Qwen 3.6, utilizing a decentralized GPU infrastructure, has successfully generated over 22.8 billion tokens on the Dolphin Network, powered by 383 GPUs. This network leverages a diverse array of graphics units, including NVIDIA’s Blackwell-generation GPUs, which have recently gained traction as advantageous for scaling high-end inference tasks in AI projects. The current setup offers a total inference bandwidth of 9,400 tokens per second, showcasing the network’s capacity for synthetic data generation—a crucial component for enhancing large language models.

NVIDIA: NVIDIA is a leading designer of graphics processing units and data‑center accelerators widely used for AI training and inference, including its RTX, A‑series, L‑series, H‑series, and Blackwell‑generation products. In this context, Dolphin Network’s hardware inventory and performance metrics are built almost entirely on NVIDIA GPUs, underscoring how decentralized AI infrastructure layers are being constructed on top of NVIDIA’s ecosystem of workstation and server‑class cards.
dphnAI: dphnAI is the team and account representing the Dolphin Network project, focused on building a distributed GPU network and APIs for AI inference and data generation. In this update, dphnAI reports strong progress in rolling out node providers for Qwen 3.6 35B inference, detailing the mix of GPUs in the network and announcing that public API access for developers is coming soon.
Dolphin Network: Dolphin Network is a decentralized GPU infrastructure and inference marketplace that aggregates idle capacity from a wide range of consumer and data‑center GPUs to run large language models and other AI workloads. In this news, Dolphin Network is hosting Qwen 3.6 35B for large‑scale synthetic data generation, showcasing hundreds of GPUs online and highlighting how its node pool can repurpose GPU memory comparable to large clusters of high‑end accelerators for high‑throughput inference.

NVIDIA_Blackwell_Adoption: Industry coverage over the past month has noted early adoption of NVIDIA’s Blackwell‑generation workstation and server GPUs by AI infrastructure platforms, which present these cards as a key upgrade path for scaling high‑end inference workloads on emerging networks.
Synthetic_Data_Generation: AI infrastructure projects have increasingly highlighted synthetic data generation as a primary workload for large language models, framing it as a way to continuously expand training corpora and fine‑tuning datasets without relying solely on web‑scraped human data.
Decentralized_GPU_Infrastructure: In recent weeks, several decentralized GPU networks have emphasized using heterogeneous consumer and data‑center GPUs to deliver competitive inference performance for large models, positioning themselves as an alternative to centralized cloud AI providers.