ARTIFACT // BENCH-PANTHER-LAKE-X2-ELITE-LLM.SH (SH)
Snapdragon X2 Elite Extreme & Intel Core Ultra X9 388H (Panther Lake) Fixed Laptop LLM Benchmark (bench-panther-lake-x2-elite-llm.sh)
Checks native binary architecture (ARM64 vs x86_64), lists active Vulkan/SYCL iGPU devices (Adreno X2-90 or Arc B390 12 Xe3), runs llama-bench pp2048/tg128 x3 with Q8_0 KV cache, and logs a sustained decode run to catch thin-and-light thermal throttling.
[KEY VERIFIED TAKEAWAY / TL;DR]This reproducible configuration artifact "Snapdragon X2 Elite Extreme & Intel Core Ultra X9 388H (Panther Lake) Fixed Laptop LLM Benchmark" (bench-panther-lake-x2-elite-llm.sh, SH) is paired with Tinker AI Lab's hardware-verified guide "snapdragon-x2-elite-extreme-128gb-vs-intel-core-ultra-x9-388h-panther-lake-xe3-local-llm". It exposes 3 customizable browser variables (MODEL_PATH=./models/Qwen3-30B-A3B-Q4_K_M.gguf, LLAMA_BIN_DIR=./build/bin, SUSTAIN_SECONDS=600) with loopback-first security defaults.
01 // HOMELABLocal AI / HomeLab· Verified: 2026-11-06
Browse All 42 Artifacts →bench-panther-lake-x2-elite-llm.sh
View Paired Guide →Snapdragon X2 Elite Extreme & Intel Core Ultra X9 388H (Panther Lake) Fixed Laptop LLM Benchmark
Checks native binary architecture (ARM64 vs x86_64), lists active Vulkan/SYCL iGPU devices (Adreno X2-90 or Arc B390 12 Xe3), runs llama-bench pp2048/tg128 x3 with Q8_0 KV cache, and logs a sustained decode run to catch thin-and-light thermal throttling.
Customizable Variables (Saved in Browser)
UTF-8 Encoded · Live Client-Side Compilation
#!/usr/bin/env bash # Tinker AI — Thin-and-Light AI Laptop Fixed Benchmark: Snapdragon X2 Elite Extreme / Intel Panther Lake (Guide #39) set -euo pipefail MODEL="./models/Qwen3-30B-A3B-Q4_K_M.gguf" BIN="./build/bin" SUSTAIN="600" OUT="bench-laptop-$(hostname)-$(date +%Y%m%d-%H%M).md" echo "=== [1/4] Verifying Model Hash & Native Binary Architecture ===" sha256sum "$MODEL" | tee -a "$OUT" echo "- host arch: $(uname -m)" | tee -a "$OUT" file "$BIN/llama-bench"* 2>/dev/null | head -n 1 | tee -a "$OUT" || true echo "=== [2/4] Enumerating Vulkan / SYCL iGPU Devices (Arc B390 Xe3 or Adreno X2-90) ===" "$BIN/llama-bench" --list-devices 2>&1 | tee -a "$OUT" || true echo "=== [3/4] Running llama-bench pp2048 / tg128 x3 (Flash Attention + Q8_0 KV) ===" "$BIN/llama-bench" -m "$MODEL" -ngl 99 -fa 1 -ctk q8_0 -ctv q8_0 -p 2048 -n 128 -r 3 -o md | tee -a "$OUT" echo "=== [4/4] Sustained Decode ($SUSTAIN s) — Thin-and-Light Thermal Throttle Check ===" timeout "$SUSTAIN" "$BIN/llama-cli" -m "$MODEL" -ngl 99 -fa -ctk q8_0 -ctv q8_0 -n -1 --ignore-eos \ -p "Write a detailed architecture analysis comparing 192-bit vs 128-bit LPDDR5X unified memory for local MoE inference." \ 2>&1 | grep -E "eval time|tokens per second" | tail -n 3 | tee -a "$OUT" || true echo "Saved benchmark report to $OUT"
PARAMETER SPECIFICATION // 参数化变量规范说明表
Customizable Parameter Reference for bench-panther-lake-x2-elite-llm.sh
This SH artifact exposes 3 template variables compiled client-side in your browser.
| Template Token | Parameter Description | Verified Default | Example Format |
|---|---|---|---|
{{MODEL_PATH}} | GGUF model path | ./models/Qwen3-30B-A3B-Q4_K_M.gguf | ./models/Qwen3-30B-A3B-Q4_K_M.gguf |
{{LLAMA_BIN_DIR}} | llama.cpp binary directory (Vulkan / SYCL / ARM64) | ./build/bin | ./build/bin |
{{SUSTAIN_SECONDS}} | Sustained decode duration (seconds) | 600 | 600 |
PAIRED HARDWARE MANIFEST // 配套实测环境与边界记录
Verified Environment for "Thin-and-Light AI Laptops for Local LLMs (Fall 2026): Snapdragon X2 Elite Extreme (192-bit · 228 GB/s · 128GB) vs Intel Core Ultra X9 388H "Panther Lake" (Arc B390 12 Xe3 · 153.6 GB/s) — Bandwidth Ceilings, NPU Myths & Reproduction Protocol"
Tested Hardware & OS
Spec basis (not lab-measured): Qualcomm Snapdragon X2 Elite Extreme X2E-96-100 (TSMC 3nm · 18-core 3rd-gen Oryon up to 5.0 GHz · Adreno X2-90 GPU · 80 TOPS Hexagon NPU · 192-bit LPDDR5X-9533 @ 228 GB/s · up to 128GB) vs Intel Core Ultra X9 388H "Panther Lake" (Intel 18A · 16C: 4P+8E+4LPE · Arc B390 12 Xe3 cores + XMX · 50 TOPS NPU 5 · 128-bit LPDDR5X-9600 @ 153.6 GB/s · up to 64GB/96GB)
Pinned Stack & Model Versions
llama.cpp master (2026-11) — Vulkan / SYCL (x86_64) on Intel Panther Lake; Vulkan / ARM64 NEON on Snapdragon X2OpenVINO GenAI 2026.x (Intel NPU 5 + Xe3 hybrid execution) / Qualcomm QNN & Genie SDK (Hexagon NPU INT4)Ollama (latest Windows x64 & Windows ARM64 native installer)Models: Llama-3.2-3B / Qwen3-8B / Qwen3-14B / Qwen3-30B-A3B / Qwen3-Next-80B-A3B (GGUF Q4_K_M)
Offline Boundary (Fully Offline Local Execution After Model Download)
All llama.cpp (Vulkan/SYCL), Ollama, OpenVINO, and QNN models execute locally offline. Windows Copilot+ Recall and OEM cloud assistants are excluded.
Known Operational Limits & Failure Modes
- Evidence level: SPEC-DERIVED ESTIMATE. Decode tok/s figures in Table 2 are computed from published memory bandwidth (bandwidth × 0.68 ÷ active-weight bytes per token) and have not yet been measured on retail laptops in our lab.
- Crucial SKU trap on Qualcomm: ONLY the top-bin **Snapdragon X2 Elite Extreme (`X2E-96-100`)** has the **192-bit memory bus (228 GB/s)**. Standard `Snapdragon X2 Elite` (`X2E-94-100` / `X2E-90-100`) drops to a **128-bit bus (~152 GB/s)**—matching Panther Lake’s bandwidth.
- Neither the 80 TOPS Hexagon NPU nor the 50 TOPS Intel NPU 5 is used by standard `llama.cpp` or `Ollama` out of the box. Running models on the NPU requires `OpenVINO GenAI` (Intel) or `Qualcomm QNN / Genie / Nexa SDK` (Snapdragon) with static INT4 compilation and typically caps out at 3B–8B dense models.
DEPLOYMENT CHECKLIST // 配置落地核验步骤
3-Step Verification Checklist After Deploying bench-panther-lake-x2-elite-llm.sh
- Verified exact SoC SKU: confirmed `X2E-96-100` (192-bit, 228 GB/s) vs non-Extreme `X2E-94/90` (128-bit, ~152 GB/s) or `Core Ultra X9 388H` (128-bit, 153.6 GB/s).
- Confirmed `llama-bench --list-devices` detects `Intel Arc B390` (SYCL/Vulkan) or `Adreno X2-90` (Vulkan) and that the binary is native (`ARM64` or `x86_64`).
- Compared initial `llama-bench` burst decode speed against a 10-minute sustained generation run on laptop cooling.