AMD · consumer

Radeon RX 6750 GRE

Radeon RX 6750 GRE has 10 GB of VRAM at 320 GB/s — about 9.30 GiB usable after driver and compositor overhead. 797 of 2118 indexed models fit at 64K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
10 GB
GDDR6
Bandwidth
320 GB/s
160-bit bus
Tensor FP16
—
dense
TDP
170 W
$269 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 622embedding 21vision language 86audio tts 19audio asr 36video 12image 1

What fits at 64K context

largest quantization that fits, per model · 797 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Llama-3.2-3B-Instruct-abliteratedI1-Q2_K3.6B1.39 GiB7.00 GiB9.30 GiB0.00 GiB24±26.5%
Llama-3.2-3B-Instruct-uncensoredQ2_K3.6B1.39 GiB7.00 GiB9.30 GiB0.00 GiB24±26.5%
nomic-embed-codeQ5_K_L7.1B4.84 GiB3.50 GiB9.30 GiB0.00 GiB24±26.5%
AfriqueGemma-12BI1-IQ2_XS12.2B3.88 GiB4.47 GiB9.30 GiB0.00 GiB24±26.5%
granite-20b-code-instruct-8kIQ3_S20.1B8.32 GiB0.00 GiB9.30 GiB0.00 GiB24±26.5%
granite-20b-code-base-8kI1-IQ3_S20.1B8.32 GiB0.00 GiB9.30 GiB0.00 GiB24±26.5%
DeepSeek-Coder-V2-Lite-BaseMoEI1-IQ3_XXS15.7B6.49 GiB1.90 GiB9.29 GiB0.01 GiB39±37%
DeepSeek-Coder-V2-Lite-InstructMoEIQ3_XXS15.7B6.49 GiB1.90 GiB9.29 GiB0.01 GiB39±37%
DeepSeek-V2-Lite-ChatMoEIQ3_XXS15.7B6.49 GiB1.90 GiB9.29 GiB0.01 GiB39±37%
Qwen3.6-35B-A3B-REAM-160-ru-agentMoEIQ2_S23.6B7.14 GiB1.25 GiB9.29 GiB0.01 GiB54±37%
gemma-4-E4B-uncensoredQ8_07.9B7.43 GiB0.94 GiB9.28 GiB0.02 GiB24±26.5%
gemma-4-E4B-it-qat-heretic_decensoredQ8_07.9B7.43 GiB0.94 GiB9.28 GiB0.02 GiB24±26.5%
gemma-4-E4B-it-QAT-SOMPOA-heresyQ8_07.9B7.43 GiB0.94 GiB9.28 GiB0.02 GiB24±26.5%
Tinman-gemma4-companion-mergedQ8_07.9B7.43 GiB0.94 GiB9.28 GiB0.02 GiB24±26.5%
Qwen3-VL-2B-InstructQ6_K_L2.1B1.39 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Lightning-1.7BQ6_K_L1.7B1.39 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Dolphin3.0-Llama3.2-3BIQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Llama-Doctor-3.2-3B-InstructI1-IQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Llama-Song-Stream-3B-InstructIQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Llama-3.2-3B-Instruct-roleplay-tunedI1-IQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Llama-3.2-3B-Instruct-heretic-ablitered-uncensoredI1-IQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
llama-3.2-Korean-Bllossom-3BIQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Llama3.2-3B-creative-writer-v0.1I1-IQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Firefly-V3.2I1-IQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Firefly-V3I1-IQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Hermes-3-Llama-3.2-3BIQ3_XS3.2B1.38 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Qwen3.6-12B-IQ-Ultra-Heretic-Uncensored-Thinking-V2-HightopQ4_K_M12.1B6.82 GiB1.50 GiB9.28 GiB0.02 GiB24±26.5%
Qwen3-0.6BBF16752M1.41 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
Qwen3-0.6B-Distilled-30B-A3B-Thinking-SFTF16752M1.41 GiB7.00 GiB9.28 GiB0.02 GiB24±26.5%
internlm3-8b-instructQ4_K_L8.8B5.35 GiB3.00 GiB9.28 GiB0.02 GiB24±26.5%
DeepScaleR-1.5B-PreviewF321.8B6.63 GiB1.75 GiB9.28 GiB0.02 GiB24±26.5%
DeepSeek-R1-Distill-Qwen-1.5BF321.8B6.63 GiB1.75 GiB9.28 GiB0.02 GiB24±26.5%
granite-vision-4.1-4bQ8_04.0B3.37 GiB5.00 GiB9.27 GiB0.03 GiB24±26.5%
Parable-Granite-4.1-3B-Claude-Fable-5Q8_03.4B3.37 GiB5.00 GiB9.27 GiB0.03 GiB24±26.5%
granite-4.0-microQ8_03.4B3.37 GiB5.00 GiB9.27 GiB0.03 GiB24±26.5%
granite-4.1-3bQ8_03.4B3.37 GiB5.00 GiB9.27 GiB0.03 GiB24±26.5%
granite-4.0-micro-baseQ8_03.4B3.37 GiB5.00 GiB9.27 GiB0.03 GiB24±26.5%
OpenClaude-1.7B-MergedQ6_K1.7B1.38 GiB7.00 GiB9.27 GiB0.03 GiB24±26.5%
Qwythos-9B-v2Q4_K_L9.7B6.34 GiB2.00 GiB9.27 GiB0.03 GiB24±26.5%
Tess-4-9BQ4_K_L9.7B6.34 GiB2.00 GiB9.27 GiB0.03 GiB24±26.5%
Qwen3.5-9B-BaseQ5_19.7B6.33 GiB2.00 GiB9.27 GiB0.03 GiB24±26.5%
Teuken-7B-instruct-research-v0.4Q6_K_L7.5B6.33 GiB2.00 GiB9.27 GiB0.03 GiB24±26.5%
gemma-2-2b-it-abliteratedF162.6B4.88 GiB3.48 GiB9.26 GiB0.04 GiB24±26.5%
Vikhr-Gemma-2B-instructBF162.6B4.88 GiB3.48 GiB9.26 GiB0.04 GiB24±26.5%
Qwen3-1.7BQ5_K_M2.0B1.37 GiB7.00 GiB9.26 GiB0.04 GiB24±26.5%
DualMinded-Qwen3-1.7BQ5_K_M2.0B1.37 GiB7.00 GiB9.26 GiB0.04 GiB24±26.5%
DualMindQ5_K_M2.0B1.37 GiB7.00 GiB9.26 GiB0.04 GiB24±26.5%
Qwen3-1.7B-Coder-Distilled-SFTQ5_K_M2.0B1.37 GiB7.00 GiB9.26 GiB0.04 GiB24±26.5%
Qwen3-1.7B-Distilled-30B-A3B-SFTQ5_K_M2.0B1.37 GiB7.00 GiB9.26 GiB0.04 GiB24±26.5%
DistilQwen3-1.7B-uncensoredQ5_K_M2.0B1.37 GiB7.00 GiB9.26 GiB0.04 GiB24±26.5%
Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-IQ1_S21.3B5.30 GiB3.00 GiB9.26 GiB0.04 GiB24±26.5%
Qwen3.6-21B-IQ-Ultra-Heretic-Uncensored-ThinkingI1-IQ1_S21.3B5.30 GiB3.00 GiB9.26 GiB0.04 GiB24±26.5%
Amaretto-3BIQ4_XS4.3B1.84 GiB6.50 GiB9.25 GiB0.05 GiB24±26.5%
orpheus-3b-0.1-ftQ2_K3.8B1.34 GiB7.00 GiB9.25 GiB0.05 GiB24±26.5%
Felldude-Uncensored-Ministral3-3B-bf16I1-IQ4_XS3.8B1.82 GiB6.50 GiB9.24 GiB0.06 GiB24±26.5%
Ministral-3-3B-Instruct-2512-BF16IQ4_XS4.3B1.82 GiB6.50 GiB9.24 GiB0.06 GiB24±26.5%
Ministral-3-3B-Instruct-2512IQ4_XS3.8B1.82 GiB6.50 GiB9.24 GiB0.06 GiB24±26.5%
Ministral-3-3B-Reasoning-2512IQ4_XS4.3B1.82 GiB6.50 GiB9.24 GiB0.06 GiB24±26.5%
EVA-Yi-1.5-9B-32K-V1I1-IQ2_XXS8.8B2.29 GiB6.00 GiB9.22 GiB0.08 GiB24±26.5%
Supertron2-Reranker-2BI1-Q6_K2.1B1.32 GiB7.00 GiB9.21 GiB0.09 GiB24±26.5%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Measured on this card

third-party benchmarks, aggregated
Workload◍MedianMiddle 50%Runs
Image generation2.78 it/s2.77–2.925
Benchmarked· n=5

Aggregated from community-submitted runs, so the spread is wide by nature — it covers different models, resolutions, step counts and settings, not one controlled configuration. Read the middle 50% rather than the median alone. These figures are reproduced with attribution from vladmandic-sd-data-benchmark, which publishes no licence — so we display and link rather than redistribute them.

Questions people ask

What AI models can a Radeon RX 6750 GRE run?
797 of 2118 indexed open-weight models fit a Radeon RX 6750 GRE at 65,536 context with f16 KV cache, the largest being Llama-3.2-3B-Instruct-abliterated at I1-Q2_K. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon RX 6750 GRE actually have?
Its nameplate is 10 GB, but about 9.30 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Radeon RX 6750 GRE fast for local AI?
Its memory bandwidth is 320 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.