specialized on-chip AI and software ecosystem can help them carve out a sustainable position.
Sources:
Sources