What can NVIDIA GeForce RTX 3060 12GB run locally?

7 models fit comfortably on NVIDIA GeForce RTX 3060 12GB with 32 GB system RAM. 38 more run with tradeoffs. Top pick: Gemma 4 E2B (Effective 2B).

What's the recommended runner for this hardware?

llama.cpp / Ollama (CUDA). This is the runner that gives best tokens-per-second on NVIDIA GeForce RTX 3060 12GB based on windows support and the GPU backend.

What can NVIDIA GeForce RTX 3060 12GB run?

Build: RTX 3060 12GB + Ryzen 5 5600 + 32GB DDR4 (cheapest path)

Memory: 12 GB VRAM + 32 GB system RAM

Runner: llama.cpp / Ollama (CUDA)

Any Chat Coding Agents Reasoning Vision Long context Creative

Runs comfortably
7 models

Full-VRAM resident, with room for context. No compromises.

#1Gemma 4 E2B (Effective 2B)

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 5.4 GBHeadroom: 6.6 GBTTFT: fast

ollama run gemma4:e2b

194

tok/s

Weights

1.21 GB

KV cache

0.25 GB

Activations

2.11 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~403 ms (fast)

Model details →Run-on benchmark page →

#2Llama 3.2 3B Instruct

llama

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 6.1 GBHeadroom: 5.9 GBTTFT: noticeable

ollama run llama3.2:3b

129

tok/s

Weights

1.81 GB

KV cache

0.38 GB

Activations

2.14 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~605 ms (noticeable)

Model details →Run-on benchmark page →

#3Phi-3.5 Mini Instruct

3.8B

phi

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 6.7 GBHeadroom: 5.3 GBTTFT: noticeable

ollama run phi3.5:3.8b

102

tok/s

Weights

2.29 GB

KV cache

0.47 GB

Activations

2.16 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~766 ms (noticeable)

Model details →Run-on benchmark page →

#4Gemma 4 E4B (Effective 4B)

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 6.9 GBHeadroom: 5.1 GBTTFT: noticeable

ollama run gemma4:e4b

tok/s

Weights

2.42 GB

KV cache

0.50 GB

Activations

2.17 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~806 ms (noticeable)

Model details →Run-on benchmark page →

#5Qwen 3 4B

qwen

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 6.9 GBHeadroom: 5.1 GBTTFT: noticeable

ollama run qwen3:4b

tok/s

Weights

2.42 GB

KV cache

0.50 GB

Activations

2.17 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~806 ms (noticeable)

Model details →Run-on benchmark page →

#6Gemma 3 4B

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 6.9 GBHeadroom: 5.1 GBTTFT: noticeable

ollama run gemma3:4b

tok/s

Weights

2.42 GB

KV cache

0.50 GB

Activations

2.17 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~806 ms (noticeable)

Model details →Run-on benchmark page →

#7Phi-3.5 Vision

4.2B

phi

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 7.0 GBHeadroom: 5.0 GBTTFT: noticeable

tok/s

Weights

2.54 GB

KV cache

0.53 GB

Activations

2.17 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~847 ms (noticeable)

Model details →Run-on benchmark page →

Runs with tradeoffs
38 models

Tight VRAM, partial CPU offload, or context-limited.

Llama 3.1 8B Instruct

llama

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 9.2 GBHeadroom: 2.8 GBTTFT: noticeable

• Tight VRAM fit — only 2.8 GB headroom left for context growth

ollama run llama3.1:8b

tok/s

Weights

4.83 GB

KV cache

0.27 GB

Activations

2.29 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~1613 ms (noticeable)

Model details →Run-on benchmark page →

Qwen 3 30B-A3B

30B

qwen

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 26.6 GBHeadroom: 4.6 GBTTFT: slow

• Partial CPU offload: ~55% of layers run on CPU
• CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here

ollama run qwen3:30b

tok/s

Weights

18.11 GB

KV cache

3.75 GB

Activations

2.95 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~6047 ms (slow)

Model details →Run-on benchmark page →

Qwen 2.5 Coder 32B Instruct

32B

qwen

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 24.7 GBHeadroom: 6.5 GBTTFT: slow

• Partial CPU offload: ~51% of layers run on CPU
• CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here

ollama run qwen2.5-coder:32b

tok/s

Weights

19.32 GB

KV cache

0.54 GB

Activations

3.01 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~6450 ms (slow)

Model details →Run-on benchmark page →

Qwen 3 32B

32B

qwen

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 28.1 GBHeadroom: 3.1 GBTTFT: slow

• Partial CPU offload: ~57% of layers run on CPU
• CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here

ollama run qwen3:32b

tok/s

Weights

19.32 GB

KV cache

4.00 GB

Activations

3.01 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~6450 ms (slow)

Model details →Run-on benchmark page →

Gemma 4 31B Dense

31B

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 27.4 GBHeadroom: 3.8 GBTTFT: slow

• Partial CPU offload: ~56% of layers run on CPU
• CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here

ollama run gemma4:31b

tok/s

Weights

18.72 GB

KV cache

3.88 GB

Activations

2.98 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~6249 ms (slow)

Model details →Run-on benchmark page →

Qwen 3 8B

qwen

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 9.9 GBHeadroom: 2.1 GBTTFT: noticeable

• Tight VRAM fit — only 2.1 GB headroom left for context growth

ollama run qwen3:8b

tok/s

Weights

4.83 GB

KV cache

1.00 GB

Activations

2.29 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~1613 ms (noticeable)

Model details →Run-on benchmark page →

DeepSeek R1 Distill Qwen 32B

32B

deepseek

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 28.1 GBHeadroom: 3.1 GBTTFT: slow

• Partial CPU offload: ~57% of layers run on CPU
• CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here

ollama run deepseek-r1:32b

tok/s

Weights

19.32 GB

KV cache

4.00 GB

Activations

3.01 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~6450 ms (slow)

Model details →Run-on benchmark page →

Gemma 4 26B MoE

26B

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 23.6 GBHeadroom: 7.6 GBTTFT: slow

• Partial CPU offload: ~49% of layers run on CPU
• CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here

ollama run gemma4:26b-moe

tok/s

Weights

15.70 GB

KV cache

3.25 GB

Activations

2.83 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~5241 ms (slow)

Model details →Run-on benchmark page →

What if you upgraded?

Hypothetical scenarios. We re-ran the compatibility engine for each.

+32 GB system RAM

~$80–150

Doubles your CPU-offload working set. Helps when models don't quite fit in VRAM.

Unlocks: 40 new tradeoff

• Llama 3.1 8B Instruct
• Qwen 3 30B-A3B
• Qwen 2.5 Coder 32B Instruct
• Llama 3.3 70B Instruct

Shop this upgrade↗

Upgrade to NVIDIA GeForce RTX 4080 Super

~$1099

16 GB VRAM (vs your 12 GB) plus a bandwidth jump from ~360 GB/s to ~736 GB/s.

Unlocks: 10 new comfortable

• Gemma 3 1B
• Llama 3.2 1B Instruct
• DeepSeek R1 Distill Qwen 7B
• Llama 3.1 8B Instruct

Shop this upgrade↗

Add a second NVIDIA GeForce RTX 3060 12GB

~$249

Tensor parallelism splits the model across both cards, effectively doubling VRAM. Bandwidth doesn't double — runs ~1.5× the single-card speed in practice.

Unlocks: 13 new comfortable

• Gemma 3 1B
• Llama 3.2 1B Instruct
• Llama 3.1 Nemotron Nano 8B
• Mistral 7B Instruct v0.3

Shop this upgrade↗

Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.

Won't run
top 5 popular models

Need more memory than you have. Shown for orientation.

Qwen 3 235B-A22B

235B

qwen

Commercial OK