Tenstorrent Introduces Wormhole Instances on Koyeb Cloud
In a significant advancement in AI technology, Tenstorrent has launched its Wormhole™ Instances on the Koyeb Serverless Platform. This announcement, made from Santa Clara, California on February 25, 2025, marks a pivotal moment for developers looking to enhance their AI model inference performance.
On-Demand Access to Cutting-Edge Technology
Tenstorrent aims to provide developers quick access to their innovative hardware. The Wormhole instances are designed to streamline the development process, allowing users to harness the power of the Tensix Processor and the TT-Metalium SDK quickly. With the Koyeb platform's serverless architecture, users can now bring up and test their models in just a matter of minutes, eliminating the traditional complexities associated with infrastructure management.
Instance Details
The Wormhole Instances come in two configurations that cater to different performance needs:
- - TT-n300s Instance: This entry-level option features one n300s chip with 24GB of GDDR6, 192MB of SRAM, up to 466 FP8 TFLOPS performance, 64GB of RAM, and 4 virtual CPUs.
- - TT-Loudbox Instance: For demanding applications, this upgraded configuration utilizes four n300s chips meshed together, offering a robust 96GB of GDDR6, 768MB of SRAM, up to 1864 FP8 TFLOPS performance, along with 256GB of RAM and 16 virtual CPUs.
Both instances integrate seamlessly with Koyeb's infrastructure, boasting rapid data storage with NVMe volumes, efficient container deployment, and autoscaling capabilities to manage fluctuating workloads.
Developer-Centric Features
One of the standout features of the Koyeb platform is its focus on the developer experience. Key functionalities include:
- - Instant Development Environments: Developers can deploy containers executing Tenstorrent SDKs within a minute, significantly reducing setup time.
- - Continuous Deployment: Automatic building of Dockerfiles ensures that teams can push updates much more effortlessly.
- - Built-in Observability: Users can analyze logs and metrics, making it easier to pinpoint issues and optimize performance.
Jim Keller, Tenstorrent's CEO, expressed enthusiasm about the advantages of on-demand access, noting, "The ability to experience Tensix processors in a matter of minutes will allow developers to quickly develop, test, and run their models with greater efficiency. Cloud access is crucial for us as we aim to enable developers to explore our technology without upfront commitment."
User Feedback
The Wormhole Instances have already garnered attention from industry engineers. Erik Kaunismäki, a Software Engineer at Hugging Face, lauded the immediate access to Tenstorrent's hardware, stating, "Having instant access to the Wormhole multi-chip solution transforms how we can prototype and innovate in our projects."
Wormhole Technology
The Wormhole architecture is noteworthy for its built-in Network-on-Chip (NoC), facilitating a staggering 3.2 Tbps Ethernet connectivity between chips. This setup enhances data movement efficiency, essential for high-performance AI workloads. The technology is anticipated to provide significant computational capabilities in various applications, effectively comparable to a TT-LoudBox system.
Availability and Future Developments
As of now, these Wormhole Instances are available for private preview. Potential users are encouraged to visit Koyeb's website to request access and explore the benefits of Tenstorrent's groundbreaking technology. Tenstorrent continues to position itself at the forefront of AI hardware innovation, bolstered by partnerships with industry leaders and a commitment to developer satisfaction.
About Tenstorrent
Tenstorrent is a cutting-edge computing company specializing in AI technologies. With headquarters in North America and locations across the globe, including Toronto and Silicon Valley, Tenstorrent brings together experts in AI software and silicon design. Their endeavors are supported by several notable investors, ensuring that they remain a leader in the advancing field of AI computing. For more information, visit
www.tenstorrent.com.