GPUs & Macs › AMD Radeon RX 6900 XT

AMD Radeon RX 6900 XT: local LLM speed

Llama 2 7B (Q4_0) generates 108 tokens/s on the AMD Radeon RX 6900 XT (Vulkan), and processes prompts at 1,901 tokens/s.

Measured results

ModelGenerate
tok/s
Prompt
tok/s
QuantBackendTestSource
Llama 2 7B1081,901Q4_0Vulkan FA offtg128llama.cpp #10879 details
Llama 2 7B1061,762Q4_0Vulkan FA ontg128llama.cpp #10879 details
Llama 2 7B88.51,890Q4_0ROCm FA offtg128llama.cpp #15021 details
Llama 2 7B85.01,948Q4_0ROCm FA ontg128llama.cpp #15021 details
Llama 3.2 1B Instruct62.84,411Q4_K_Mllamafile LocalScore avgLocalScore details
Llama 3.1 8B Instruct42.41,082Q4_K_Mllamafile LocalScore avgLocalScore details
Qwen2.5 14B Instruct18.5569Q4_K_Mllamafile LocalScore avgLocalScore details

What fits on the AMD Radeon RX 6900 XT

Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 108 tok/s (Vulkan, source). Fit assumes the whole model in GPU memory with ~8K context.

ModelBest quant that fitsFileExpected tok/s
Qwen3.8-27BUD-IQ3_XXS10.9 GB37.8 estimate
gemma-4-12B-itQ8_012.7 GB32.6 estimate
Qwen3.5-9BQ8_09.5 GB43.3 estimate
Qwen3.5-4BQ8_04.5 GB92.1 estimate
Qwen3.6-35B-A3BUD-IQ3_XXS13.2 GB182 estimate
gemma-4-26B-A4B-itUD-IQ4_XS13.6 GB98.7 estimate
MiMo-V2.6-Distill-Qwen-9BQ8_09.5 GB43.2 estimate
gemma-4-E4B-itQ8_08.2 GB50.4 estimate
Muse-Glimmer-30BUD-IQ3_XXS13.1 GB31.4 estimate
Qwen3-4BQ8_04.3 GB96.4 estimate
gemma-4-31B-itUD-IQ3_XXS11.8 GB34.9 estimate
gemma-4-12B-it-qat-q4_0-unquantizedUD-Q4_K_XL6.7 GB61.5 estimate
gemma-3-1b-itQ8_01.1 GB344 estimate
gemma-4-E2B-itQ8_05.0 GB83.1 estimate

Open in the picker

Similar hardware

AMD Radeon RX 9070 XT · AMD Radeon RX 9070 · AMD Radeon RX 7800 XT · AMD Radeon RX 7900 GRE · AMD Radeon RX 6800 · AMD Radeon VII · AMD Radeon RX 6800 XT · AMD Radeon RX 9060 XT