Tenstorrent has two distinct
software approaches:
Great for production customers who want to get models up and running with ease. They want flexibility, but don’t have time to program new operations or contact support every time something changes.
Great for development customers who want to customize their models, write new ones or even run non-machine learning code. No black boxes, encrypted APIs, or hidden functions.
TT-Metalium is a low-level software platform for a heterogeneous collection of CPUs and Tenstorrent devices, giving users direct access to the RISC-V processors, NoC (Network-on-Chip), and Matrix and Vector engines within the Tensix Core.
The figure below shows the software layers that can be built on top of the TT-Metalium platform. With TT-Metalium, developers can write host and kernel programs that can implement a specific math operation (e.g., matrix multiplication, image resizing etc.), which are then packaged into libraries. Using the libraries as building blocks, various frameworks provide the user with a flexible high-level environment in which they can develop a variety of HPC and ML applications.
- Python APIs
- Easy to Use
- Intuitive, PyTorch-like
Beginner
- Batch
- Library of Fused OPs
- Op Selection & Parallelization
- Data formats & Math Fidelity
- Tensor Storage L1 or DRAM
- Augmented Python APIs
Intermediate
Write your own custom kernels & OPs
- Fused Kernels
- TMs
Go into Metalium Code
- Write C++ kernels
- Understand the Hardware
- Low-level Optimization
Expert