Mastodawn

TechRadar (@techradar)

Meta가 GenAI 추론을 효율적으로 처리하기 위해 MTIA 맞춤형 칩과 1700W 슈퍼칩을 개발하고 있으며, 주류 반도체 업체에 의존하지 않는 방향을 추진한다는 내용입니다.

https://x.com/techradar/status/2034738865760870740

TechRadar (@techradar) on X

Meta develops MTIA custom chips and a 1700W superchip to run GenAI inference efficiently without relying on mainstream silicon vendors. https://t.co/tFkeMMV4XC

X (formerly Twitter)

KINEWS24 Mar 12, 2025

Meta entwickelt eigenen KI-Chip MTIA

MTIA-Chip: Reduziert Abhängigkeit von Nvidia
Kostenreduktion: Günstigere KI-Infrastruktur
Effiziente KI: Bessere Leistung für Empfehlungen

#ai #ki #artificialintelligence #kuenstlicheintelligenz #meta #mtia #nvidia #chipentwicklung

Jetzt lesen und folgen!

https://kinews24.de/meta-mtia/

Meta MTIA der eigene Chip gegen Nvidia-Dominanz

Meta MTIA der eigene Chip gegen Nvidia-Dominanz: Meta testet eigenen KI-Chip, um Nvidia herauszufordern & Kosten zu senken. Alle Infos zum Meta MTIA Chip!

KINEWS24.de

eicker.news ᳇ tech news Apr 11, 2024

»#Meta’s new #AIchips run faster than before: The next-generation #MTIA chip could be expanded to train generative #AImodels.« https://www.theverge.com/2024/4/10/24125924/meta-mtia-ai-chips-algorithm-training?eicker.news #tech #media

Meta’s new AI chips run faster than before

Meta’s planned next-generation MTIA chip will train and optimize inference for its ranking and recommendation models — and maybe soon its generative AI models, too.

The Verge

Data Center Café Feb 5, 2024

Meta が今年後半に自社データセンターにAIチップを導入へ
https://cafe-dc.com/hpc/meta-to-deploy-its-own-ai-supported-chips-in-its-data-centers/
#cafedc #meta #mtia

Meta が今年後半に自社データセンターにAIチップを導入へ | Data Center Café

Metaは、今年中に自らのデータセンターにAIに特化したカスタムチップの最新版を導入する計画を立てています。ロイター通信が入手し報道した文書によると、同社は自社で設計し、2023年に開発開始が報じられた第2世代のチップを導入することで、Nvidiaへの依存度を下げたい考えのようです。

Data Center Cafe

Florian - RISC-V Ambassador May 19, 2023

Facebook introduced #MTIA which is their own AI accelerator which is manufactured in TSMC 7nm and clocked at 800MHz and has 64 PE (Processing Elements) inside.

Each PE include 2 Processors, one based on the Vector extension and both based on RISC-V.

It is optimized for pyTorch and comes with a full SW stack and with less than 25W per chip, having 2 chips on a PCIe board.

I'm looking forward to get more detailed information!

https://ai.facebook.com/blog/meta-training-inference-accelerator-AI-MTIA

#riscv #semiconductor #AI