跳到正文
TechPowerUp· AleksandarK·· 1 天前AI 評分58

Microsoft 將在 Azure 自研晶片採用 Intel EMIB 先進封裝技術

Microsoft to Use Intel's EMIB Advanced Packaging Technology

AI 導讀

Microsoft 宣佈,將在 Azure 自研晶片採用 Intel Foundry 的 EMIB 先進封裝技術,並使用具矽通孔(TSV)的 EMIB-T。下一代自研晶片將由 Intel Foundry 採用 EMIB 及其變體封裝;Microsoft 亦確認繼續使用 Intel Xeon,以及即將推出的「Diamond Rapids」Xeon 7 CPU 支援 Azure。

正文

Microsoft announced today that it will officially adopt Intel Foundry's Embedded Multi-die Interconnect Bridge (EMIB) advanced packaging technology for its custom silicon used in Azure. Specifically, Microsoft will use Intel's EMIB-T, which includes through-silicon vias (TSVs) to enable vertical power delivery across multiple stacked chips, supporting the concept of vertical power delivery for 3D silicon modules at the package level. As a reminder, Microsoft's fleet of custom silicon solutions includes valuable accelerators, such as the current Maia 200, which powers AI inferencing workloads like large-scale reasoning, and the Cobalt 200 CPUs, designed for agentic workloads. The next-generation versions of these silicon projects will be packaged at Intel Foundry, supported by EMIB and its variants like EMIB-T.

Additionally, Microsoft confirmed its commitment to using Intel's Xeon processors and the upcoming "Diamond Rapids" Xeon 7 CPUs to power Azure infrastructure. The company will collaborate with Intel to integrate Intel Xeons on Compute Express Link (CXL) to achieve greater memory efficiency and system-level control across its data centers. Microsoft will also utilize Intel QuickAssist (Intel QAT) and confidential computing through Intel Trust Domain Extensions (Intel TDX). The company has measured a 2.5x increase in performance and flexibility on some workloads, such as SQL backups, thanks to CXL and other optimizations.

Naga Chandrasekaran, EVP, Chief Technology and Operations Officer and General Manager of Intel FoundryMicrosoft and Intel share a clear view of where AI infrastructure is headed-increasingly complex systems designed through the co-optimization of silicon, packaging, compute memory and software. Through use of our advanced packaging capabilities with EMIB to scale future generations of Azure infrastructure, Intel and Microsoft can combine the flexibility of heterogeneous integration with the manufacturing scale required for the AI era—giving customers new levels of performance and economics when interacting with Azure.

來源:TechPowerUp · techpowerup.com