Qwen3-32B on AMD GPUs
7 results · 3 sources · 16 measurements
Grouped by reported model identity. Exact weights, quantizations and software can differ. How results are reviewed
Documented results
7 selected tests | | | | | | |
|---|
View
ReportLocalMaxxing | Qwen3-32BQ4_K_MModel ref. ↗ | Token speed | — | 57.95 tok/sPrefill · unknown | 3.22 tok/sGeneration · unknown | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration- Memory
- 32 GB VRAM/GPU
- OS
- Windows
- Engine
- llama.cpp
- Backend
- cuda
- Harness
- LocalMaxxing community speed test; exact client where reported
- Prompt / output
- 512 / 128 tokens
- Context
- 2048
- Concurrency
- Not reported
- Date
- 2026-06-23
Full record → |
|---|
View
ReportGeneric LLM | Qwen3-32BQuant not reportedModel ref. ↗ | Token speed | 30.21 tok/sThroughput · unknown | — | — | 1 × RX 7900 XTXSingle GPU · 24 GB VRAM/GPUConfiguration- Memory
- 24 GB VRAM/GPU
- OS
- Not reported
- Engine
- Ollama
- Backend
- Not reported
- Harness
- Not reported
- Prompt / output
- ? / ? tokens
- Context
- Not reported
- Concurrency
- Not reported
- Date
- Not reported
Full record → |
|---|
View
ReportGeneric LLM | Qwen3-32BQ6Model ref. ↗ | Relative performance | 4.47 x vs RTX 5080Reported speedup · Per request | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration- Memory
- 32 GB VRAM/GPU
- OS
- Windows 11 Pro 24H2
- Engine
- LM Studio / llama.cpp
- Backend
- Vulkan
- Harness
- Not reported
- Prompt / output
- ? / ? tokens
- Context
- Not reported
- Concurrency
- Not reported
- Date
- Not reported
Full record → |
|---|
View
ReportLocalMaxxing | Qwen3-32BFP8Model ref. ↗ | Token speed | 79.31 tok/sThroughput · peak · aggregate | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration- Memory
- 32 GB VRAM/GPU
- OS
- Arch Linux
- Engine
- vllm
- Backend
- rocm
- Harness
- LocalMaxxing community speed test; exact client where reported
- Prompt / output
- 534 / 256 tokens
- Context
- 32768
- Concurrency
- Not reported
- Date
- 2026-05-05
Full record → |
|---|
View
ReportGeneric LLM | Qwen3-32BQ6Model ref. ↗ | Relative performance | 4.96 x vs RTX 5080Reported speedup · Per request | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration- Memory
- 32 GB VRAM/GPU
- OS
- Windows 11 Pro 24H2
- Engine
- LM Studio / llama.cpp
- Backend
- Vulkan
- Harness
- Not reported
- Prompt / output
- ? / ? tokens
- Context
- Not reported
- Concurrency
- Not reported
- Date
- Not reported
Full record → |
|---|
View
ReportLocalMaxxing | Qwen3-32BFP8Model ref. ↗ | Token speed | — | — | 22.87 tok/sGeneration · unknown | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration- Memory
- 32 GB VRAM/GPU
- OS
- Arch Linux
- Engine
- vllm
- Backend
- rocm
- Harness
- LocalMaxxing community speed test; exact client where reported
- Prompt / output
- 534 / 256 tokens
- Context
- 32768
- Concurrency
- Not reported
- Date
- 2026-05-05
Full record → |
|---|
View
ReportLocalMaxxing | Qwen3-32BFP8Model ref. ↗Conflicting evidence | Token speed | — | — | 22.76 tok/sGeneration · unknown | GPU count unknownTensor parallel · 32 GB VRAM/GPUConfiguration- Memory
- 32 GB VRAM/GPU
- OS
- Arch Linux
- Engine
- vllm
- Backend
- rocm
- Harness
- LocalMaxxing community speed test; exact client where reported
- Prompt / output
- 530 / 256 tokens
- Context
- 786
- Concurrency
- Not reported
- Date
- 2026-05-06
Full record → |
|---|
Quantizations and software
Quantizations: FP8, Q4_K_M, Q6
Engines: llama.cpp, LM Studio / llama.cpp, Ollama, vllm
This is a selection across sources, not a ranking or a matched experiment. Reading benchmark reports
GPU benchmark pages: R9700 · RX 7900 XTX