AMD Radeon RX 9070 GRE

GPU

AMD RDNA 4 (a cut-down Navi 48): 12 GB GDDR6 on a 192-bit bus, 432 GB/s of memory bandwidth, 220 W board power. Launched in China on 2025-05-08 and worldwide on 2026-06-02. Its bandwidth is roughly a third below the RX 9070's, which is the binding constraint.

14

Run records

10

Models covered

4

Frameworks covered

L2 Cross-framework verified × 14

Evidence mix

Typical measured performance

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

  1. vLLM · IQ1_S Decode 165.6 · Prefill 729.98 ·
  2. Strata · IQ1_S Decode 212.06 · Prefill 968.79 ·
  3. llama.cpp · IQ2_M Decode 264.62 · Prefill 925.29 ·
  4. Strata · IQ1_S Decode 369.89 · Prefill 1215.87 ·
  5. Ollama · Q2_K Decode 89.93 · Prefill 344.74 ·
  6. Strata · Q3_K_S Decode 124.72 · Prefill 470.9 ·
  7. llama.cpp · Q2_K Decode 124.6 · Prefill 437.09 ·
  8. Ollama · Q8_0 Decode 79.82 · Prefill 368.94 ·
  9. llama.cpp · Q3_K_M Decode 100.99 · Prefill 537.73 ·
  10. llama.cpp · Q5_K_S Decode 85.62 · Prefill 376.76 ·
  11. Ollama · IQ3_XS Decode 244.89 · Prefill 1100.19 ·
  12. Strata · Q5_K_M Decode 128.98 · Prefill 579.29 ·
  13. llama.cpp · IQ3_S Decode 241.94 · Prefill 776.44 ·
  14. llama.cpp · Q2_K Decode 298.41 · Prefill 1440.77 ·

Fit check

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

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

Key specs

Vendor
AMD
Architecture
RDNA 4
VRAM
12 GB
Memory bandwidth
432 GB/s
TDP
220 W
Release date
2025-05-08

Run records

14 configurations

ModelFrameworkQuantizationDecodeGen timeSamples
Qwen3.5-4BvLLMIQ1_S165.6 tok/s—1
Qwen3.5-4BStrataIQ1_S212.1 tok/s—1
Qwen3.5-0.8Bllama.cppIQ2_M264.6 tok/s—1
Qwen3.5-0.8BStrataIQ1_S369.9 tok/s—1
Qwen3.8-9BOllamaQ2_K89.9 tok/s—1
Qwen3-8BStrataQ3_K_S124.7 tok/s—1
Qwen3-4Bllama.cppQ2_K124.6 tok/s—1
Qwen3-1.7BOllamaQ8_079.8 tok/s—1
Qwen3-4B-Thinking-2507llama.cppQ3_K_M101 tok/s—1
Qwen3-4B-Thinking-2507llama.cppQ5_K_S85.6 tok/s—1
Qwen2.5-0.5B-InstructOllamaIQ3_XS244.9 tok/s—1
Qwen2.5-Omni-3BStrataQ5_K_M129 tok/s—1
Qwen3-Embedding-0.6Bllama.cppIQ3_S241.9 tok/s—1
Qwen3-Embedding-0.6Bllama.cppQ2_K298.4 tok/s—1