inference/research
verifiedhigh · 0.85updated 2026-10-11

cadamcat: Eight ROCm kernel findings narrow to three on RDNA3

benchmark

Fields

id
sources/cadamcat-dual-radeon-rdna3-second-class
URL
https://cadamcat.github.io/dual-radeon-vllm/articles/rdna3-second-class.html
Publisher
cadamcat (dual-radeon-vllm)
Kind
benchmark
Accessed
2026-10-11
notes
Empirical analysis of ROCm kernel and framework behavior across dual RX 7900 XT (gfx1100) vs A100 under vLLM, documenting GQA ratio gates and PCIe AtomicOps hostcall dispatch issues.

Notes

Empirical analysis of ROCm kernel and framework behavior across dual RX 7900 XT (gfx1100) vs A100 under vLLM, documenting GQA ratio gates and PCIe AtomicOps hostcall dispatch issues.

Cross-references

This record references no other records.