Status

● Connected

ROCm

6.1.2

System

MI300X · 8 GPUs

Project

Retail-LLM-Inference

AK

Admin User

Administrator

Advanced Kernel Preview

Kernel Lab

Experimental module for kernel-level analysis, attention path optimization, and memory transfer improvement on AMD GPUs.

Kernel Speedup

1.82x

vs baseline path

Target Operation

Attention Kernel

Memory-bound LLM inference path

Correctness

Passed

Output validation successful

Why this matters

Kernel Lab gives YOUSUN Axion advanced technical depth. It shows that optimization does not stop at basic code changes. Axion can also identify kernel-level speedup opportunities for AMD GPU workloads, making the project stronger for technical judging.