ROCm 7.0.0

eol
Released
2025-09-16
Supported GPU architectures
CDNACDNA2CDNA3CDNA4RDNA2RDNA3RDNA4
Exact GPU targets (gfx)
gfx908gfx90agfx942gfx950gfx1030gfx1100gfx1101gfx1200gfx1201
Commercial GPU names
Instinct MI100Instinct MI200 series (MI210 / MI250 / MI250X)Instinct MI300 series (MI300A / MI300X / MI325X)Instinct MI350 series (MI350X / MI355X)Radeon RX 6800 / 6900 series (Navi 21)Radeon RX 7900 XTX / XT (Navi 31)Radeon RX 7800 XT / 7700 XT (Navi 32)Radeon RX 9060 / 9060 XT (Navi 44)Radeon RX 9070 / 9070 XT (Navi 48)
PyTorch builds available
2.5.02.6.02.7.0
Minimum kernel / driver
AMD GPU driver 30.10.x or the legacy 6.4.x/6.3.x/6.2.x KMD lines. Kernel per OS varies — e.g. Ubuntu 24.04.3: kernel 6.8 or 6.14. See AMD's ROCm 7.0.0 compatibility matrix for the full per-OS table.
Verified 2026-08 against AMD's official ROCm 7.0.0 compatibility matrix (docs-7.0.0). First ROCm 7.x release — adds CDNA4 (gfx950, MI350X/MI355X) support. Also supports TensorFlow 2.17.1-2.19.1, JAX 0.6.0, and ONNX Runtime 1.22.0 per the same matrix, not tracked in this site's schema yet. Release date confirmed 2025-09-16 against AMD's official ROCm release-history page (rocm.docs.amd.com/en/latest/release/versions.html). Status changed from an earlier 'supported' to 'eol' now that ROCm 7.14.0 and 10.0.0 have shipped and superseded it — see the 6.4.3 entry's notes for the full reasoning on this site's simplified status heuristic.

← All ROCm versions