Koyeb Tenstorrent Instances

Product · API / developer platform by Koyeb

Access open, scalable, and powerful next-gen AI hardware in seconds.

Koyeb Tenstorrent Instances provide on-demand access to next-generation AI hardware through a serverless platform. Users can build, optimize, and deploy AI models efficiently with support for various data formats like FP8 and BF16. The platform automates infrastructure management with features like scale-to-zero, instant API endpoints, and built-in service mesh. It allows teams to transition from single-user development to multi-user production environments in seconds.

Languages
Python, C++
Accelerator Type
Tenstorrent Wormhole PCIe Boards
Architecture
RISC-V Powered Tensix Cores
On-chip RAM per board
Up to 96GB
Data Formats
FP8, FP16, BF16, FP32
vCPU (N300s)
4 vCPU
RAM (N300s)
32GB RAM

Overview

Koyeb Tenstorrent Instances provide on-demand access to next-generation AI hardware through a serverless platform. Users can build, optimize, and deploy AI models efficiently with support for various data formats like FP8 and BF16. The platform automates infrastructure management with features like scale-to-zero, instant API endpoints, and built-in service mesh. It allows teams to transition from single-user development to multi-user production environments in seconds.

AI Hardware Acceleration

  • Wormhole PCIe Boards: Access Wormhole PCIe boards powered by Tensix Cores for AI acceleration.
  • Data Precision Support: Supports broad data precision formats including FP8, FP16, BF16, and FP32.
  • High-Performance Memory: Features up to 96GB of on-chip RAM and high-speed interconnects for large-scale training.
  • RISC-V Architecture: Utilizes RISC-V architecture for a customizable and scalable AI development environment.

Developer Experience

  • Instant Dev Environments: Deploy ready-to-use development environments with one click and connect via VSCode.
  • ML Libraries: Use TT-NN libraries as building blocks for flexible high-level ML applications.
  • Container Support: Run full-fledged Docker containers with zero configuration required.

Infrastructure Management

  • Scale-to-Zero: Resources automatically scale up with demand and down to zero when traffic drops.
  • Instant API Endpoints: Provision API endpoints ready to handle requests in seconds with no manual config.

Pricing

N300s (24GB VRAM)
Request access
4x N300s (96GB VRAM)
Request access

Specifications

Accelerator Type
Tenstorrent Wormhole PCIe Boards
Architecture
RISC-V Powered Tensix Cores
On-chip RAM per board
Up to 96GB
Data Formats
FP8, FP16, BF16, FP32
vCPU (N300s)
4 vCPU
RAM (N300s)
32GB RAM
Disk (N300s)
320GB NVMe

Added by AntSpace from public sources (koyeb.com).

Related

More products by Koyeb