Unlocking AI with Open Compute
Traditional AI processors force customers into closed hardware-software ecosystems that limit innovation. Ventana offers an open alternative. Built on RISC-V and delivered as soft chiplet IP, the Veyron E2X provides scalable, standards-based AI acceleration that customers can integrate and customize freely.
The Problem: Fragmented AI Acceleration
AI hardware is fragmented. Each major vendor has invented its own accelerator, compiler stack, and development model. This results in:
- Fragmented architectures
- Reinvented and incompatible software stacks
- No shared investment—no open alternative to CUDA
- Vendor lock-in and dependency
This fragmentation blocks ecosystem scale.
The Solution: RISC-V with RUCA
Ventana’s RUCA (RISC-V Unified Compute Architecture) provides the unified foundation AI needs:
- Shared ISA across CPUs and accelerators
- Scalable to fit across AI tiers from training to edge inference
- Common toolchains and compiler infrastructure
- Ecosystem leverage—shared investment, not vendor lock-in that limit innovation. Ventana offers an open alternative. Built on RISC-V and delivered as soft chiplet IP, the Veyron E2X provides scalable, standards-based AI acceleration that customers can integrate and customize freely.
Why RISC-V Works for AI
RISC-V has matured:
- Standard vector (RVV 1.0) extensions are here
- Matrix extensions are in progress, and Ventana is aligned with the emerging specification
- RISC-V provides a standard AI ISA than can leverages across vendors and market segements
- Frameworks like PyTorch, TensorFlow, LiteRT, and ONNX are gaining RISC-V support
- The RISE project is accelerating AI software ecosystem development
Ventana contributes to this movement while delivering practical, production-ready IP. Our early implementations include elements necessary to meet performance needs now, with a clear path to full compliance as the standard evolves. Customers get a clean software interface and a roadmap toward long-term RISC-V standardization.
Veyron E2X and RUCA: A Unified Soft Chiplet Architecture
Ventana’s Veyron E2X is part of our broader RUCA (RISC-V Unified Compute Architecture) platform. RUCA brings together scalar, vector, and matrix compute in a unified, scalable framework that spans data center AI to edge inference. Designed as a soft chiplet, Veyron E2X is portable, configurable, and optimized for integration into larger systems alongside general-purpose compute.
Each Veyron E2X core tightly integrates scalar, vector, and matrix execution with an efficient shared state microarchitecture:
- Supports a variety of data types, with the flexibility to extend support for additional ones as needed
- Aligned with the emerging RISC-V standard matrix extensions, with early implementations that address current performance needs while maintaining a clear path to full compliance
- Designed to push the boundaries of AI compute density—delivering significantly higher FP8 TOPS per core than any previous Ventana design. Early estimates suggest E2X will significantly exceed V3 in AI performance, with final numbers forthcoming from silicon and RTL evaluation.
- Enables customers to tailor tile count, precision, and memory interface
Build a Modular AI Platform
Ventana provides a complete, open platform for building scalable AI systems—from chiplets to silicon to servers. Customers get a head start with fully integrated system components:
- Veyron V2 or V3 compute clusters with AI compute for general-purpose workloads
- Open interconnects for modular integration, including chiplet-based packaging and optional 3D stacking
- Platform-level infrastructure including boot, debug, RAS, trace/profiling, and software stack
All that remains is tailoring the Veyron E2X soft chiplet IP to serve as the dedicated AI engine. This flexible approach enables customers to deploy high-performance AI compute on their terms, without building from scratch or locking into a fixed architecture.
Software Strategy and Ecosystem Alignment
AI software across RISC-V is still evolving, and PyTorch support is a top priority for enabling real-world deployment of machine learning models. Ventana actively participates in the broader RISE initiative and works closely with partners focused on AI framework enablement, ensuring strong alignment with evolving software standards without duplicating ecosystem efforts. Ventana is:
- Leveraging contributions from partners and the RISE AI working groups
- Supporting key frameworks like TVM, ONNX, TFLite, and tracking PyTorch enablement within RISE
- Delivering validated software stacks to early customers based on their V2+E2 configurations
While many companies raced ahead with proprietary AI accelerators, fragmenting the ecosystem and limiting software portability, Ventana is taking a grounded, collaborative approach—combining high performance with a commitment to open standards and long-term ecosystem alignment.
