Qualcomm Snapdragon X Elite

Other

37

Run records

14

Models covered

4

Frameworks covered

L1 Reproduced × 1 · L2 Cross-framework verified × 36

Evidence mix

Typical measured performance

Published benchmark averages per model × framework × quantization on this hardware.

  1. llama.cpp · Q2_K Decode 83.91 · Prefill 412.44 ·
  2. vLLM · FP16 Decode 32.76 · Prefill 107.43 ·
  3. Strata · Q4_K_S Decode 59.96 · Prefill 299.18 ·
  4. llama.cpp · Q4_K_M Decode 112.42 · Prefill 515.25 ·
  5. Ollama · IQ4_XS Decode 117.58 · Prefill 399 ·
  6. Ollama · Q8_0 Decode 69.82 · Prefill 301.4 ·
  7. llama.cpp · Q3_K_L Decode 50.63 · Prefill 217.32 ·
  8. vLLM · Q3_K_L Decode 51.08 · Prefill 205.33 ·
  9. vLLM · Q2_K Decode 69.03 · Prefill 343.27 ·
  10. vLLM · Q3_K_S Decode 55.63 · Prefill 229.65 ·
  11. Strata · IQ2_M Decode 86.12 · Prefill 422.03 ·
  12. Strata · Q3_K_M Decode 73.79 · Prefill 405.35 ·
  13. vLLM · IQ4_XS Decode 66.12 · Prefill 360.69 ·
  14. vLLM · Q3_K_M Decode 74.16 · Prefill 385.13 ·
  15. Strata · FP16 Decode 39.4 · Prefill 159.35 ·
  16. vLLM · IQ4_XS Decode 100.93 · Prefill 494.53 ·
  17. Ollama · Q3_K_M Decode 88.26 · Prefill 401.25 ·
  18. llama.cpp · Q2_K Decode 81.63 · Prefill 398.71 ·
  19. Strata · Q8_0 Decode 48.7 · Prefill 212.72 ·
  20. llama.cpp · Q3_K_L Decode 48.96 · Prefill 229.97 ·
  21. Ollama · IQ2_M Decode 54.16 · Prefill 255.41 ·
  22. Ollama · IQ4_XS Decode 131.69 · Prefill 717.01 ·
  23. Strata · FP16 Decode 96.34 · Prefill 381.84 ·
  24. vLLM · IQ2_XS Decode 103.58 · Prefill 468.29 ·
  25. vLLM · Q5_K_S Decode 65.67 · Prefill 300.03 ·
  26. Ollama · Q2_K Decode 89.58 · Prefill 387.16 ·
  27. Ollama · Q3_K_S Decode 76.04 · Prefill 366.79 ·
  28. vLLM · Q4_K_S Decode 152.62 · Prefill 807.12 ·
  29. Strata · Q6_K Decode 148.28 · Prefill 791.63 ·
  30. Ollama · Q4_K_M Decode 76.52 · Prefill 395.44 ·
  31. Strata · Q2_K Decode 154.6 · Prefill 695.42 ·
  32. Strata · Q3_K_L Decode 121.82 · Prefill 519.2 ·
  33. vLLM · Q3_K_L Decode 70.71 · Prefill 337.65 ·
  34. vLLM · Q4_K_M Decode 62.62 · Prefill 304.49 ·
  35. Ollama · Q3_K_L Decode 58.9 · Prefill 275.23 ·
  36. Ollama · Q8_0 Decode 35.6 · Prefill 173.17 ·
  37. llama.cpp · Q8_0 Decode 18.5 · Prefill 2.2 ·

Fit check

Physics-based estimate from VRAM and memory bandwidth: pick a model to see which quantizations fit and the expected speed

VRAM is not recorded for this hardware; fit cannot be computed

Formula: tok/s = (bandwidth ÷ model size) × 0.55, adapted from llmfit; see run records for measured numbers

Key specs

Vendor
Qualcomm
Memory bandwidth
136 GB/s

Run records

37 configurations

ModelFrameworkQuantizationDecodeGen timeSamples
Qwen3.5-4Bllama.cppQ2_K83.9 tok/s—1
Qwen3.5-4BvLLMFP1632.8 tok/s—1
Qwen3.5-9BStrataQ4_K_S60 tok/s—1
Qwen3.5-0.8Bllama.cppQ4_K_M112.4 tok/s—1
Qwen3.5-0.8BOllamaIQ4_XS117.6 tok/s—1
Qwen3.5-0.8BOllamaQ8_069.8 tok/s—1
Qwen3.8-9Bllama.cppQ3_K_L50.6 tok/s—1
Qwen3.8-9BvLLMQ3_K_L51.1 tok/s—1
Qwen3-8BvLLMQ2_K69 tok/s—1
Qwen3-8BvLLMQ3_K_S55.6 tok/s—1
Qwen3-8BStrataIQ2_M86.1 tok/s—1
Qwen3-8BStrataQ3_K_M73.8 tok/s—1
Qwen3-4BvLLMIQ4_XS66.1 tok/s—1
Qwen3-4BvLLMQ3_K_M74.2 tok/s—1
Qwen3-4BStrataFP1639.4 tok/s—1
Qwen3-1.7BvLLMIQ4_XS100.9 tok/s—1
Qwen3-1.7BOllamaQ3_K_M88.3 tok/s—1
Qwen3-4B-Thinking-2507llama.cppQ2_K81.6 tok/s—1
Qwen3-4B-Thinking-2507StrataQ8_048.7 tok/s—1
Qwen3-Coder-Nextllama.cppQ3_K_L49 tok/s—1
Qwen3-Coder-NextOllamaIQ2_M54.2 tok/s—1
Qwen2.5-0.5B-InstructOllamaIQ4_XS131.7 tok/s—1
Qwen2.5-0.5B-InstructStrataFP1696.3 tok/s—1
Qwen2.5-Omni-3BvLLMIQ2_XS103.6 tok/s—1
Qwen2.5-Omni-3BvLLMQ5_K_S65.7 tok/s—1
Qwen2.5-Omni-3BOllamaQ2_K89.6 tok/s—1
Qwen2.5-Omni-3BOllamaQ3_K_S76 tok/s—1
Qwen3-Embedding-0.6BvLLMQ4_K_S152.6 tok/s—1
Qwen3-Embedding-0.6BStrataQ6_K148.3 tok/s—1
gemma-4-E2B-itOllamaQ4_K_M76.5 tok/s—1
gemma-4-E2B-itStrataQ2_K154.6 tok/s—1
gemma-4-E2B-itStrataQ3_K_L121.8 tok/s—1
gemma-4-E4B-itvLLMQ3_K_L70.7 tok/s—1
gemma-4-E4B-itvLLMQ4_K_M62.6 tok/s—1
gemma-4-E4B-itOllamaQ3_K_L58.9 tok/s—1
gemma-4-E4B-itOllamaQ8_035.6 tok/s—1
Qwen3-4B-Thinking-2507llama.cppQ8_018.5 tok/s—1