AMD · consumer

Radeon RX 6750 GRE

Radeon RX 6750 GRE has 12 GB of VRAM at 384 GB/s — about 11.16 GiB usable after driver and compositor overhead. 1599 of 2118 indexed models fit at 64K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
12 GB
GDDR6
Bandwidth
384 GB/s
192-bit bus
Tensor FP16
—
dense
TDP
230 W
$289 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1362vision language 135image 2audio tts 21audio asr 39video 14embedding 26

What fits at 64K context

largest quantization that fits, per model · 1599 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
EVA-abliterated-TIES-Qwen2.5-14BI1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Neuron-V1-14B-InstructI1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Ektome-Qwen2.5-Coder-14B-Instruct-PristinelyUncensoredI1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Qwen2.5-14B-Instruct-1M-abliteratedI1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
DeepCoder-14B-PreviewQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Deepseeker-Kunou-Qwen2.5-14bI1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
SuperNova-MediusQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
14B-Qwen2.5-Kunou-v1I1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Sugoi-14B-Ultra-HFI1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Qwen2.5-14B-Instruct-abliterated-v2Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Qwen2.5-14B-Instruct-UncensoredQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Qwen2.5-Coder-14B-Instruct-abliteratedQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
OpenCodeReasoning-Nemotron-14BQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
DeepSeek-R1-Distill-Qwen-14B-abliterated-v2I1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
C1-TachuI1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
DeepSeek-R1-Distill-Qwen-14B-abliteratedI1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
0x-liteQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Qwen2.5-Coder-14B-InstructQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Tessera-4I1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
AceReason-Nemotron-14BQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Qwen2.5-14B-InstructQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
FinetunedQwen14BQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Tessera-4.1I1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Qwen2.5-14B-Instruct-1MQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Qwen2.5-Coder-14BQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
DeepSeek-R1-Distill-Qwen-14BQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
UwU-14B-Math-v0.2I1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
EVA-Qwen2.5-14B-v0.2I1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
EVA-Qwen2.5-14B-v0.0I1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
EVA-Qwen2.5-14B-v0.1I1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
oxy-1-smallQ3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Impish_QWEN_14B-1MI1-Q3_K_M14.8B6.84 GiB3.38 GiB11.16 GiB0.00 GiB24±26.5%
Qwen3-16B-A3BMoEIQ4_NL16.0B8.58 GiB1.69 GiB11.16 GiB0.00 GiB42±37%
Qwen2.5-14BQ3_K_M14.8B6.83 GiB3.38 GiB11.15 GiB0.01 GiB24±26.5%
Lamarck-14B-v0.7I1-Q3_K_M14.8B6.83 GiB3.38 GiB11.15 GiB0.01 GiB24±26.5%
QwenStock-14BI1-Q3_K_M14.8B6.83 GiB3.38 GiB11.15 GiB0.01 GiB24±26.5%
DeepSeek-R1-Distill-Qwen-14B-UncensoredI1-Q3_K_M14.8B6.83 GiB3.38 GiB11.15 GiB0.01 GiB24±26.5%
Assistant_Pepe_8BQ8_0—7.96 GiB2.25 GiB11.15 GiB0.01 GiB24±26.5%
gemma-4-19B-A4B-it-INSTRUCT-Heretic-UncensoredMoEI1-Q3_K_L19.0B9.48 GiB0.79 GiB11.15 GiB0.01 GiB24±26.5%
gemma-4-19B-A4B-it-The-DECKARD-Heretic-Uncensored-ThinkingMoEI1-Q3_K_L19.0B9.48 GiB0.79 GiB11.15 GiB0.01 GiB24±26.5%
gemma-4-19b-a4b-it-REAP-hereticMoEI1-Q3_K_L19.0B9.48 GiB0.79 GiB11.15 GiB0.01 GiB24±26.5%
Gemma-4-19BMoEI1-Q3_K_L19.0B9.48 GiB0.79 GiB11.15 GiB0.01 GiB24±26.5%
Aya-Medikal-V2Q8_08.0B7.95 GiB2.25 GiB11.15 GiB0.01 GiB24±26.5%
Qwen3-Coder-REAP-25B-A3BMoEQ2_K24.9B8.57 GiB1.69 GiB11.15 GiB0.01 GiB47±37%
Qwen3-VL-8B-Instruct-HereticI1-Q3_K_M8.8B7.68 GiB2.53 GiB11.15 GiB0.01 GiB24±26.5%
ZAYA1-8B-CoderMoEQ8_08.8B8.83 GiB1.41 GiB11.15 GiB0.01 GiB24±26.5%
Foundation-Sec-8B-Instruct-hereticQ8_08.0B7.96 GiB2.25 GiB11.15 GiB0.01 GiB24±26.5%
Foundation-Sec-8B-InstructQ8_08.0B7.96 GiB2.25 GiB11.15 GiB0.01 GiB24±26.5%
dolphin-2.9-llama3-8bQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
saiga_llama3_8bQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
llama-3-8b-Instruct-bnb-4bitQ8_08.2B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
Meta-Llama-3-8B-InstructQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
MiniCPM-Llama3-V-2_5Q8_08.5B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
Llama3-ChatQA-1.5-8BQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
Meta-Llama-3-8BQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
Foundation-Sec-8B-ReasoningQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
Llama-3.1-Tulu-3-8BQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
Llama-3-Groq-8B-Tool-UseQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
llama3.1-heretic-unsensoredQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 GiB24±26.5%
Dolphin3.0-Llama3.1-8BQ8_08.0B7.95 GiB2.25 GiB11.14 GiB0.02 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 generation3.37 it/s2.97–3.4528
Benchmarked· n=28

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?
1599 of 2118 indexed open-weight models fit a Radeon RX 6750 GRE at 65,536 context with q4_0 KV cache, the largest being EVA-abliterated-TIES-Qwen2.5-14B at I1-Q3_K_M. 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 12 GB, but about 11.16 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 384 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.