Repository navigation
bench: RX 7900 XTX + Ryzen 5 7600X tuning on Linux; FORK.md - #1
Merged
Merged
Conversation
A/B sweeps on this fork's machine against setup's configuration, all runs in runs.csv with the measurement scripts: - --kv-resident 32768 + --prefill auto:32768: decode +4%, a 53.7k prompt read 22% faster; -4% decode at 109k tokens (K/V streamed from RAM). - --pcie-frac 0.35 (the probe picks 0.55): decode +6% on top, the only PC-dependent setting that pays here; 0.25 and 0.75 are slower. - Neutral: --host-core last, --pool-workers 4, --spec-min-p 0.3/0.7, the GPU COMPUTE power profile. --lookup-chain trades prose for edits. - Images through the CPU encoder: no VRAM cost, ~1.4 s per 1280x800 PNG. FORK.md records what ash/main carries on top of upstream, what was tested and not merged, and how the fork is updated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Measurements from the machine this fork runs on (RX 7900 XTX 24 GB, Ryzen 5 7600X, 64 GB DDR5, Ubuntu 26.04, HIP source build), plus a FORK.md that records what
ash/maincarries. No code changes.Report:
bench/results/2026-10-07-rx-7900-xtx-linux-tuning/- README, every run inruns.csv, the measurement scripts and the test image. Strata's PC-dependent defaults were measured on a Ryzen 5 7600 + RTX 5070 and setup doesn't calibrate on AMD yet, so these are A/B sweeps against setup's configuration:--kv-resident 32768+--prefill auto:32768--pcie-frac 0.35(probe picks 0.55)--host-core last,--pool-workers 4,--spec-min-p 0.3/0.7, GPU COMPUTE profile--lookup-chain 2/4FORK.md: the A/B-only merge policy, what
ash/mainadds on top of upstream 0.1.40.2 (Niko1221#1126 + thene1 == 0fix, Niko1221#1107, the Niko1221#1107/Niko1221#789 conflict resolution), the PRs tested and not merged, and the update steps.Checked: no hostnames, credentials or home paths in the added files; the scripts take the llama.cpp source from
$LLAMA_CPP_SRC.🤖 Generated with Claude Code