Meta aims to address its extensive AI requirements through a strategic collaboration with Broadcom, focusing on the joint development of next-generation AI silicon, referred to as “XPUs.”
Meta and Broadcom Join Forces to Create a Multi-Gigawatt AI Ecosystem with Custom AI Silicon “XPUs”
Last month, Meta introduced four specialized AI chips under its MTIA AI lineup, designed specifically to optimize various tasks associated with artificial intelligence. These XPUs, which integrate multiple intellectual properties, are tailored for different workloads that encompass Generative AI inference, general-purpose tasks, and training functions. Below are the specifications for each chip:
| Metric | PATH 300 | PATH 400 | PATH 450 | PATH 500 |
| Workload Focus | R&R Training | General | GenAI Inference | GenAI Inference |
| Module TDP | 800 W | 1200 W | 1400 W | 1700 W |
| HBM Bandwidth | 6.1 TB/s | 9.2 TB/s | 18.4 TB/s | 27.6 TB/s |
| HBM Capacity | 216 GB | 288 GB | 288 GB | 384-512 GB |
| MX4 Performance | — | 12 PFLOPs | 21 PFLOPs | 30 PFLOPs |
| FP8/MX8 Performance | 1.2 PFLOPs | 6 PFLOPs | 7 PFLOPs | 10 PFLOPs |
| BF16 Performance | 0.6 PFLOPs | 3 PFLOPs | 3.5 PFLOPs | 5 PFLOPs |
| Scale-up Domain Size | 16 | 72 | 72 | 72 |
| Scale-up Network (unidirectional bandwidth)* | 1TB/s | 1.2TB/s | 1.2TB/s | 1.2TB/s |
| Scale-out Network (unidirectional bandwidth)* | 200GB/s** | 100GB/s | 100GB/s | 100GB/s |
In a recent announcement, Meta revealed its commitment to advancing its custom MTIA chips in collaboration with Broadcom. This strategic partnership will encompass chip design, advanced packaging, and networking over several years, resulting in multiple custom XPU designs.
The initial phase of this ambitious partnership aims to establish an AI ecosystem with a capacity exceeding 1 Gigawatt. Meta is positioning itself to directly compete with existing market solutions, targeting a consistent rollout of multiple products annually to meet its burgeoning demands.
Press Release: Today, Meta has announced an expanded collaboration with Broadcom to co-develop multiple generations of its next-generation MTIA (Meta Training and Inference Accelerator) chips. This custom silicon will enhance AI capabilities across all of Meta’s applications and services.
Broadcom will collaborate on chip design, advanced packaging, and networking, establishing the fundamental computing structure required to deliver real-time AI interactions to consumers worldwide. The partnership leverages Broadcom’s XPU platform, specifically engineered for the creation of custom AI accelerators to optimize Meta’s AI infrastructure across forthcoming silicon generations. Additionally, their advanced Ethernet technologies will facilitate high-bandwidth networking for Meta’s rapidly growing AI computing clusters.
The agreement entails a commitment surpassing 1 Gigawatt as part of a structured multi-gigawatt rollout strategy. This solidifies a mutual roadmap aimed at co-designing and scaling the necessary hardware to facilitate the emergence of personal superintelligence for billions across the globe.
“We are pleased to expand our strategic collaboration with Meta as they pioneer the next frontier of artificial intelligence, ” said Hock Tan, President and CEO of Broadcom.“This initial MTIA deployment is just the beginning of a sustained, multigeneration roadmap to serve the trajectory of massive growth over the next few years that highlights Broadcom’s unmatched leadership in AI networking and the power of our foundational XPU custom accelerator platform.”
“Meta is partnering with Broadcom across chip design, packaging, and networking to build out the massive computing foundation we need to deliver personal superintelligence to billions of people, ” said Meta founder and CEO Mark Zuckerberg.“As we roll out more than 1 Gigawatt of our custom silicon to start and then multiple gigawatts over time, this partnership will give us greater performance and efficiency for everything we’re building.”
This expanded collaboration with Broadcom, coupled with the evolution of the MTIA, is vital to Meta’s overarching strategy for AI infrastructure, emphasizing the necessity for a tailored compute foundation that can support its long-term aspirations for AI at a global scale.
Related Articles:
Microsoft Introduces Xbox Mode for All Windows 11 PCs, Merging Console and Desktop Gaming Experiences
20:22
Samsung Ends LPDDR4 Memory Production Amid 50x Profit Surge, Predicts Increasing Memory Shortages by 2027
19:07
Ultimate Guide to Windrose PC Performance Analysis and Tuning for Optimal PC Experience
18:18
Leave a Reply