Profile Example 07 GEMM All-Scatter with IntelliKit Metrix

[Copilot]

Runs the GEMM all-scatter benchmark (Example 07) on 2× MI300X GPUs and captures memory + compute metrics via IntelliKit Metrix.

Results

Throughput (fp16, 2 GPUs):

Shape	TFLOPS	Latency
4096×2048×4096	146.5	0.469 ms
8192×4608×36864 (default)	611.8	4.549 ms

persistent_gemm_all_scatter kernel metrics (M=4096, N=2048, K=4096, 3-replay avg):

Metric	Value
Kernel duration	602.9 µs avg (301–1177 range)
L1 / L2 hit rate	66.7% / 72.0%
HBM arithmetic intensity	~1206 FLOP/byte — compute-bound
Coalescing efficiency	25% — optimization opportunity
LDS bank conflicts	0

Changes

• profiling/profile_with_metrix.py — reusable script that runs memory + compute Metrix profiles against the benchmark and emits a merged JSON report
• profiling/metrix_profiling_results.json — captured results from MI300X
• .gitignore — allow examples/**/profiling/*.json to be tracked

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.