Các tài liệu gốc của sự kiện này
4-
Chip
Groq LPU
The deterministic inference engine whose design NVIDIA licensed in 2025.
Groq's first-generation Language Processing Unit strips out caches and dynamic scheduling: the compiler knows where every byte is every cycle, using 230 MB of on-chip SRAM at 80 TB/s on GlobalFoundries 14 nm. In December 2025 Groq signed a non-exclusive licensing deal with NVIDIA; founder Jonathan Ross and president Sunny Madra joined NVIDIA, while Groq…
Mở hồ sơ → -
Công ty
Groq
Deterministic inference hardware from the engineer who started Google's TPU.
Founded by TPU originator Jonathan Ross, Groq serves open models at headline speed from racks of its SRAM-only LPUs and sells capacity through GroqCloud. A 2025 investment round valued it near $7B, with a multi-billion-dollar Saudi deployment agreement its largest disclosed commitment.
Mở hồ sơ → -
Công ty
NVIDIA
The company the AI build-out is denominated in: GPUs, systems, networking and CUDA.
NVIDIA turned a graphics-chip business into the platform underneath modern AI: Hopper, Blackwell and Rubin GPUs, NVLink rack-scale systems, InfiniBand and Spectrum-X networking, and the CUDA software layer. Data-center revenue was $89.0B in Q2 FY2027 (quarter ended July 2026), up 117% year over year, and Vera Rubin production shipments began in August 2026.…
Mở hồ sơ → -
Hệ thống
NVIDIA Vera Rubin NVL72
The third-generation NVLink rack: 72 Rubin GPUs and 36 Vera CPUs in the Oberon frame.
Vera Rubin NVL72 is NVIDIA's 2026 rack-scale system and the successor to GB300 NVL72. It keeps the Oberon rack layout of 72 GPU sockets and 36 CPU sockets, filled with Rubin GPUs, Vera CPUs, NVLink 6 switches, ConnectX-9 SuperNICs and BlueField-4 DPUs. NVIDIA's spec table rates the rack at 3,600 PFLOPS of NVFP4 inference (sparse) with 20.7 TB of HBM4. First…
Mở hồ sơ →
