RX 9070 XT LLM benchmarks
530 results · 932 measurements · 21 sources
Device specifications
| Variant | Memory | Memory type | Bus | Bandwidth | FP16 matrix | INT8 matrix |
|---|---|---|---|---|---|---|
| RX 9070 XT ↗ | 16 GB | GDDR6 | 256-bit | 640 GB/s | 195 TFLOPS | 389 TOPS |
Manufacturer specifications. Shared system RAM is not dedicated VRAM. Unreported values remain —; measured LLM results retain their reported configuration.
Selected reported tests
24 of 530 resultsSource-reported results, selected across reports by documentation quality. Match configurations before comparing values.
| View ReportCustom evaluation | Qwen3.5-27B | Local LLM custom suite4b · Math / math_basic | 100 %Score · 1 / 1 · Custom answer parser 6.42 tok/sThroughput · per request | — | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
|---|---|---|---|---|---|---|
| View ReportGeneric LLM | Qwen3.8-27B | LongBench v225 short/medium problems, seed42; mixed easy/hard1 concurrent request | 56 %Score · n = 100 | — | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3-14B | Custom C++ corpus perplexity~960KB C++ corpus2K configured context | 1.7031 perplexityScore · Perplexity · Lower is better | — | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View ReportCoding session | Qwen3.6-27B | Token speed | 17.14 tok/sThroughput · per request | 1,592.45 tok/sPrefill · per request | 23.5 tok/sGeneration · per request | 2 × RX 9070 XTLayer split · 16 GB VRAM/GPUConfiguration
|
| View ReportCoding session | Qwen3.8-27B | Token speed | 14.2 tok/sThroughput · per request | 2,061.11 tok/sPrefill · per request | 24.7 tok/sGeneration · per request | 2 × RX 9070 XTLayer split · 16 GB VRAM/GPUConfiguration
|
| View ReportLocalMaxxing | gemma-4-12B-it | Token speed | — | 955.8 tok/sPrefill · median · unknown | 110.7 tok/sGeneration · median · unknown | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | Llama-3.2-3B-Instruct | Token speed | — | 7,533.54 tok/sPrefill · mean · per request | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | Llama-2-7B | Token speed | — | 4,343.73 tok/sPrefill · mean · per request | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | Qwen3-14B | Token speed | — | — | 8.1 tok/sGeneration · per request | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | Qwen3-0.6B | Token speed | — | 18,116 tok/sPrefill · per request | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | Laguna XS.2 33B-A3B | Token speed | — | 2,548.1 tok/sPrefill · per request | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | Qwen3.8-27B | Token speed | — | 1,268.76 tok/sPrefill · per request | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | DeepSeek-R1-Distill-Llama-8B | Token speed | — | — | 71.31 tok/sGeneration · mean · per request | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | deepseek2 16B Q4_K (Lite) (checkpoint unspecified) | Token speed | — | 5,887.59 tok/sPrefill · per request | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | Gemma 4-26B-A4B | Token speed | — | 1,949.9 tok/sPrefill · per request | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Llama-2-13b-chat | Procyon | — | — | 61.13 tok/sGeneration · per request | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | Qwen3.6-35B-A3B | Token speed2,048 input tokens · 0 output tokens · 48,000 existing KV tokens | — | 1,875.05 tok/sPrefill · mean · per request | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Qwen3.6-35B-A3B | Token speed | — | 1,055 tok/sPrefill · per request | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | gpt-oss-20b | Token speed | 179.24 tok/sThroughput · unknown | — | — | GPU count unknownUnknown · 16 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Qwen3-14B | Token speed | 53.73 tok/sThroughput · unknown | — | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | MLPerf Client v1.0 (selected models unreported) | Token speed1.0 · MLPerf Client | — | — | 117 tok/sGeneration · per request | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.5-27B | Local LLM custom suite4b · Math / math_discount | 100 %Score · 1 / 1 · Custom answer parser 6.4 tok/sThroughput · per request | — | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Qwen3.8-27B | GSM8K30 problems with199.7K resident context1 concurrent request | 96.7 %Reported accuracy · n = 100 | — | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3-14B | Custom C++ corpus perplexity~960KB C++ corpus2K configured context | 1.7044 perplexityScore · Perplexity · Lower is better | — | — | 1 × RX 9070 XTSingle GPU · 16 GB VRAM/GPUConfiguration
|
Models and software in this collection
Model families: DeepSeek R1, GPT-OSS, Gemma 4, Granite 3.0, Laguna, Llama 2, Llama 3.1, Llama 3.2, Mistral, Phi 3.5, Qwen3, Qwen3.5, Qwen3.6, Qwen3.8
Engines: Custom HIP LLM runner, LLAMA-DX / llama.cpp fork, ONNX Runtime, Ollama, custom, hipfire, llama.cpp, llama.cpp RDNA lab fork, llama.cpp TurboQuant fork
Methodology and evidence standards · Download the research dataset