Graphcore Tech Stack
AI accelerator silicon and software stack for datacenter inference and training
Graphcore designs custom silicon and software for AI workloads, operating as a vertically integrated semiconductor house under SoftBank ownership. The tech stack—PyTorch, JAX, TensorFlow paired with PCIe Gen 6, 800G Ethernet, and high-performance interconnect fabrics (NCCL, MPI, libfabric)—reflects a company building end-to-end compute from chip through datacenter orchestration. Active hiring in validation and firmware engineering, combined with heavy project focus on silicon bringup and post-silicon characterization, signals they are scaling manufacturing readiness while optimizing kernel efficiency on their hardware.
Tech Stack 167 technologies
What Graphcore Is Building
◆Challenges
- Improving test infrastructure
- Detecting hardware defects
- Ensuring silicon reliability before production
- Improving coverage
- Scaling ai models to thousands of accelerators
- Improving yield
- Optimizing validation efficiency
- Ensuring supply continuity
- Improving power reliability and efficiency
- High-quality product delivery
▲Active Projects
- Silicon validation tests
- Optimising kernels for efficiency on hardware
- Reference applications
- Post-silicon validation
- Prototype bring-up and validation
- Zephyr-based firmware for hyperscale server and rack management platforms
- Automated test framework development
- Bring-up and characterisation of silicon devices
- Silicon bringup sequences
- Post-silicon characterisation of advanced socs
Hiring Activity
Department
Seniority
Notable leadership hires: AI SoC Validation Lead, Hardware Validation Lead
