Repository navigation
[Docs] add a Windows Strix Halo (gfx1151) deployment guide - #3029
HaiYunChuang wants to merge 2 commits into
Conversation
|
Have you seen my community test suite reports #3032 for the Radeon 8060S (gfx1151) on a Ryzen AI Max+ 395 (Windows 11)? They’re for version 0.44.1, though. |
|
Thanks for the pull request. We're resetting the tracker for 0.2.0-beta. There are more than 1,300 open issues and pull requests, too many for us to review properly, so we're closing everything that isn't being worked on right now. Closed here does not mean rejected. To reopen it takes about two minutes:
The rules are in CONTRIBUTING.md. |
|
/reopen |
|
Not reopened yet. Still missing:
Edit the pull request and comment |
|
/reopen |
|
Reopened, thank you. A maintainer will pick it up from here. If it conflicts with main, rebase it first; that speeds up review. |
…n on 0.2.7 (#3029) Structure from the guide in PR #3029 by HaiYunChuang (what the PC reaches, memory and the BIOS, the configuration, what is on by default, traps), rewritten for the 0.2.7 install and with every number measured again: the Terminal and the gfx1151 engines built from f-027-int2 (3e468f5de), a local signed test feed, strata launch as a standard test user on the NucBox EVO-X2 (Adrenalin 26.10.1), tools/strata_bench.py over a tunnel. Qwen IQ3_S: 44 tok/s chat, prompts 755 / 821 / 792 tok/s at 4K / 16K / 32K. DeepSeek 2.00 bpw: 12 to 14 tok/s chat, prompts 214 to 342 tok/s. An interleaved A/B of STRATA_SH_STREAM on Windows gfx1151 (three pairs): off is 6.9% faster in chat decode, prompts unchanged. What launch chose and logged is quoted instead of a hand-made config. The first guide's tuning rows (engines 0.1.41 to 0.1.47) are not repeated. Co-authored-by: HaiYunChuang <97429634+HaiYunChuang@users.noreply.github.com>
Ported PRs (authors kept) and our fixes, plus the owner's part-2 items: the restored TROUBLESHOOTING entries with a link check, #3895 (no shared-expert fork calibration on Windows HIP with --kv-resident; a warning on a hand-set STRATA_SH_STREAM=1), #3930 (docs), #3421 (Docker stops when models are present and MODEL_PATH is unset), #3570 (launch --show-key reads setup's config), #3311 (engine paragraph only) and #3029 (Strix Halo Windows guide, re-measured on 0.2.7).
|
Thank you for the guide and the A/B table. 0.2.7 has docs/STRIX_HALO_WINDOWS.md, which follows the structure of yours and was measured again on a Ryzen AI Max+ 395 (Windows 11, Adrenalin 26.10.1) with the 0.2.7 candidate. Your finding that STRATA_SH_STREAM is slower on Windows held (decode 43.6 to 46.6 tok/s, +6.9%, same token ids), so it is off by default on Windows gfx1151 now. Fixed in 0.2.7. Release: https://github.com/Niko1221/Strata/releases/tag/v0.2.7 |
Problem and result
Windows users of Strix Halo (Ryzen AI Max, gfx1151) have no deployment guide: the defaults that work on Linux do not all transfer to Windows, and the setup notes are spread across issues. This PR adds a guide for a working 256K configuration and, in section 3.6, the first interleaved A/B of the gfx1151 defaults measured on Windows (engine 0.1.47).
Before: no Windows gfx1151 document. After:
docs/STRIX_HALO_WINDOWS.md.Related discussion
Issue #3032 (community reports for the same machine). No maintainer agreement on scope was requested for a docs-only change.
Approach
One new file under
docs/. It records the configuration, the Windows-specific traps, the diagnostics, and the measured results. No engine, default or code change. The change touches only that file, so it applies tomainas it is.Validation
Revision tested:
mainat011c9e60(v0.2.2), engine 0.1.47 (strata-engine-0.1.47-windows-x86_64-hip-rdna3, gfx1151), Windows 11, Radeon 8060S, 128 GB. All checks below were run on that revision, after updating to it.Passed:
/healthloaded=true, max_context 262144;/metricsengine.version=0.1.47; logsession is up (engine 0.1.47);hipBLASLt tuning enabled (32 rows, gfx1151, version 100500).STRATA_GFX1151_DEFAULTS=0534.5 (-6.2%) / 46.5 (-17%);STRATA_REFILL_PARTS=1561.8 (+1.0%) / 43.1;tools/hiptable 594.5 / 45.8 vs the package table 576.6 (-3.0%) / 43.9 (-4.1%).STRATA 7391and a red circle from a test image.Failed:
Not run:
Limits
The guide reflects one machine (Ryzen AI Max+ 395 / 8060S, Windows 11, 128 GB). The 18 defaults were measured as a group; individual switches were not bisected. Absolute numbers are valid only within one interleaved run: the same configuration drifted +28% over a session.