Models & Labs
NVIDIA Spectrum-6 Powers Gigascale AI Factories
NVIDIA Bloghigh confidence
Why it matters
- →Spectrum-6 doubles the capacity of previous Ethernet systems, crucial for AI scalability.
- →It enables seamless coordination across thousands of GPUs, enhancing AI model training.
- →The integration with NVIDIA Vera Rubin platform optimizes AI operations at gigascale.
NVIDIA has unveiled the Spectrum-6 Ethernet switch system, a major leap in AI networking technology, offering 102.4 terabits per second capacity. This system is part of the NVIDIA Vera Rubin platform and is designed to support the massive data exchange required in gigascale AI factories. Companies like CoreWeave and Microsoft are among the first to adopt this technology, which promises to enhance the performance and efficiency of AI model training and inference. Spectrum-6's introduction signifies a shift towards more integrated and high-performance AI infrastructure.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
More in Models & Labs
Models & Labsother
llama.cpp b12340 adds CUDA 13 and ROCm 10.0
This release quietly closes the hardware gap for local inference by adding default support for CUDA 13 and ROCm 10.0 across Linux and Windows. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get performance, while Apple Silicon builds now ship KleidiAI disabled by default, shifting the focus back to standard ARM optimizations. The inclusion of Snapdragon support for Linux and Android marks a significant step toward on-device AI beyond the x86/ARM desktop ecosystem. It is less about new features and more about ensuring llama.cpp remains the universal runtime for every available accelerator.
llama.cpp Releases
Models & Labscoding
llama.cpp b12342 adds few-row MMQ for speculative decoding
Speculative decoding just got a significant speed boost on NVIDIA GPUs. This release introduces a specialized kernel for 'few-row' matrix multiplications, targeting the small batch sizes typical of verification passes in speculative decoding workflows. By optimizing these specific operations for quantized formats like Q8_0 and IQ4 variants, llama.cpp reduces latency where it matters most during token generation. The change is CUDA-only for now but leverages hardware capabilities from Ampere onwards, making local inference faster without requiring new model architectures or complex configuration changes.
