SambaNova Systems DataScale
Future-proof your data center for next-generation AI workloads and beyond
SambaNova Systems DataScale™ lets you break free from the constraints of today’s legacy technologies by providing you with the core infrastructure to run cutting-edge AI applications at scale from the data center to the cloud—and to the edge.
Architected using SambaNova Systems Reconfigurable Dataflow Architecture (RDA) and built using open standards and user interfaces, SambaNova DataScale is an integrated software and hardware systems platform optimized from algorithms to silicon. Our software-defined-hardware approach delivers unmatched efficiency across applications from training, inference, data analytics, high performance computing (HPC) and more—and all can run on SambaNova DataScale.
- Delivering highest performance and efficiency for training and inference
- Raising the standard for state-of-the-art accuracy
- Extensibility to run models of all sizes at multi-rack scale
EASE of USE
- CUDA-free computing without lock-in
Efficiency at Any Scale
Built with a highly flexible modular architecture, DataScale can scale seamlessly from 1 to 100s of systems to meet the relentless demands of modern AI computing—with no code change necessary.
With SambaNova RDU-Direct ™ technology, multiple DataScale systems can be connected at scale through a high throughput, low latency fabric using standard interfaces. Whether it’s state-of-the-art models today, or the bigger, more sophisticated models of tomorrow—the DataScale platform will help accelerate your discovery of the next breakthrough.
At SambaNova, we took a software-first approach and fully integrated SambaFlow™, our software stack. SambaFlow is designed to take your existing or new models and automatically determine the optimal way to take full advantage of SambaNova Systems Reconfigurable Dataflow Unit™ (RDU), with little to no change to your algorithms.
There is no need to rewrite your proven model for SambaNova DataScale or even understand our architecture, just recompile using our tools and experience high performance out of the box.
SambaFlow eliminates months of learning, tuning and optimizing to allow you to focus on what matters the most: the application.
Break free from the constraints of today’s conventional solutions and advance your AI innovations with SambaNova DataScale System.
Reconfigurable Dataflow Unit
SambaNova Systems Reconfigurable Dataflow Unit (RDU) is the industry’s next-generation processor and is at the core of SambaNova DataScale. RDUs are designed to allow the data to flow through the processor in ways in which the model was intended to run, freely and without any bottlenecks.
RDUs eliminate constant data caching and excess data movement inherent to today’s core-based architectures. This unlocks significant silicon utilization to unleash more compute than any other solution available today.
The core infrastructure to develop and deploy next-generation AI algorithms and applications at scale with unmatched performance, state-of-the-art accuracy, scale and ease of use.
Take a 3D Tour of DataScale
Peek inside our technology with this interactive tool.
Hear What Our Customers Say
Lawrence Livermore National Laboratory
“The support for running multiple models built into the hardware was a significant aspect of our selection of SambaNova Systems. … We’re seeing roughly a 5X improvement compared toa comparable GPU running the same models.”
Bronis de Supinski, Chief Technology Officer
Argonne National Laboratory
“We’re interested in technology … that can train models faster than individual GPUs, that has more scalability and hasmore performance, and so the SambaNova DataScale system met that criteria.”
Rick Stevens, Associate Laboratory Director
Hugging Face Demo
Hugging Face is the most popular model repository for developing transformer-based deep learning apps. Integrating Hugging Face with DataScale you can run thousands of Transformer models with state-of-the-art accuracy in seconds and with zero code changes.
Unlock new capabilities with SambaNova DataScale
- Trillion parameter NLP models
- 50k x 50k high resolution deep learning
- Recommendation models with huge 100 GB embedding tables
- Exascale data processing
Get Started with SambaNova Systems
Accelerate workloads deployed in your own data center with a subscription service designed to jump-start your AI initiatives.
SambaNova AI Cloud Platform
A new AI systems platform in the cloud for university researchers. Compelling research proposals are being accepted now.