NVIDIA · consumer

GeForce RTX 3060 OEM

GeForce RTX 3060 OEM has 6 GB of VRAM at 336 GB/s — about 5.58 GiB usable after driver and compositor overhead. 1188 of 2118 indexed models fit at 8K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
6 GB
GDDR6
Bandwidth
336 GB/s
192-bit bus
Tensor FP16
57 TF
dense
TDP
185 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1010embedding 26audio asr 38audio tts 19vision language 91video 3image 1

What fits at 8K context

largest quantization that fits, per model · 1188 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
L3-Dark-Planet-8BQ3_K_L8.0B4.21 GiB0.53 GiB5.58 GiB0.00 GiB50±12.9%
LFM2-8B-A1BMoEQ4_K_L8.3B4.73 GiB0.05 GiB5.58 GiB0.00 GiB142±37%
Qwen3-VL-Embedding-8BQ4_K_S8.1B4.15 GiB0.60 GiB5.58 GiB0.00 GiB50±12.9%
qwen-indic-v1I1-Q4_K_S7.6B4.15 GiB0.60 GiB5.58 GiB0.00 GiB50±12.9%
Qwen3-Embedding-8BQ4_K_S7.6B4.15 GiB0.60 GiB5.58 GiB0.00 GiB50±12.9%
Ministral-8B-Instruct-2410IQ4_XS8.0B4.14 GiB0.60 GiB5.57 GiB0.01 GiB50±12.9%
canary-qwen-2.5bBF162.6B4.73 GiB0.00 GiB5.57 GiB0.01 GiB50±12.9%
EXAONE-Deep-7.8BQ4_K_L7.8B4.73 GiB0.00 GiB5.57 GiB0.01 GiB50±12.9%
EXAONE-3.5-7.8B-InstructQ4_K_L7.8B4.73 GiB0.00 GiB5.57 GiB0.01 GiB50±12.9%
Apertus-8B-Instruct-2509IQ4_XS8.1B4.17 GiB0.53 GiB5.57 GiB0.01 GiB51±12.9%
Qwen3-TTS-12Hz-0.6B-BaseQ4_K_M915M4.72 GiB0.00 GiB5.57 GiB0.01 GiB50±12.9%
VoxCPM2F162.3B4.72 GiB0.00 GiB5.57 GiB0.01 GiB50±12.9%
Rocinante-XL-16B-v1I1-IQ1_M16.1B3.82 GiB0.90 GiB5.56 GiB0.02 GiB50±12.9%
t5-v1_1-xxlQ2_K4.8B4.72 GiB0.00 GiB5.56 GiB0.02 GiB50±12.9%
Gemma-4-12B-StyleTuneI1-IQ2_S13.0B4.20 GiB0.51 GiB5.56 GiB0.02 GiB50±12.9%
gemma-4-12b-heretic-styletune-headI1-IQ2_S12.0B4.20 GiB0.51 GiB5.56 GiB0.02 GiB50±12.9%
syrian-gemma-12bI1-IQ2_S13.0B4.20 GiB0.51 GiB5.56 GiB0.02 GiB50±12.9%
Qwen3.6-12B-IQ-Ultra-Heretic-Uncensored-Thinking-V2-HightopQ2_K12.1B4.60 GiB0.10 GiB5.56 GiB0.02 GiB51±12.9%
zeta-2Q3_K_L8.3B4.19 GiB0.53 GiB5.56 GiB0.02 GiB50±12.9%
gemma-4-E4B-uncensoredI1-Q3_K_L7.9B4.65 GiB0.09 GiB5.56 GiB0.02 GiB50±12.9%
gemma-4-E4B-it-qat-q4_0-unquantized-hereticI1-Q3_K_L7.9B4.65 GiB0.09 GiB5.56 GiB0.02 GiB50±12.9%
gemma-4-E4B-it-qat-heretic_decensoredI1-Q3_K_L7.9B4.65 GiB0.09 GiB5.56 GiB0.02 GiB50±12.9%
gemma-4-E4B-it-QAT-SOMPOA-heresyI1-Q3_K_L7.9B4.65 GiB0.09 GiB5.56 GiB0.02 GiB50±12.9%
gemma4-e4b-mahou-nsfwI1-Q3_K_L7.9B4.65 GiB0.09 GiB5.56 GiB0.02 GiB50±12.9%
gemma-4-E4B-it-mentalchat16kI1-Q3_K_L7.9B4.65 GiB0.09 GiB5.56 GiB0.02 GiB50±12.9%
gemma4-E4B-it-abliteratedI1-Q3_K_L7.9B4.65 GiB0.09 GiB5.56 GiB0.02 GiB50±12.9%
gemma-4-E4B-it-OBLITERATEDI1-Q3_K_L8.0B4.65 GiB0.09 GiB5.56 GiB0.02 GiB50±12.9%
next-8bI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Supertron2-Reranker-8BI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
next-ocrI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Qwen3-VL-8B-GLM-4.7-Flash-Heretic-Uncensored-ThinkingI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Midas-FableAgent-8BI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Qwen3-VL-8B-Heretic-1.3.0I1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Qwen3-VL-8B-Thinking-Unredacted-MAXI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Qwen3-VL-8B-Instruct-Minecraft-MT-en-zhI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Qwen-3-VL-8B-Instruct-hereticI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Poe-8B-GLM5-Opus4.6-Sonnet4.5-Kimi-Grok-Gemini-3-pro-preview-HERETICI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
ToolCUA-8BI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Huihui-Qwen3-VL-8B-Instruct-abliteratedI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Qwen3-VL-8B-Instruct-Unredacted-MAXQ3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Qwen3-VL-Reranker-8BI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Salience-1-9BI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Qwen3-VL-8B-Instruct-Uncensored-V2I1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Maestro1-9BI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
nsfwcaption-qwen3-vl-8b-v3-safetensorsQ3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
GRaPE-2-FlashI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Jan-v2-VL-medI1-Q3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Jan-v2-VL-highQ3_K_L8.8B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Parable-Qwen3-8B-Claude-Fable-5I1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
ReasonCritic-7BI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
mythos-9b-unhinged-hereticI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Finch-8B-KTOI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Finch-8BI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
MathSmith-hc-Qwen3-8BI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
MiroThinker-v1.0-8BI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
mythos-9b-unhingedI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Ektome-Qwen3-8B-PristinelyUncensoredI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
Marco-DeepResearch-8BI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
mythos-9b-mergedI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±12.9%
qwen3-8b-apostateI1-Q3_K_L8.2B4.13 GiB0.60 GiB5.56 GiB0.02 GiB50±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 3060 OEM run?
1188 of 2118 indexed open-weight models fit a GeForce RTX 3060 OEM at 8,192 context with q8_0 KV cache, the largest being L3-Dark-Planet-8B at Q3_K_L. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 3060 OEM 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 3060 OEM fast for local AI?
Its memory bandwidth is 336 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.