Local LLM Setup Live

GTX 1080 Ti + Xeon E5-2697A v4 running Qwen3.6-35B via llama-server-turbo

Hardware

ComponentDetails
CPUIntel Xeon E5-2697A v4 — 16c / 32t, 2.6 GHz, AVX2, 40 MB L3
GPUNVIDIA GTX 1080 Ti — 11 GB GDDR5X, Compute Capability 6.1
RAM64 GB DDR4 ECC
Storage458 GB NVMe SSD (23 GB used)
OSDebian 13 (bookworm), Linux 6.12.85, OpenClaw 2026.5.2

Model

Qwen3.6-35B-A3B-UD-Q4_K_M

Quantized MoE variant of Qwen3.6, 35B total params with ~3B active per token. Q4_K_M quantization keeps quality high while fitting comfortably on 11 GB VRAM.

PropertyValue
FormatGguf (GGUF)
QuantizationQ4_K_M
Context256,000 tokens
VRAM Usage~10 GB (with -ngl 99)
Active Params~3B (MoE routing)

Benchmarks

605 tok/s
Medium Prefill (6,422 tokens)
40 tok/s
Generation Speed
124 tok/s
Short Prompt Eval

Improvement Summary

Baseline prefill was ~420 tok/s (~15.2 s for 6.4k tokens). Adding --batch-size 4096 and --ubatch-size 1024 pushed prefill to ~605 tok/s (~10.7 s) — a ~43% improvement in prompt processing time.

Server Flags

llama-server-turbo
  -m /models/downloads/Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
  --port 8081 --host 0.0.0.0
  --cache-type-k turbo4 --cache-type-v turbo3
  --n-cpu-moe 30 -ngl 99 --no-mmap --mlock
  --jinja --chat-template-kwargs '{"enable_thinking": false}'
  --flash-attn auto --prio 2 --poll 80
  -c 256000 --ctx-size 256000
  --threads 16 --threads-batch 32 --parallel 1
  --batch-size 4096 --ubatch-size 1024

Flag Notes

FlagWhy
-ngl 99Offload all layers to GPU (11 GB VRAM handles it)
--cache-type-k turbo4 --cache-type-v turbo3TurboQuant KV cache — ~43% faster prefill
--flash-attn autoFlash attention enabled for compute 6.1+ (auto = safe for 1080 Ti)
--batch-size 4096 --ubatch-size 1024Larger batch = faster prefill, no generation impact
--threads-batch 32 --parallel 1Optimal thread/parallel balance for this CPU
--prio 2 --poll 80Higher scheduler priority, 80% CPU poll = lower latency
--chat-template-kwargs {"enable_thinking": false}Disables thinking output — required for OpenClaw/Telegram compatibility
--no-mmap --mlockLock model in RAM for consistent latency

Deployment

Managed via systemd at /etc/systemd/system/llama-server.service

sudo systemctl enable --now llama-server.service
curl http://127.0.0.1:8081/health
# → {"status":"ok"}

API endpoint: http://127.0.0.1:8081/v1/chat/completions

Known Issues

Context Overflow on Compaction

When the conversation summary grows to ~60k+ tokens, the full prompt re-processing can take 2+ minutes on the 1080 Ti. OpenClaw's default LLM timeout (30s) was too short, causing FailoverError: LLM request timed out.

Fix applied: Set models.providers.local-qwen.timeoutSeconds = 480 in ~/.openclaw/openclaw.json — gives the server 8 minutes to process large prompts.

Links