Intel Unveils “Jaguar Shores”: Next-Gen AI Chips for Inference at SC24

Intel has kept things pretty mum about its upcoming project, Jaguar Shores, but it’s a compelling development that promises to blend Gaudi ASICs with Xe-HPC GPUs in an intriguing way.

## Intel’s New AI GPU, Jaguar Shores, Set to Revolutionize AI Inference with Cutting-Edge 18A Node

While Intel might not look like the industry leader it once was, it’s still pushing forward with plans for both client and server platforms. Recently, Intel’s upcoming Battlemage GPUs were leaked, but it turns out there’s more in their pipeline than just Battlemage.

According to HPC Wire, Intel is pushing ahead with their next-generation AI chip strategies. Historically, their high-performance solutions emerged from the Gaudi platform by Intel’s Habana Labs. While Gaudi 3 is expected to hit the market in considerable volumes next year, there’s even more on the horizon with Jaguar Shores. This cutting-edge chip promises to cater to high computing demands, marking a significant leap forward for Intel’s product line.

This revelation came during the SC2024 conference, where Intel rolled out its plans for AI-intensive workloads focusing on top-tier performance chips. Jaguar Shores is destined to succeed Falcon Shores, the latter set for a 2025 debut. Much like Falcon Shores, it is poised to be a GPU rather than a conventional processor designed for Intel’s Gaudi Accelerator.

Initially, Falcon Shores was imagined to be a hybrid GPU-CPU solution. However, it has now transitioned into a dedicated GPU specifically for AI inference. Jaguar Shores seems to follow this path but with a twist that might incorporate a GPU-ASIC design, significantly enhancing AI Inferencing capabilities. Intel is putting a lot of stock in the 18A manufacturing node, suggesting that both Falcon and Jaguar Shores will make the most of homegrown 18A tech, utilizing advancements like RibbonFET and backside power to boost efficiency.

Intel has articulated its commitment to complementing its x86 franchise with investments in AI, zeroing in on enterprise-level, cost-efficient inferencing. This underscores its ongoing strategy, despite some of its chips being outsourced to TSMC—such as Lunar and Arrow Lake—Intel is aiming to gain a significant edge using the 18A node, which is also set to feature in next year’s Clearwater Forest. Looking further ahead, Intel plans to deploy its in-house 14A process node by 2027.

Facing off against major players like NVIDIA and AMD, Intel’s Jaguar Shores will need to hold its own. These competitors have well-defined plans for high-performance GPUs stretching through 2027, with both giants commanding the AI training and inference sectors. Therefore, Intel has strategically focused on capturing the inference market with these new chips.

Intel regularly reassesses its roadmap to make sure it aligns with customer needs, though no fresh updates have been released. Still, they are committed to advancing superior enterprise AI through their CPU and accelerator/GPU offerings, as emphasized by an Intel spokesperson.

It’s no secret that Intel has encountered difficulties, culminating in the cancellation of the Rialto Bridge GPU last year. Flaws in planning were evident as Falcon Shores shifted from a CPU+GPU hybrid to a standalone GPU. Similarly, the Ponte Vecchio GPU got the axe after the Aurora supercomputer went operational.

To navigate these choppy waters, Intel must meticulously plan and implement its strategies, especially amid recent leadership shake-ups within the GPU realm and the company at large. While the client CPU market has seen a slowdown, Intel can’t afford a similar scenario in the AI arena.

[Source: HPCWire]

—

Share this story on:
– Facebook
– Twitter