Go Beyond Tops
AI Software Platform
Bring your workloads to CGLA.
The LENZO AI Software Platform combines AI and C compilers with remote FPGA access, giving developers a way to explore how their models and algorithms execute on LENZO’s programmable TokenProcessor™ architecture.
Discover what’s possible beyond theoretical compute ratings.
From your code to TokenProcessor™
The AI SOFTWARE PLATFORM provide a path to port existing workloads to the TokenProcessor™ development environment through our AI Compiler, C Compiler and even remote FPGA access.
AI Compiler
Bring your AI models to CGLA.
Built around MLIR, the AI Compiler translates supported model operations into a form that can execute within the CGLA environment.
Explore how AI workloads map to the architecture without starting from low-level accelerator programming.
C Compiler
Accelerate beyond neural networks.
The C Compiler is being developed to map selected C and C++ kernels onto CGLA’s parallel execution architecture.
Extend acceleration opportunities to compute-intensive algorithms, including image processing, signal processing, and custom operators.
Remote FPGA Access
Explore the architecture without local hardware.
Connect to LENZO’s FPGA-based development environment to test supported workloads and evaluate CGLA remotely.
Performance is more than a number
TOPS measures theoretical compute capacity. But it doesn’t capture the full picture of how a processor handles real workloads.
Actual performance depends on how computation is organized, how data moves through the architecture, and how much energy is required to get the work done.
The real question isn’t just how much a processor can compute. It’s how efficiently it puts that compute to work.
Parallel by design
Efficient by architecture
At the heart of TokenProcessor™ is CGLA — CPU Grounded Linear Array — LENZO’s programmable dataflow architecture.
CGLA distributes suitable computations across an array of processing elements, allowing multiple operations to execute in parallel.
Rather than repeatedly fetching and interpreting instructions for every operation, CGLA is designed to organize computation and data movement around the workload itself. This reduces processing overhead and creates opportunities for more efficient execution.
Unlike accelerators limited to a fixed set of neural-network operations, CGLA can be programmed for different computational patterns, including matrix operations, image processing, signal processing, and custom algorithms.
Parallel Computation
Multiple processing elements work concurrently on suitable operations, exploiting the parallelism within a workload.
Dataflow Efficiency
Computation and data movement are organized together to reduce unnecessary overhead and support efficient execution.
Programmable Versatility
The same underlying architecture can be used for different classes of computational workloads, extending its potential beyond conventional AI inference.
Don’t benchmark our model
Test yours
The most meaningful way to evaluate a processor architecture is with the workloads that matter to you.
The LENZO AI Software Platform provides a development path from existing AI models and computational kernels to CGLA, with FPGA-based access for testing supported workloads.
Explore how your workload maps onto the architecture, which operations can be accelerated, and where further optimization may be possible.
More than models
More of the workload
Physical AI systems rely on a continuous flow of computation — from processing sensor data to interpreting the environment and preparing actions.
Not all of that computation happens inside a neural network.
CGLA’s programmable architecture creates opportunities to accelerate suitable operations across these stages.
The AI Compiler and C Compiler provide complementary paths for mapping supported computation to CGLA, while the host CPU continues handling general-purpose application and control tasks.
One architecture. More of the workload.
Explore what’s possible with CGLA
The LENZO AI Software Platform is being developed for engineers and teams looking to evaluate new approaches to AI compute.
Whether you’re working with an existing AI model, a custom algorithm, or a computational bottleneck, we want to understand your workload.
Work with LENZO to explore compatibility, evaluate execution, and help shape the platform’s continued development.
Your workload
Our architecture
Contact LENZO to discuss your workload and access to the AI Software Platform.