Common questions about OXMIQ's licensable GPU IP and software. Don't see what you're looking for? Get in touch.
OxCore is licensable GPU silicon IP that fuses CPU, GPU, and TPU architectures into a single unified core. It includes three engines: OxOrc (scheduler), OxTen (tensor engine), and OxSIMT (CUDA-compatible engine).
OxCore unifies CPU, GPU, and TPU-style architecture in a single core. OxTen is its tensor engine, designed for leadership tensor operations, and it scales from 1 to thousands of cores.
Google's TPU is dedicated tensor silicon currently available only on Google Cloud. OxCore is licensable IP that fuses scalar (CPU-style), vector (GPU-style), and matrix (TPU-style) compute into one unified core that semiconductor companies can customize and tape out themselves.
Yes. OxSIMT, one of OxCore's three engines, is a CUDA-compatible engine that runs PTX directly, preserving the broadest CUDA ecosystem without a proprietary toolchain.
Yes. OxCore is delivered as licensable GPU IP that you can customize and tape out for your own SoC. Contact OXMIQ Labs to discuss licensing terms.
OxCore is designed to scale from chiplet to full SoC, and from single-core AI deployments at the edge to large-scale datacenter configurations.
OxQuilt is supply-chain-adaptive chiplet packaging IP. It provides the fabric layer that connects modular chiplets, from mobile-class designs to full AI pod configurations that are tuned precisely for the workload.
Monolithic GPUs commit to a single die at a single foundry, locking in cost and supply risk. OxQuilt assembles compute from modular chiplets, letting you mix foundries, process nodes, and memory ratios per workload, which is the basis of supply-chain-adaptive packaging.
OxQuilt is foundry-agnostic. It's designed to work across major semiconductor foundries so customers can source chiplets from the supplier that best matches their cost, performance, and geopolitical constraints.
OxQuilt is meant to be licensed alongside OxCore as a full-stack solution, not on its own. You can, however, license OxCore without OxQuilt.
Yes. Our configuration tool lets you lay out and customize your design before you commit, so you can adjust the layout to fit your requirements.
Yes. If you already have a chiplet in production, we can incorporate it into your custom Quilt design rather than starting from scratch.
Supply-chain-adaptive means the packaging strategy adjusts to which foundries, memory suppliers, and chiplet vendors are available, instead of committing to one supplier. Every ratio, from mobile to AI pods, can be tuned precisely for the target workload and supply chain.
OxCapsule is the developer tool for running AI models on any accelerator. It enables local-first inference for engineers building agentic AI without vendor lock-in.
No. OxCapsule is designed to run AI workloads across GPUs, accelerators, and CPUs, including NVIDIA, AMD, Intel, and Tenstorrent silicon. Additional architectures can be supported on request.
OxCapsule is currently in free beta. Sign up at oxmiq.ai/beta for early access.
OxCapsule supports any open-source AI models out of the box.
OxCapsule runs on Linux, macOS, and Windows.
OxPython is a runtime optimized for heterogeneous silicon. It lets developers run AI models seamlessly across different accelerators without code changes.
No. OxPython supports PyTorch workflows out of the box. Your existing code runs on OxPython without rewrites.
OxPython supports OxCore and Tenstorrent silicon natively. Additional architectures can be supported on request.
OxPython is currently a proprietary runtime developed by OXMIQ Labs. Contact us for licensing details.
OxPython is the name of OXMIQ's AI stack. It natively supports PyTorch code that targets CUDA devices and is implemented in both Python and C++.