NVIDIA · consumer

GeForce RTX 3060 Ti GDDR6X

GeForce RTX 3060 Ti GDDR6X has 8 GB of VRAM at 608 GB/s — about 7.44 GiB usable after driver and compositor overhead. 1223 of 2118 indexed models fit at 16K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
8 GB
GDDR6X
Bandwidth
608 GB/s
256-bit bus
Tensor FP16
65 TF
dense
TDP
225 W
$399 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1034embedding 26vision language 96audio asr 38video 8image 1audio tts 20

What fits at 16K context

largest quantization that fits, per model · 1223 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3-VL-Embedding-8BQ4_K_M8.1B4.36 GiB2.25 GiB7.44 GiB0.00 GiB64±12.9%
qwen-indic-v1I1-Q4_K_M7.6B4.36 GiB2.25 GiB7.44 GiB0.00 GiB64±12.9%
Qwen3-Embedding-8BQ4_K_M7.6B4.36 GiB2.25 GiB7.44 GiB0.00 GiB64±12.9%
Teuken-7B-instruct-research-v0.4I1-Q6_K7.5B6.10 GiB0.50 GiB7.44 GiB0.00 GiB65±12.9%
OmniAtlas-Qwen3-30B-A3BI1-IQ1_M31.7B6.59 GiB0.00 GiB7.44 GiB0.00 GiB65±12.9%
Qwen3-Omni-30B-A3B-CaptionerI1-IQ1_M31.7B6.59 GiB0.00 GiB7.44 GiB0.00 GiB65±12.9%
CycleGRPO-4BQ8_04.8B4.37 GiB2.25 GiB7.43 GiB0.01 GiB64±12.9%
Falcon3-10B-InstructIQ3_XS10.3B4.07 GiB2.50 GiB7.43 GiB0.01 GiB65±12.9%
Jan-v3-4B-base-instructQ8_04.4B4.37 GiB2.25 GiB7.43 GiB0.01 GiB64±12.9%
Jan-code-4bQ8_04.4B4.37 GiB2.25 GiB7.43 GiB0.01 GiB64±12.9%
Ministral-8B-Instruct-2410Q4_08.0B4.35 GiB2.25 GiB7.43 GiB0.01 GiB64±12.9%
Qwen3-4B-BaseQ8_04.0B4.37 GiB2.25 GiB7.43 GiB0.01 GiB64±12.9%
Bonsai-8B-unpackedIQ4_XS8.2B4.35 GiB2.25 GiB7.43 GiB0.01 GiB65±12.9%
Assistant_Pepe_8BQ4_K_M—4.59 GiB2.00 GiB7.43 GiB0.01 GiB65±12.9%
Foundation-Sec-8B-InstructI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Foundation-Sec-8B-Instruct-hereticI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
dolphin-2.9-llama3-8bQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
saiga_llama3_8bQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
llama-3-8b-Instruct-bnb-4bitQ4_K_M8.2B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Meta-Llama-3-8B-InstructQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
MiniCPM-Llama3-V-2_5Q4_K8.5B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
llama-3-8b-bnb-4bitQ4_K_M8.2B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama3-ChatQA-1.5-8BQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Meta-Llama-3-8BQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Foundation-Sec-8B-ReasoningQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama-3.1-Tulu-3-8BQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama-3-Groq-8B-Tool-UseQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
llama3.1-heretic-unsensoredI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Dolphin3.0-Llama3.1-8BQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
dolphin-2.9.4-llama3.1-8bQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Dolphin3.0-Llama3.1-8B-abliteratedQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
LLAMA-3_8B_Unaligned_BETAQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama-3.1-8BQ4_K8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Anubis-Mini-8B-v1I1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Meta-Llama-3-8BQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Anubis-Mini-8B-v1-hereticQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Deepseek-R1-Distill-NSFW-RPv1Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama-3.3-8B-Instruct-128KQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
L3.1-Dark-Reasoning-LewdPlay-evo-Hermes-R1-Uncensored-8B-hereticI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
L3.1-Dark-Reasoning-LewdPlay-evo-Hermes-R1-Uncensored-8BI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Gluon-8BI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-ReasoningI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama3.3-8B-Instruct-Thinking-Claude-4.5-Opus-High-ReasoningI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Hypnos-i1-8BQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
grok-oss-Apollyon-8B-hereticI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama-3.3-8B-Instruct-128K-JbliteratedI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
PsyCoPref-Llama3-8BI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama3.1-GptDeluxe-8BI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
SelfCite-8B-CC-SFTI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
ai-girlfriend-v2I1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama-3.3-8B-Instruct-128K_AbliteratedI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Aisha-Uncensored-v2I1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
grok-oss-Apollyon-8BI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
HARC-Llama-3.1-8B-InstructI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Lumen-1.2.5I1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Stoicism1_Llama3.1-8b-instructI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
grok-oss-Thanatos-8BI1-Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama-3.1-8B-Lexi-Uncensored-V2Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
Llama-3.1-8B-InstructQ4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±12.9%
L3-8B-Sunfall-v0.5-Stheno-v3.2Q4_K_M8.0B4.58 GiB2.00 GiB7.42 GiB0.02 GiB65±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 Ti GDDR6X run?
1223 of 2118 indexed open-weight models fit a GeForce RTX 3060 Ti GDDR6X at 16,384 context with f16 KV cache, the largest being Qwen3-VL-Embedding-8B at Q4_K_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 3060 Ti GDDR6X 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 3060 Ti GDDR6X fast for local AI?
Its memory bandwidth is 608 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.