NVIDIA · consumer

GeForce RTX 3050 OEM

GeForce RTX 3050 OEM has 8 GB of VRAM at 224 GB/s — about 7.44 GiB usable after driver and compositor overhead. 1212 of 2118 indexed models fit at 32K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
8 GB
GDDR6
Bandwidth
224 GB/s
128-bit bus
Tensor FP16
32 TF
dense
TDP
130 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1023embedding 26vision language 96video 8audio asr 38image 1audio tts 20

What fits at 32K context

largest quantization that fits, per model · 1212 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.6-14B-A3B-FableVibesMoEQ3_K_M13.8B6.30 GiB0.33 GiB7.44 GiB0.00 GiB72±37%
Qwen3.6-14B-A3B-VibeForged-v2MoEQ3_K_M13.8B6.30 GiB0.33 GiB7.44 GiB0.00 GiB72±37%
OmniAtlas-Qwen3-30B-A3BI1-IQ1_M31.7B6.59 GiB0.00 GiB7.44 GiB0.00 GiB25±12.9%
Qwen3-Omni-30B-A3B-CaptionerI1-IQ1_M31.7B6.59 GiB0.00 GiB7.44 GiB0.00 GiB25±12.9%
Hy-MT2-7BQ4_18.0B4.47 GiB2.13 GiB7.44 GiB0.00 GiB24±12.9%
Hunyuan-7B-InstructQ4_17.5B4.47 GiB2.13 GiB7.43 GiB0.01 GiB25±12.9%
DeepSeek-R1-Distill-Llama-8B-AbliteratedI1-IQ2_XXS8.0B4.47 GiB2.13 GiB7.43 GiB0.01 GiB25±12.9%
Nemotron-3-Embed-8B-BF16Q4_K_S8.0B4.33 GiB2.26 GiB7.43 GiB0.01 GiB25±12.9%
GLM-4.7-Flash-REAP-23B-A3B-absolute-heresyMoEI1-IQ2_XXS23.0B5.74 GiB0.88 GiB7.43 GiB0.01 GiB54±37%
Qwen3-VL-8B-Instruct-HereticI1-IQ1_M8.8B4.20 GiB2.39 GiB7.43 GiB0.01 GiB25±12.9%
gemma-3-12b-it-vl-Gemini-3-Pro-Preview-Heretic-Uncensored-ThinkingI1-IQ3_M12.2B5.27 GiB1.31 GiB7.42 GiB0.02 GiB25±12.9%
gemma-3-12b-it-vl-Deepseek-v3.1-Heretic-Uncensored-ThinkingI1-IQ3_M12.2B5.27 GiB1.31 GiB7.42 GiB0.02 GiB25±12.9%
gemma-3-12b-it-vl-GLM-4.7-Flash-Heretic-Uncensored-ThinkingI1-IQ3_M12.2B5.27 GiB1.31 GiB7.42 GiB0.02 GiB25±12.9%
gemma-3-12b-it-ultra-uncensored-hereticIQ3_M12.2B5.27 GiB1.31 GiB7.42 GiB0.02 GiB25±12.9%
Floppa-12B-Gemma3-UncensoredI1-IQ3_M12.2B5.27 GiB1.31 GiB7.42 GiB0.02 GiB25±12.9%
gemma-3-12b-it-hereticI1-IQ3_M12.2B5.27 GiB1.31 GiB7.42 GiB0.02 GiB25±12.9%
gemma-3-12b-it-abliteratedIQ3_M12.2B5.27 GiB1.31 GiB7.42 GiB0.02 GiB25±12.9%
glm-4v-9bQ5_K_M13.9B6.57 GiB0.00 GiB7.41 GiB0.03 GiB25±12.9%
Hubble-4B-v1Q8_04.5B4.47 GiB2.13 GiB7.41 GiB0.03 GiB24±12.9%
Aura-4BQ8_04.5B4.47 GiB2.13 GiB7.41 GiB0.03 GiB24±12.9%
magnum-v2-4bQ8_04.5B4.47 GiB2.13 GiB7.41 GiB0.03 GiB24±12.9%
Impish_LLAMA_4BQ8_04.5B4.47 GiB2.13 GiB7.41 GiB0.03 GiB24±12.9%
Llama-3.1-Minitron-4B-Width-BaseQ8_04.5B4.47 GiB2.13 GiB7.41 GiB0.03 GiB24±12.9%
Tiger-Gemma-12B-v3IQ3_XS12.8B5.25 GiB1.31 GiB7.41 GiB0.03 GiB25±12.9%
AfriqueGemma-12BI1-IQ3_XS12.2B5.25 GiB1.31 GiB7.41 GiB0.03 GiB25±12.9%
starcoder2-7bKV unresolvedQ6_K7.2B5.49 GiB1.06 GiB7.41 GiB0.03 GiB25±12.9%
stable-code-3bI1-Q3_K_M2.8B1.30 GiB5.31 GiB7.41 GiB0.03 GiB24±12.9%
rocket-3BQ3_K_M2.8B1.30 GiB5.31 GiB7.41 GiB0.03 GiB24±12.9%
Anubis-Mini-8B-v1Q4_K_S8.0B4.44 GiB2.13 GiB7.41 GiB0.03 GiB25±12.9%
Mistral-NeMo-Minitron-8B-InstructQ3_K_M8.4B3.92 GiB2.66 GiB7.41 GiB0.03 GiB25±12.9%
LFM2-8B-A1BMoEQ6_K_L8.3B6.41 GiB0.20 GiB7.40 GiB0.04 GiB67±37%
Kepler-8B-Instruct-v2Q2_K7.6B5.61 GiB0.93 GiB7.40 GiB0.04 GiB25±12.9%
Marco-Mini-InstructMoEI1-IQ2_XS17.3B4.76 GiB1.86 GiB7.40 GiB0.04 GiB40±37%
Gemma-The-Writer-N-Restless-Quill-10B-UncensoredI1-IQ2_XS10.0B3.07 GiB3.48 GiB7.40 GiB0.04 GiB25±12.9%
Qwen3.5-9B-CoderI1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Qwythos-9B-Claude-Mythos-5-1M-MTPI1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliteratedI1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Qwen3.5-9B-Fable-5-v1I1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Qwythos-9B-v2I1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
PINQWEN-3.5-9B-1M-BF16I1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Openprose-2-FlashI1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Qwen3.5-9B-Nikusui-v1I1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Ornstein-3.5-9B-V1.5I1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Ornith-1.0-9B-heretic-MTPI1-Q5_K_S9.4B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Tess-4-9BI1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
dotwebs-1I1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
liftQ5_09.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
Hemlock-Qwopus3.5-9B-CoderI1-Q5_K_S9.7B6.03 GiB0.53 GiB7.39 GiB0.05 GiB25±12.9%
EVA-abliterated-TIES-Qwen2.5-14BI1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
Neuron-V1-14B-InstructI1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
Ektome-Qwen2.5-Coder-14B-Instruct-PristinelyUncensoredI1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
Qwen2.5-14B-Instruct-1M-abliteratedI1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
Deepseeker-Kunou-Qwen2.5-14bI1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
14B-Qwen2.5-Kunou-v1I1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
Sugoi-14B-Ultra-HFI1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
DeepSeek-R1-Distill-Qwen-14B-abliterated-v2I1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
C1-TachuI1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
DeepSeek-R1-Distill-Qwen-14B-abliteratedI1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
Tessera-4I1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
Tessera-4.1I1-IQ1_S14.8B3.36 GiB3.19 GiB7.39 GiB0.05 GiB25±12.9%
●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.

Questions people ask

What AI models can a GeForce RTX 3050 OEM run?
1212 of 2118 indexed open-weight models fit a GeForce RTX 3050 OEM at 32,768 context with q8_0 KV cache, the largest being Qwen3.6-14B-A3B-FableVibes at Q3_K_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 3050 OEM actually have?
Its nameplate is 8 GB, but about 7.44 GiB is available to a model once driver and compositor overhead is accounted for.
Is a GeForce RTX 3050 OEM fast for local AI?
Its memory bandwidth is 224 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.