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. 607 of 2118 indexed models fit at 128K context with q4_0 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 467audio asr 36vision language 63audio tts 17embedding 21video 3

What fits at 128K context

largest quantization that fits, per model · 607 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Felldude-Uncensored-Ministral3-3B-bf16I1-IQ2_XS3.8B1.10 GiB3.66 GiB5.57 GiB0.01 GiB25±12.9%
Amaretto-3BI1-IQ2_XS4.3B1.10 GiB3.66 GiB5.57 GiB0.01 GiB25±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%
granite-vision-4.1-4bQ4_K_M4.0B1.96 GiB2.81 GiB5.57 GiB0.01 GiB25±12.9%
Parable-Granite-4.1-3B-Claude-Fable-5I1-Q4_K_M3.4B1.96 GiB2.81 GiB5.57 GiB0.01 GiB25±12.9%
granite-4.0-microQ4_K_M3.4B1.96 GiB2.81 GiB5.57 GiB0.01 GiB25±12.9%
granite-4.1-3bQ4_K_M3.4B1.96 GiB2.81 GiB5.57 GiB0.01 GiB25±12.9%
granite-4.0-micro-baseQ4_K_M3.4B1.96 GiB2.81 GiB5.57 GiB0.01 GiB25±12.9%
granite-3.3-2b-instructQ6_K_L2.5B1.96 GiB2.81 GiB5.57 GiB0.01 GiB25±12.9%
granite-3.2-2b-instructQ6_K_L2.5B1.96 GiB2.81 GiB5.57 GiB0.01 GiB25±12.9%
granite-3.1-2b-instructQ6_K_L2.5B1.96 GiB2.81 GiB5.57 GiB0.01 GiB25±12.9%
granite-vision-3.2-2bQ6_K_L3.0B1.96 GiB2.81 GiB5.57 GiB0.01 GiB25±12.9%
gemma-3n-E4B-itQ4_K_M7.8B4.23 GiB0.51 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%
Supertron2-Reranker-2BI1-IQ3_M2.1B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
Uni-MuMER-Qwen3-VL-2BI1-IQ3_M2.1B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
Qwen3-VL-Reranker-2BI1-IQ3_M2.1B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
OpenCaption-2B-VL-SFT-v1.0I1-IQ3_M2.1B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
Atomight-V2.5-1.7BI1-IQ3_M1.7B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
OpenClaude-1.7B-MergedI1-IQ3_M1.7B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
Qwen3-VL-2B-InstructIQ3_M2.1B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
gaon-1.7b-v2-translateI1-IQ3_M1.7B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
gaon-1.7b-v2-instructI1-IQ3_M1.7B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
Lightning-1.7BIQ3_M1.7B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
DorsetHeatwaveLLM2I1-IQ3_M1.7B0.83 GiB3.94 GiB5.57 GiB0.01 GiB25±12.9%
MARTHA-LXVII.8B_QWEN-3.5-3.6_prune_9b-3.6_baseI1-IQ3_M8.1B3.74 GiB0.98 GiB5.56 GiB0.02 GiB26±12.9%
t5-v1_1-xxlQ2_K4.8B4.72 GiB0.00 GiB5.56 GiB0.02 GiB26±12.9%
gemma-4-E4B-it-abliteratedI1-IQ3_XS8.0B4.23 GiB0.51 GiB5.56 GiB0.02 GiB26±12.9%
gemma-4-E4B-uncensoredI1-IQ3_XS7.9B4.23 GiB0.51 GiB5.56 GiB0.02 GiB26±12.9%
gemma-4-E4B-it-qat-q4_0-unquantized-hereticI1-IQ3_XS7.9B4.23 GiB0.51 GiB5.56 GiB0.02 GiB26±12.9%
gemma-4-E4B-it-qat-heretic_decensoredI1-IQ3_XS7.9B4.23 GiB0.51 GiB5.56 GiB0.02 GiB26±12.9%
gemma-4-E4B-it-QAT-SOMPOA-heresyI1-IQ3_XS7.9B4.23 GiB0.51 GiB5.56 GiB0.02 GiB26±12.9%
gemma4-e4b-mahou-nsfwI1-IQ3_XS7.9B4.23 GiB0.51 GiB5.56 GiB0.02 GiB26±12.9%
gemma-4-E4B-it-mentalchat16kI1-IQ3_XS7.9B4.23 GiB0.51 GiB5.56 GiB0.02 GiB26±12.9%
gemma4-E4B-it-abliteratedI1-IQ3_XS7.9B4.23 GiB0.51 GiB5.56 GiB0.02 GiB26±12.9%
gemma-4-E4B-it-OBLITERATEDI1-IQ3_XS8.0B4.23 GiB0.51 GiB5.56 GiB0.02 GiB26±12.9%
Ornith-1.0-9BIQ2_M9.2B3.60 GiB1.13 GiB5.56 GiB0.02 GiB26±12.9%
granite-3.1-3b-a800m-instructMoEQ6_K_L3.3B2.54 GiB2.25 GiB5.56 GiB0.02 GiB24±37%
Qwen3-1.7BIQ3_XXS2.0B0.83 GiB3.94 GiB5.56 GiB0.02 GiB25±12.9%
Llama-Doctor-3.2-3B-InstructI1-IQ1_S3.2B0.81 GiB3.94 GiB5.55 GiB0.03 GiB26±12.9%
Llama-3.2-3B-Instruct-roleplay-tunedI1-IQ1_S3.2B0.81 GiB3.94 GiB5.55 GiB0.03 GiB26±12.9%
Llama-3.2-3B-Instruct-heretic-ablitered-uncensoredI1-IQ1_S3.2B0.81 GiB3.94 GiB5.55 GiB0.03 GiB26±12.9%
Llama3.2-3B-creative-writer-v0.1I1-IQ1_S3.2B0.81 GiB3.94 GiB5.55 GiB0.03 GiB26±12.9%
Firefly-V3.2I1-IQ1_S3.2B0.81 GiB3.94 GiB5.55 GiB0.03 GiB26±12.9%
Firefly-V3I1-IQ1_S3.2B0.81 GiB3.94 GiB5.55 GiB0.03 GiB26±12.9%
granite-4.0-h-tinyMoEQ5_K_S6.9B4.50 GiB0.28 GiB5.55 GiB0.03 GiB70±37%
Teuken-7B-instruct-research-v0.4I1-IQ3_S7.5B3.58 GiB1.13 GiB5.55 GiB0.03 GiB26±12.9%
Nanbeige4.2-3BQ2_K4.2B1.64 GiB3.09 GiB5.55 GiB0.03 GiB26±12.9%
granite-vision-3.3-2bQ6_K3.0B1.94 GiB2.81 GiB5.55 GiB0.03 GiB25±12.9%
Qwen3-VL-2B-ThinkingQ3_K_S2.1B0.81 GiB3.94 GiB5.54 GiB0.04 GiB26±12.9%
Qwen3-VL-Embedding-2BQ3_K_S2.1B0.81 GiB3.94 GiB5.54 GiB0.04 GiB26±12.9%
granite-4.0-h-tiny-baseMoEQ5_06.9B4.48 GiB0.28 GiB5.53 GiB0.05 GiB71±37%
InternVL3_5-8BQ4_K_M8.5B4.68 GiB0.00 GiB5.53 GiB0.05 GiB26±12.9%
Vero-Qwen35-9B-BaseI1-Q2_K9.4B3.56 GiB1.13 GiB5.52 GiB0.06 GiB26±12.9%
Vero-Qwen35-9BI1-Q2_K9.4B3.56 GiB1.13 GiB5.52 GiB0.06 GiB26±12.9%
Qwen3.5-9B-Claude-4.6-Opus-Deckard-V4.2-Uncensored-Heretic-ThinkingI1-Q2_K9.4B3.56 GiB1.13 GiB5.52 GiB0.06 GiB26±12.9%
Morphos-9BI1-Q2_K9.0B3.56 GiB1.13 GiB5.52 GiB0.06 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?
607 of 2118 indexed open-weight models fit a GeForce RTX 3050 at 131,072 context with q4_0 KV cache, the largest being Felldude-Uncensored-Ministral3-3B-bf16 at I1-IQ2_XS. 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.