#United States #San Jose #Together AI #NVIDIA Blackwell #GPU Clusters

Transforming AI Infrastructure: Together AI at NVIDIA GTC 2025

Together AI has made notable strides in AI cloud computing, hosting a major unveiling at NVIDIA GTC 2025. As the leading provider in AI acceleration, the company introduced the highly anticipated Instant GPU Clusters, in addition to ramping up deployments of the state-of-the-art NVIDIA Blackwell GPUs. This merger of technology promises significant advancements for AI professionals and enterprises aiming to streamline their AI workloads.

Instant GPU Clusters: Flexibility and Speed

At the heart of this launch is the Together Instant GPU Clusters. These clusters come with up to 64 NVIDIA Hopper GPUs, designed for maximum efficiency and performance. They are interconnected through NVIDIA’s Quantum-2 InfiniBand and NVIDIA NVLink, allowing users to experience ultra-low latency and high-bandwidth — essential features for AI teams requiring quick access to high-performance computational resources.

AI teams can now enjoy a fully self-service experience by creating their clusters within minutes through the Together AI console. This innovation accelerates AI research and experimentation, reducing time wasted in lengthy procurement processes. Whether it's peak computing demands, validation of models before significant investments, or extensive training and inference tasks, these clusters fulfill various needs effectively.

Enhanced Performance with NVIDIA Blackwell GPUs

The latest generation of NVIDIA Blackwell GPUs marks a significant leap in AI performance. Built on a cutting-edge 5nm process technology, these GPUs incorporate advanced technologies like FP8 Tensor Cores, enabling up to 35 times the performance compared to earlier models. This enhancement means that AI training runs will be up to 90% faster than those using previous architectures. Such efficiency is pivotal for machine learning applications, particularly for large-scale models requiring substantial computational power.

As highlighted by the unveiling, the Together GPU Clusters now demonstrate a remarkable capacity to manage 15,200 tokens/second/node during training for a 70 billion parameter large language model. This kind of performance is a game-changer for developers focusing on complex tasks.

GPU Clusters Tailored for All Stages of AI Development

Together AI offers a spectrum of GPU clusters to cater to diverse enterprise needs:

• - Instant GPU Clusters with up to 64 GPUs for quick and efficient deployments.
• - Dedicated GPU Clusters, which can house between 64 to 1,000 NVIDIA GPUs, are designed for deep and extensive training and inference tasks.
• - Custom GPU Clusters are tailored for hyperscale projects, accommodating upwards of 1,000 GPUs for expansive AI supercomputing tasks.

Seamless Integration and Future Prospects

As a recognized NVIDIA Cloud Partner, Together AI facilitates the deployment of NVIDIA’s NIM microservices, optimizing the infrastructure for running AI applications. This integration allows for ease in adapting AI applications, scaling infrastructure according to real-time demands, and leveraging NVIDIA AI Enterprise’s capabilities for enhanced operational efficiency.

Together AI illustrates a commitment to pushing the boundaries of AI infrastructure, ensuring enterprises can meet the growing demands of modern AI research and applications. With significant players like Cartesia, Salesforce, and Captions already leveraging Together AI’s bespoke infrastructure, the field is ripe for innovative breakthroughs.

For access to the newly introduced Instant GPU Clusters, interested parties can visit together.ai/instant. This leap forward beckons a promising future for generative AI, reinforcing stability, speed, and performance in computational resources.

Conclusion

As the AI landscape evolves, innovations like those showcased by Together AI at GTC 2025 ensure that powerful and efficient computational resources remain at the forefront. The collaborative focus on transparency and flexibility allows diverse sectors to embrace and adapt AI solutions more effectively than ever before. Together AI's advancements reinforce a future where AI models can be trained and deployed with unprecedented speed, marking a new chapter for AI development and application.