row bsw-r-ls-iq3-24g (R-LS-IQ3-24G, ROW_KIND=llama-server) · started 2026-10-09T09:36:37Z · uid 1000 · python 3.14.4
CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=1 TZ=UTC
pins: 34 files read their sha256 again in 0.6 s (the 2 shards were read at the window's open)
the build target: /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/CMakeCache.txt reads CMAKE_CUDA_ARCHITECTURES=86
band of record: predicted 31-59 tok/s (decode at 4K, median of 3, N = 34; C(34) = 127.80; outside 15-31 or 59-127.80; under the floor below 15; VOID above 127.80; N = 34, the laptop's fit, not refitted on this card: the prior road, section 2.2): AMENDMENT section 3.2
identity before the launch: the client's nine prompt digests equal L-IQ3-t8's (9/9)
port withheld-2026-10-13 reads free on the box (ss -ltn)
server argv: /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/bin/llama-server -m /workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf --host 127.0.0.1 --port withheld-2026-10-13 -ngl 99 --n-cpu-moe 34 -c 131072 -fa on -ctk q8_0 -ctv q8_0 -lm mmap --lazy-mode on -fit off -np 1 --cache-ram 0 --no-webui --jinja -lv 4 -t 8 -tb 8
the second fence (the unit's IP rules), inner.py's ipfilter probe: NOT ENFORCED (the connect to the box's own non-loopback address was not dropped: failed (ECONNREFUSED) (a refusal means the packet reached the host's stack))
sandbox: /usr/bin/bwrap --unshare-net --unshare-pid --die-with-parent --ro-bind / / --dev-bind /dev /dev --proc /proc --tmpfs /run --tmpfs /tmp --bind /workshop/card24-strata/results/R-LS-IQ3-24G/R-LS-IQ3-24G.run /workshop/card24-strata/results/R-LS-IQ3-24G/R-LS-IQ3-24G.run --ro-bind /run/nvidia-persistenced /run/nvidia-persistenced -- /usr/bin/python3 /workshop/card24-strata/kit/client/inner.py --spec /workshop/card24-strata/results/R-LS-IQ3-24G/R-LS-IQ3-24G.run/spec.json
sandbox note: /var/run is a link to /run (or absent): the /run tmpfs covers it
sandbox note: /run/nvidia-persistenced bound read-only (the driver's persistence socket)
cards before:
index, pci.bus_id, name, memory.used [MiB], memory.total [MiB], enforced.power.limit [W], temperature.gpu
0, withheld-2026-10-13, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 85276 MiB, 97887 MiB, 420.00 W, 32
1, withheld-2026-10-13, NVIDIA GeForce RTX 3090 Ti, 1 MiB, 24564 MiB, 300.00 W, 54
the pinned card's memory.used before the launch: 1.0 MiB
placement: hybrid (--n-cpu-moe 34) -- the registered line keeps 34 expert layers in host memory: recorded, never judged whole; offloaded 49/49; buffers ['load_tensors:   CPU_Mapped model buffer size = 35835.49 MiB', 'load_tensors:        CUDA0 model buffer size = 18870.50 MiB', 'load_tensors:   CPU_Mapped model buffer size = 27465.95 MiB']
build target: the server's start line reads CUDA : ARCHS = 860, the pinned build target
server maps: /workshop/bench-flashnext/cuda-13.4/targets/x86_64-linux/lib/libcublas.so.13.8.0.4, /workshop/bench-flashnext/cuda-13.4/targets/x86_64-linux/lib/libcublasLt.so.13.8.0.4, /workshop/bench-flashnext/cuda-13.4/targets/x86_64-linux/lib/libcudart.so.13.4.92, /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/bin/libggml-base.so.0.24.0, /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/bin/libggml-cpu.so.0.24.0, /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/bin/libggml-cuda.so.0.24.0, /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/bin/libggml.so.0.24.0, /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/bin/libllama-common.so.0.4.1, /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/bin/libllama-server-impl.so, /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/bin/libllama.so.0.4.1, /workshop/card24-strata/build/llama.cpp-3cf03257-sm86/bin/libmtmd.so.0.4.1, /usr/lib/x86_64-linux-gnu/libcuda.so.595.84
cards after:
index, pci.bus_id, name, memory.used [MiB], memory.total [MiB], enforced.power.limit [W], temperature.gpu
0, withheld-2026-10-13, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 85282 MiB, 97887 MiB, 420.00 W, 38
1, withheld-2026-10-13, NVIDIA GeForce RTX 3090 Ti, 1 MiB, 24564 MiB, 300.00 W, 56
ended 2026-10-09T10:15:54Z · inner rc 0 · last phase done · identity 9/9 · placement hybrid (--n-cpu-moe 34) · fences held ['netns']