NVIDIA · consumer

GeForce RTX 3050

GeForce RTX 3050 has 6 GB of VRAM at 168 GB/s — about 5.58 GiB usable after driver and compositor overhead. 1197 of 2118 indexed models fit at 4K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
6 GB
GDDR6
Bandwidth
168 GB/s
96-bit bus
Tensor FP16
27 TF
dense
TDP
70 W
$179 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1020vision language 90audio asr 38audio tts 19embedding 26image 1video 3

What fits at 4K context

largest quantization that fits, per model · 1197 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
gemma-3-12b-it-vl-Gemini-3-Pro-Preview-Heretic-Uncensored-ThinkingI1-IQ2_M12.2B4.01 GiB0.72 GiB5.58 GiB0.00 GiB26±12.9%
gemma-3-12b-it-vl-Deepseek-v3.1-Heretic-Uncensored-ThinkingI1-IQ2_M12.2B4.01 GiB0.72 GiB5.58 GiB0.00 GiB26±12.9%
gemma-3-12b-it-vl-GLM-4.7-Flash-Heretic-Uncensored-ThinkingI1-IQ2_M12.2B4.01 GiB0.72 GiB5.58 GiB0.00 GiB26±12.9%
Floppa-12B-Gemma3-UncensoredI1-IQ2_M12.2B4.01 GiB0.72 GiB5.58 GiB0.00 GiB26±12.9%
gemma-3-12b-it-hereticI1-IQ2_M12.2B4.01 GiB0.72 GiB5.58 GiB0.00 GiB26±12.9%
gemma-3-12b-it-abliteratedIQ2_M12.2B4.01 GiB0.72 GiB5.58 GiB0.00 GiB26±12.9%
Olmo-3-7B-ThinkQ2_K_L7.3B2.75 GiB2.00 GiB5.58 GiB0.00 GiB25±12.9%
dolphin-2.8-mistral-7b-v02Q4_17.2B4.24 GiB0.50 GiB5.58 GiB0.00 GiB26±12.9%
Kunoichi-DPO-v2-7BKV unresolvedQ4_17.2B4.24 GiB0.50 GiB5.58 GiB0.00 GiB26±12.9%
LFM2-8B-A1BMoEQ4_K_L8.3B4.73 GiB0.05 GiB5.58 GiB0.00 GiB75±37%
legitus-instruct-v1IQ4_XS8.1B4.21 GiB0.50 GiB5.58 GiB0.00 GiB26±12.9%
canary-qwen-2.5bBF162.6B4.73 GiB0.00 GiB5.57 GiB0.01 GiB26±12.9%
EXAONE-Deep-7.8BQ4_K_L7.8B4.73 GiB0.00 GiB5.57 GiB0.01 GiB26±12.9%
EXAONE-3.5-7.8B-InstructQ4_K_L7.8B4.73 GiB0.00 GiB5.57 GiB0.01 GiB26±12.9%
deepseek-coder-6.7b-instructQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
deepseek-coder-6.7b-baseQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
deepseek-coder-6.7B-kexerI1-IQ3_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
Magicoder-S-DS-6.7BI1-IQ3_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
Anubis-Mini-8B-v1IQ4_XS8.0B4.23 GiB0.50 GiB5.57 GiB0.01 GiB26±12.9%
MathCoder2-CodeLlama-7BQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
CodeLlama-7b-instruct-hfQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
CodeLlama-7b-hfQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
WizardLM-7B-UncensoredI1-IQ3_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
Llama-2-7B-32K-InstructI1-IQ3_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
Luna-AI-Llama2-UncensoredI1-IQ3_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
Llama-2-7b-chat-hfQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
Swallow-7b-NVE-instruct-hfI1-IQ3_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
llava-v1.5-7bQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
CodeLlama-7b-python-hfQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
Wizard-Vicuna-7B-UncensoredQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
llama2_7b_chat_uncensoredQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
WizardLM-7B-V1.0-UncensoredQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
Llama-2-7b-hfQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
pygmalion-2-7bQ3_K_S6.7B2.75 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
Qwen3-TTS-12Hz-0.6B-BaseQ4_K_M915M4.72 GiB0.00 GiB5.57 GiB0.01 GiB26±12.9%
VoxCPM2F162.3B4.72 GiB0.00 GiB5.57 GiB0.01 GiB26±12.9%
EVA-Yi-1.5-9B-32K-V1I1-Q3_K_L8.8B4.37 GiB0.38 GiB5.57 GiB0.01 GiB26±12.9%
Yi-Coder-9B-ChatQ3_K_L8.8B4.37 GiB0.38 GiB5.57 GiB0.01 GiB26±12.9%
Yi-1.5-9B-ChatQ3_K_L8.8B4.37 GiB0.38 GiB5.57 GiB0.01 GiB26±12.9%
Olmo-3-7B-InstructUD-IQ3_XXS7.3B2.74 GiB2.00 GiB5.57 GiB0.01 GiB26±12.9%
t5-v1_1-xxlQ2_K4.8B4.72 GiB0.00 GiB5.56 GiB0.02 GiB26±12.9%
MiMo-VL-7B-RLI1-Q4_K_S8.3B4.17 GiB0.56 GiB5.56 GiB0.02 GiB26±12.9%
Kuwutu-7B-CYOA-v2I1-Q4_K_S7.6B4.17 GiB0.56 GiB5.56 GiB0.02 GiB26±12.9%
Aya-Medikal-V2I1-Q3_K_L8.0B4.22 GiB0.50 GiB5.56 GiB0.02 GiB26±12.9%
Bonsai-8B-unpackedQ3_K_L8.2B4.17 GiB0.56 GiB5.56 GiB0.02 GiB26±12.9%
Falcon3-10B-InstructIQ3_XS10.3B4.07 GiB0.63 GiB5.56 GiB0.02 GiB26±12.9%
Falcon3-7B-InstructQ4_K_M7.5B4.26 GiB0.44 GiB5.56 GiB0.02 GiB26±12.9%
Qwen3.6-12B-IQ-Ultra-Heretic-Uncensored-Thinking-V2-HightopQ2_K12.1B4.60 GiB0.09 GiB5.55 GiB0.03 GiB26±12.9%
Huihui-gemma-3n-E4B-it-abliteratedQ5_K_M7.8B4.61 GiB0.11 GiB5.55 GiB0.03 GiB26±12.9%
gemma-3n-E4B-itQ5_K_S7.8B4.61 GiB0.11 GiB5.55 GiB0.03 GiB26±12.9%
SambaLingo-Japanese-ChatI1-IQ3_XS6.9B2.73 GiB2.00 GiB5.55 GiB0.03 GiB26±12.9%
dolphin-2.9.2-Phi-3-MediumKV unresolvedIQ2_XS14.0B3.91 GiB0.78 GiB5.55 GiB0.03 GiB26±12.9%
rnj-1-instructQ3_K_L8.3B4.20 GiB0.50 GiB5.55 GiB0.03 GiB26±12.9%
gemma-4-E2B-itQ8_05.1B4.70 GiB0.05 GiB5.55 GiB0.03 GiB25±12.9%
gemma-4-E2B-itQ8_05.1B4.70 GiB0.05 GiB5.55 GiB0.03 GiB25±12.9%
L3-Dark-Planet-8BQ3_K_L8.0B4.21 GiB0.50 GiB5.55 GiB0.03 GiB26±12.9%
Vero-Qwen35-9B-BaseI1-Q3_K_L9.4B4.59 GiB0.13 GiB5.55 GiB0.03 GiB26±12.9%
Vero-Qwen35-9BI1-Q3_K_L9.4B4.59 GiB0.13 GiB5.55 GiB0.03 GiB26±12.9%
Qwen3.5-9B-Claude-4.6-Opus-Deckard-V4.2-Uncensored-Heretic-ThinkingI1-Q3_K_L9.4B4.59 GiB0.13 GiB5.55 GiB0.03 GiB26±12.9%
Morphos-9BI1-Q3_K_L9.0B4.59 GiB0.13 GiB5.55 GiB0.03 GiB26±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.

Measured on this card

third-party benchmarks, aggregated
Workload◍MedianMiddle 50%Runs
Image generation0.31 it/s0.21–2.479
Benchmarked· n=9

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 GeForce RTX 3050 run?
1197 of 2118 indexed open-weight models fit a GeForce RTX 3050 at 4,096 context with f16 KV cache, the largest being gemma-3-12b-it-vl-Gemini-3-Pro-Preview-Heretic-Uncensored-Thinking at I1-IQ2_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 3050 actually have?
Its nameplate is 6 GB, but about 5.58 GiB is available to a model once driver and compositor overhead is accounted for.
Is a GeForce RTX 3050 fast for local AI?
Its memory bandwidth is 168 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.