Devlog Research Notes
Personal archive of coding experiments, research threads, and draft ideas.
Current Focus
- CRUSHCOMPILE
- CRUSHBLAS
- CPPDL
- Fused Transformer Kernels and Speculative CPU Decoding
- Optimal Hardware-Aware GEMM Kernel Code Generation
Hardware Specs
Main PC:
- CPU: Intel i9-9900K 5.1 GHz OC
- GPU: NVIDIA RTX 2080 Super 8GB VRAM
- RAM: 128GB DDR4 3600 MHz
- OS: Windows 10 LTSC
Compute Box:
- CPU: Ryzen 9-9950X3D 5.7 GHz
- GPU: 2X AMD 7900XTX 24GB VRAM (48GB VRAM Total)
- RAM: 192GB DDR5 5600 MHz
- OS: Ubuntu Server 24.04.3 LTS
last updated: 2026-01-30