Archive matched Qwen Turbo Q2 image and memory evidence

runs/3cc53336-2555-5e90-99c3-7b94dcb40acc/run.md

@@ -1,6 +1,6 @@
 # Images, music, and the limits of video on a 6 GB GTX 1060
 
-An older gaming GPU can still produce a useful library of images and short music clips. Our October 2026 Packet Foundry campaign on a Windows GTX 1060 with 6 GB VRAM now retains **143 generated files: 130 images, 11 videos and two music clips**, across **150 recorded attempts**. The images include the 100-request FLUX study, ten earlier checks, ten Qwen Turbo Q4 follow-up images and ten matched Q8 comparison outputs. The user judged both music samples usable; videos range from recognizable but flawed motion to colored noise. Every poor output and all seven no-output failures remain in the archive.
+An older gaming GPU can still produce a useful library of images and short music clips. Our October 2026 Packet Foundry campaign on a Windows GTX 1060 with 6 GB VRAM now retains **153 generated files: 140 images, 11 videos and two music clips**, across **160 recorded attempts**. The images include the 100-request FLUX study, ten earlier checks, ten Qwen Turbo Q4 follow-up images ten matched Q8 comparison outputs and ten Q2 loading-policy comparison outputs. The user judged both music samples usable; videos range from recognizable but flawed motion to colored noise. Every poor output and all seven no-output failures remain in the archive.
 
 The most useful result is the range: roughly a minute for a 512-pixel image in the large FLUX.3 study, around eleven minutes for fifteen seconds of music, and anything from under two minutes to more than an hour for a short video. Those numbers describe these exact configurations and timing boundaries. They are not a general ranking of the models.
 
@@ -108,6 +108,28 @@ File size leaves out dequantization buffers, compute workspaces, activations, en
 
 [Compare all ten Q4/Q8 image pairs](https://benchmarks.weeyuga.com/benchmarks/ef874435.html) · [Download exact matched timings, cache context and pair mapping](https://benchmarks.weeyuga.com/data/predator-media-20261021/qwen-turbo-q4-q8-comparison.json).
 
+### Qwen Turbo Q4, Q8 and Q2: the same ten briefs and seeds
+
+Ten Q2_K outputs reuse the exact ten Q4/Q8 briefs and seeds 42–51; the twenty earlier originals remain unchanged. The dimensions, eight Euler steps, CFG1, sigma schedule, CPU encoder/VAE and FP32 compute remain fixed. Q2 also uses a new Comfy process, default normal VRAM with a 0.6 GiB reserve, forced full-generator loading and explicit instrumentation. Those runtime changes matter when reading the timing table.
+
+| Matched workload | Q4 native | Q8 native | Q2 native |
+|---|---:|---:|---:|
+| Initial request; actual process/cache state | 320.246s | 541.410s | 217.040s |
+| Same first brief, next seed | 95.234s | 93.391s | 90.960s |
+| Eight new briefs, retained-runtime median | 199.712s | 446.477s | 196.433s |
+
+At the observed load boundary, sampler entry, all eight post-denoiser/pre-Euler-update callbacks and final completion, Q2’s registered generator parameters and buffers were on CUDA with zero partial-offload accounting. Peak Torch allocated memory across the ten requests was 3.278 GiB. The original sampler and callbacks remain in use; synchronization/inventory checks add measurement overhead.
+
+Real one-second target per-process WDDM observations preserve their actual successful-sample gaps and idle baseline. The three-second coverage check passed for 7/10 whole-graph windows and 9/10 generator windows; the largest measured gaps were 6.406083s and 6.074722s respectively. Coverage failures stay marked as failures. The completed observer retained 3 query errors in total; 3 graph windows and 1 generator windows contained errors. Failed queries remain errors rather than synthetic samples. A separate post-run analysis retains the unchanged strict verifier's outcome and distinguishes these coverage windows; no additional image was generated to repair a measurement gap.
+
+The largest observed shared-memory increase over that process/adapter baseline in a generator window was 4.0 MiB. These are observations at sampled times, not bounds on unobserved gaps or proof that every generator weight remained physically in dedicated VRAM throughout execution. Idle shared usage can already be nonzero. The exact coverage and counters are included beside each result.
+
+A setup gate first rejected the unsupported `--normalvram` argument before generation. Removing `--lowvram` selects this pinned version’s normal default. The first WDDM observer also stopped after a CIM counter-query error, with zero verified samples and zero observed generation requests; its cause remains unresolved, and its failed state, logs and source remain retained. These are setup/measurement failures outside the generation-attempt count. Twelve source-hash differences were verified as exact LF-to-CRLF serialization changes, with native byte hashes still pinned.
+
+All ten outputs passed actual relay/acknowledgement, full PNG decoding/checksums, trace attachment retrieval and three comparable native ten-case experiments. Every original was visually inspected, with defects retained. No automatic aesthetic winner or isolated quantization-speed effect follows from these unrandomized runs at different times. Only the first two requests share a brief; five variants were not measured.
+
+[Compare all ten Q4/Q8/Q2 image triples](https://benchmarks.weeyuga.com/benchmarks/af3885e2.html) · [Exact matched timings, stage intervals and residency observations](https://benchmarks.weeyuga.com/data/predator-media-20261021/qwen-turbo-q4-q8-q2-comparison.json).
+
 ## Music: usable to the listener, expensive per second
 
 YuE2 INT8 ConvRot produced two fully decoded fifteen-second stereo FLAC files at 48 kHz. Factory traces took **700.26 and 634.17 seconds**, approximately 11 minutes 40 seconds and 10 minutes 34 seconds. The native graph durations were 696.921 and 630.994 seconds. Loader state remained available between requests; the second request changed the style wording and seed, so semantic generation and acoustic sampling both ran again.
@@ -155,4 +177,4 @@ Other retained records include the cancelled YuE2 CPU trial, Cog's initial atten
 
 For accepted outputs, the delivery checks verify checksums, full decoding, durable collection and acknowledgement. Saved LangSmith attachment bytes were independently decoded, and native experiment exports were checked where recorded. LangSmith's media-player UI was not separately inspected. The original noisy Cog videos passed file and delivery checks while failing visual usefulness: the separation is deliberate.
 
-The practical lesson from this machine is specific. The FLUX study completed all 100 planned images, the Qwen Turbo Q4 follow-up completed ten, and the matched Q8 comparison added ten further outputs; each measured subset remains separate. Two short music samples were usable to their listener, at a substantial time cost. Short video could complete, sometimes quickly, but our tested clips did not establish reliable requested motion. The published prompts, settings, poor outputs and failed attempts give the next builder something more actionable than a success headline.
+The practical lesson from this machine is specific. The FLUX study completed all 100 planned images; the Qwen Turbo Q4, Q8 and Q2 follow-ups each added ten outputs. Every measured subset remains separate. Two short music samples were usable to their listener, at a substantial time cost. Short video could complete, sometimes quickly, but our tested clips did not establish reliable requested motion. The published prompts, settings, poor outputs and failed attempts give the next builder something more actionable than a success headline.