Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.
C++
docs: the CPU picture encoder stays where the OS puts it (pinning it one per core was 7 to 19 times slower on a 16-core Windows PC)
Niko12214cb96acd814a
diff --git a/docs/DETAILS.md b/docs/DETAILS.mdindex e5e94af57..4677b4da3 100644--- a/docs/DETAILS.md+++ b/docs/DETAILS.md@@ -2095,10 +2095,15 @@ Only the workers and the main thread are pinned. Every other thread (the promptthe graphics driver's threads) is placed by the operating system, and the picture encoder is a separate process whosethreads are not pinned either. A task manager that shows only the first few logical processors busy is therefore notnecessarily the pool: while the experts are in graphics memory the workers have nothing to do. On a Ryzen AI MAX+ 395 (16-cores, 32 threads, Windows 11) with every expert in graphics memory, five snapshots of a running engine showed the-unpinned threads on logical processors 0 to 15 only (the first eight cores, both threads of each) and 16 to 31 idle, while-the 15 workers sat parked on 2, 4, ..., 30. With the experts on the CPU the same machine's workers were each at 100% on-exactly those even numbers (CPU pool alone on random expert data, so only the placement and the memory traffic are real).+cores, 32 threads, Windows 11, Balanced power plan) with every expert in graphics memory, snapshots of four runs of the engine+showed the unpinned threads on logical processors 0 to 15 only (the first eight cores, both threads of each) and 16 to 31+idle; Windows listed 16 to 31 as parked ("Parking Status" 1) while nothing pinned ran there. The 15 workers sat on 2, 4, ..., 30+and did nothing. With the experts on the CPU the same machine's workers were each at 100% on exactly those even numbers (CPU+pool alone on random expert data, so only the placement and the memory traffic are real). The CPU picture encoder+(`strata-vision`, 16 threads there) showed the same picture: 16 threads on logical processors 0 to 15, 10.1 s for a+1024x1024 picture. Pinning those 16 threads to one per core (2, 4, ..., 30) made it far slower (74 to 193 s) on that PC,+and 8 threads pinned to 8 real cores took 12.3 s, so the encoder stays where the operating system puts it (the cause was+not looked into; the parked second chiplet is the obvious suspect).That is why the default is not "the first N logical processors": on that machine, 8 workers on logical processors 1 to 8(four cores, both threads) read 67 GB/s, 8 workers on real cores 2, 4, ..., 16 read 94 GB/s, and the default 15 workers on2, 4, ..., 30 read 146 GB/s (median of three interleaved 10-second runs; 31 workers on 1 to 31 read 123 GB/s).