It's been 1 years working this research project Heterogeneous AI Super Cluster(HASC). A experimental computing platform can perform Quantum Simulations around 32-35 Qubits with ~15TFLOPS compute Power.
Achieved computing on PCHL, OpenCL/vulkan based custom AI distributed computing framework with multipal CPU architecture with model training system under low memory with PCHL's streaming system & for Quantum Research & development.
Designing PCHL Dual-State Computing Pipeline architecture in simulation and traditional model training workloads to efficiently distribute workloads under low resources.
Designed Custom Memory Engine Architecture (MEA) to pause and preserve computing workload states in the background enabling efficient context switching between different workloads under constrained memory resources with highly compressed state math
can be achievable under 1GB compressed state.
Designing Custom neural network architecture system highly modified version of transformer architecture redesigning for HASC,
researching on replacing the core architectural training process into something new.(experimental)
With the help of low-level Programming language PCHL (Parallel computing hybrid language) including Cell build for computing and use as a universal lang on HASC.
Currently PCHL runs on x86, Cell B.E.
future support on E.E ARM.
Note- Proper research and documentation still unavailable due to slow research.
Top comments (0)