Groq

Tin tức, mô hình và bản phát hành AI mới nhất từ Groq. ['Groq 3 LP30', 'Groq 3 LPU', 'Groq 3 LPX', 'llama-3.3-70b', 'LPU', 'RealScale']

Trang chính thức ↗

Phần cứng & Suy luận 🇷🇺 ⚡

NVIDIA Groq 3: SRAM, Disaggregation, and Determinism in the New AI Platform

After acquiring Groq for $20 billion, NVIDIA integrated LPU accelerators into the Vera Rubin platform, creating a heterogeneous architecture for inference. The new Groq 3 LPX rack accelerator, based on Groq 3 LP30 chips with 500 MB of SRAM and 150 TB/s memory bandwidth, ensures deterministic execution and low latency. Phase-based disaggregation of decoding (AFD) allows separating FFN computations from attention, distributing the workload between GPUs and LPUs.

NVIDIA

Groq

ServerNews27.07 · 15:06

Phần cứng & Suy luận 🇷🇺

NVIDIA Unveils Vera Rubin POD Cluster: Five Racks, Seven Chips, 1.2 Quadrillion Transistors

NVIDIA announced the Vera Rubin POD cluster, built on the third-generation MGX architecture. It includes five rack types: NVL72, LPX, Vera, STX, and SPX, integrating 1,152 Rubin GPUs, 36 Vera CPUs, and other components. The cluster delivers 60 exaflops performance and 10 PB/s interconnect speed.

NVIDIA

Groq

ServerNews27.07 · 15:05

Xem thêm ↓