[Perf Regression] 20 config(s) regressed @ 345d6a53

[github-actions]

Performance Regression Detected

Commit: 345d6a53
Run: https://github.com/ROCm/ATOM/actions/runs/27974660529
Date: 2026-06-23T08:41:21.568770+00:00

Regressed Configurations

Model	ISL/OSL	Conc	Tput (cur)	Tput (base)	Δ%	TPOT (cur)	TPOT (base)	Δ%
DeepSeek-R1-0528	1024/1024	16	1175.0	1126.7	4.3%	13.17	13.79	-4.5%
DeepSeek-R1-0528	8192/1024	4	335.4	284.5	17.9%	11.29	13.42	-15.8%
DeepSeek-R1-0528 MTP3	1024/1024	4	522.5	633.9	-17.6%	7.16	5.82	23.1%
DeepSeek-R1-0528 MTP3	1024/1024	8	957.6	835.8	14.6%	7.87	9.14	-13.8%
DeepSeek-R1-0528 MTP3	1024/1024	16	1455.5	1578.8	-7.8%	10.49	9.51	10.3%
DeepSeek-R1-0528 MTP3	1024/1024	32	2127.8	2507.6	-15.2%	14.41	12.12	18.8%
DeepSeek-R1-0528 MTP3	8192/1024	64	2147.6	2208.8	-2.8%	27.66	27.18	1.8%
DeepSeek-R1-0528-MXFP4	1024/1024	256	5856.5	5621.4	4.2%	42.01	43.89	-4.3%
DeepSeek-R1-0528-MXFP4 MTP3	1024/1024	4	627.3	639.5	-1.9%	6.03	6.06	-0.6%
DeepSeek-V4-Pro	1024/1024	4	248.8	244.0	2.0%	15.36	15.72	-2.3%
DeepSeek-V4-Pro	1024/1024	16	857.3	872.4	-1.7%	17.98	17.73	1.4%
DeepSeek-V4-Pro DPA	1024/1024	64	1963.0	1981.7	-0.9%	30.07	30.25	-0.6%
DeepSeek-V4-Pro DPA	1024/1024	512	8776.9	8532.5	2.9%	54.02	56.92	-5.1%
DeepSeek-V4-Pro DPA TBO	8192/1024	512	4278.3	4061.6	5.3%	103.17	115.13	-10.4%
DeepSeek-V4-Pro MTP3	1024/1024	4	332.5	422.7	-21.3%	11.44	8.80	30.0%
DeepSeek-V4-Pro MTP3	1024/1024	32	1734.6	1837.8	-5.6%	17.01	16.55	2.8%
DeepSeek-V4-Pro MTP3	1024/1024	128	3669.0	3699.8	-0.8%	32.99	33.08	-0.3%
DeepSeek-V4-Pro MTP3	8192/1024	4	415.6	444.3	-6.5%	8.78	8.33	5.4%
DeepSeek-V4-Pro MTP3	8192/1024	64	1844.5	1857.5	-0.7%	32.65	32.44	0.6%
GLM-5.2-FP8	8192/1024	8	415.6	434.5	-4.4%	18.38	17.60	4.4%

Performance Summary

Summary not available

Profiler Traces

Download from workflow artifacts.
Open in Perfetto UI or Chrome chrome://tracing for analysis.

Next Steps

1. Download profiler-analysis-27974660529 artifact
2. Open trace files in Perfetto UI
3. Compare kernel durations against previous traces
4. Identify bottleneck changes