macOS: benchmark and expose a low-cost render profile without changing Windows defaults

[tcconnally]

Problem

The current macOS profile uses high-quality downscaling and debanding. The M5 trace showed about 92% GPU active residency and Heavy thermal pressure for most samples. We need measured render tiers, not an unverified global profile=fast switch.

Scope

• Add an explicitly named macOS-only low-cost profile behind opt-in.
• Benchmark deband, downscale, upscale, and correction changes one at a time.
• Keep Windows/Linux defaults unchanged.
• Record render time, normalized render CPU, GPU residency, paint gaps, and visual notes.

Acceptance criteria

• No profile is promoted without a fixed-corpus comparison.
• A selected profile meets the responsiveness gate without unacceptable quality loss.
• Platform tests prove option precedence and prevent Windows/Linux changes.

[tcconnally]

Implemented in PR #81, merged as 300e14c. HYPERWALL_RENDER_PROFILE=low-cost is available as an explicit macOS-only tier; HQ defaults and Windows/Linux options are unchanged. Fixed-corpus visual and responsiveness comparison remains the qualification step.

[tcconnally]

Native acceptance update

Current main is 4475e0ff8f9207f28f92f0067d9d0ba8b739094d. The explicit macOS-only HYPERWALL_RENDER_PROFILE=low-cost tier is present and Windows/Linux defaults remain unchanged. Promotion is not justified yet.

No current fixed-corpus HQ versus low-cost A/B exists with both responsiveness metrics and visual-quality evidence on the M5 target. The native target and current Emby configuration are unavailable in this session. Keep HQ as the default; this issue remains open.

[tcconnally]

Post-merge status: main is now b94be1ce3d86f06841ff458e3708d66efacd4ef1 after PR #84. PR #84 changed only the #79 capacity-evaluator contract; it produced no Mac/native measurements and does not change the open native blockers for this issue. The issue remains open.

[tcconnally]

Native pilot update for run 20260905T165329Z/profile-8-docked-clamshell-fixed on commit 2ce6de1c66ce7f4217249c59fa710268c47bc2bc. This was an HQ run with HYPERWALL_RENDER_PROFILE=hq; it was not an HQ versus low-cost A/B comparison. No fixed-corpus visual-quality evidence was collected, so no render profile is promoted and Windows/Linux defaults remain unaffected. Issue remains open.

[tcconnally]

HQ versus low-cost A/B update on commit 074258ddb6b1b2e3e1e8bd5cbe09742310c5d99b.

Both sides used the same 67-item Favorites manifest (0c1b04d2739bd0a151768b2f781644b505e440ad7a4e44562b52ce367895be5e), software decode, LG ULTRAGEAR+ (1) Wall 1x2, and a 2-minute phase.

HQ reference run: 20260905T195016Z/profile-2-favorites-initial

• p95 loop lag: 23.3 ms
• maximum paint gap: 502.145292 ms
• freezes/audio underruns/A/V drift: zero

Low-cost run: 20260905T204815Z/render-low-cost-2-favorites-match-hq

• p95 loop lag: 16.2 ms
• maximum paint gap: 492.153459 ms
• freezes/audio underruns/A/V drift: zero
• powermetrics complete; CPU mean 92.80%, GPU idle residency mean 79.96%, thermal Nominal=120

The earlier low-cost run on LG ULTRAGEAR+ (2) was excluded because it changed the display/layout variable. No visual-quality evidence was captured, so low-cost is not promoted and this issue remains open.

[tcconnally]

Current closeout boundary — main a858488152f1e57453c7ad98ad9c620a69dda407

The matched fixed-corpus HQ/low-cost pilot remains evidence for responsiveness only: HQ p95 loop lag 23.3 ms versus low-cost 16.2 ms, with no freezes, underruns, or A/V drift. No visual-quality capture was made, so low-cost is not promoted. The target Mac is unreachable from the current Linux host; the issue remains open pending visual comparison evidence.