What can NVIDIA GeForce RTX 4090 run locally?

143 models fit comfortably on NVIDIA GeForce RTX 4090 with 32 GB system RAM. 56 more run with tradeoffs. Top pick: Gemma 4 E4B (Effective 4B).

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 4090 based on windows support and the GPU backend.

What can NVIDIA GeForce RTX 4090 run for long context?

Build: NVIDIA GeForce RTX 4090 + — + 32 GB RAM (windows)

Memory: 24 GB VRAM + 32 GB system RAM

Runner: llama.cpp / Ollama (CUDA)

Any Chat Coding Agents Reasoning Vision Long context Creative

Runs comfortably
143 models

Ranked by fit for long context use case + predicted speed. Click a row for VRAM breakdown.

#1Gemma 4 E4B (Effective 4B)

gemma

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 8.4 GBHeadroom: 15.6 GBTTFT: fast

ollama run gemma4:e4b

154

tok/s

Estimated

Weights

4.40 GB

KV cache

2.00 GB

Activations

0.23 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~124 ms (fast)

Model details →

#2Phi-3.5 Mini Instruct

3.8B

phi

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 8.0 GBHeadroom: 16.0 GBTTFT: fast

ollama run phi3.5:3.8b

162

tok/s

Estimated

Weights

4.10 GB

KV cache

1.90 GB

Activations

0.21 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~118 ms (fast)

Model details →

#3Command R7B (12-2024)

command-r

Quant: Q4_K_MContext: 8,192VRAM: 10.4 GBHeadroom: 13.6 GBTTFT: fast

136

tok/s

Estimated

Weights

4.40 GB

KV cache

4.00 GB

Activations

0.23 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~248 ms (fast)

Model details →

#4Falcon Mamba 7B

falcon

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 9.7 GBHeadroom: 14.3 GBTTFT: fast

155

tok/s

Estimated

Weights

4.20 GB

KV cache

3.50 GB

Activations

0.22 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~217 ms (fast)

Model details →

#5Codestral Mamba 7B

mistral

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 9.7 GBHeadroom: 14.3 GBTTFT: fast

155

tok/s

Estimated

Weights

4.20 GB

KV cache

3.50 GB

Activations

0.22 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~217 ms (fast)

Model details →

#6Ministral 3B Instruct

mistral

Quant: Q4_K_MContext: 8,192VRAM: 5.3 GBHeadroom: 18.7 GBTTFT: instant

362

tok/s

Estimated

Weights

1.90 GB

KV cache

1.50 GB

Activations

0.10 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~93 ms (instant)

Model details →

#7Ministral 8B Instruct

mistral

Quant: Q4_K_MContext: 8,192VRAM: 11.1 GBHeadroom: 12.9 GBTTFT: fast

136

tok/s

Estimated

Weights

5.00 GB

KV cache

4.00 GB

Activations

0.26 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~248 ms (fast)

Model details →

#8Qwen 2.5 7B Instruct

qwen

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 10.8 GBHeadroom: 13.2 GBTTFT: fast

ollama run qwen2.5:7b

tok/s

Estimated

Weights

8.10 GB

KV cache

0.47 GB

Activations

0.41 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~217 ms (fast)

Model details →

#9Mistral Nemo 12B Instruct

12B

mistral

Commercial OK

Quant: Q5_K_MContext: 8,192VRAM: 16.9 GBHeadroom: 7.1 GBTTFT: fast

ollama run mistral-nemo:12b

tok/s

Estimated

Weights

8.70 GB

KV cache

6.00 GB

Activations

0.44 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~372 ms (fast)

Model details →

#10Qwen 3 14B

14B

qwen

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 17.6 GBHeadroom: 6.4 GBTTFT: fast

ollama run qwen3:14b

tok/s

Estimated

Weights

8.40 GB

KV cache

7.00 GB

Activations

0.43 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~434 ms (fast)

Model details →

#11Qwen 3 8B

qwen

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 14.4 GBHeadroom: 9.6 GBTTFT: fast

ollama run qwen3:8b

tok/s

Estimated

Weights

8.20 GB

KV cache

4.00 GB

Activations

0.42 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~248 ms (fast)

Model details →

#12Qwen 2.5 14B Instruct

14B

qwen

Commercial OK

Quant: Q5_K_MContext: 8,192VRAM: 19.8 GBHeadroom: 4.2 GBTTFT: fast

ollama run qwen2.5:14b

tok/s

Estimated

Weights

10.50 GB

KV cache

7.00 GB

Activations

0.53 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~434 ms (fast)

Model details →

Runs with tradeoffs
56 models

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

Nemotron 3 Nano (30B-A3B)

30B

other

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 35.7 GBHeadroom: 7.5 GBTTFT: noticeable

• Partial CPU offload: ~33% of layers run on CPU

ollama run nemotron3:nano

tok/s

Estimated

Weights

18.00 GB

KV cache

15.00 GB

Activations

0.91 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~930 ms (noticeable)

Model details →

Jamba 1.5 Mini

52B

other

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 39.8 GBHeadroom: 3.4 GBTTFT: fast

• Partial CPU offload: ~40% of layers run on CPU

tok/s

Estimated

Weights

30.00 GB

KV cache

6.50 GB

Activations

1.50 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~372 ms (fast)

Model details →

Gemma 4 Turkish 26B (4B active)

26B

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 20.1 GBHeadroom: 3.9 GBTTFT: noticeable

• Tight VRAM fit — only 3.9 GB headroom left for context growth

tok/s

Estimated

Weights

14.30 GB

KV cache

3.25 GB

Activations

0.72 GB

Runtime

1.80 GB

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

Model details →

Gemma 3 27B

27B

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 22.0 GBHeadroom: 2.0 GBTTFT: noticeable

• Tight VRAM fit — only 2.0 GB headroom left for context growth

ollama run gemma3:27b

tok/s

Estimated

Weights

16.00 GB

KV cache

3.38 GB

Activations

0.80 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~837 ms (noticeable)

Model details →

Qwen 3 30B-A3B

30B

qwen

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 35.7 GBHeadroom: 7.5 GBTTFT: noticeable

• Partial CPU offload: ~33% of layers run on CPU

ollama run qwen3:30b

tok/s

Estimated

Weights

18.00 GB

KV cache

15.00 GB

Activations

0.91 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~930 ms (noticeable)

Model details →

Muse Glimmer 30B

30B

other

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 20.4 GBHeadroom: 3.6 GBTTFT: noticeable

• Tight VRAM fit — only 3.6 GB headroom left for context growth

ollama run muse-glimmer

tok/s

Estimated

Weights

17.00 GB

KV cache

0.71 GB

Activations

0.86 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~930 ms (noticeable)

Model details →

Gemma 4 31B Dense

31B

gemma

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 36.2 GBHeadroom: 7.0 GBTTFT: noticeable

• Partial CPU offload: ~34% of layers run on CPU

ollama run gemma4:31b

tok/s

Estimated

Weights

18.00 GB

KV cache

15.50 GB

Activations

0.91 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~961 ms (noticeable)

Model details →

Qwen 3 32B

32B

qwen

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 37.8 GBHeadroom: 5.4 GBTTFT: noticeable

• Partial CPU offload: ~36% of layers run on CPU

ollama run qwen3:32b

tok/s

Estimated

Weights

19.00 GB

KV cache

16.00 GB

Activations

0.96 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~992 ms (noticeable)

Model details →

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: 41 new comfortable, 73 new tradeoff

• Qwen 3 0.6B
• Qwen 3 1.7B
• Gemma 3 270M
• SmolLM2 135M Instruct

Shop this upgrade↗

Upgrade to NVIDIA RTX PRO 4500 Blackwell

see current pricing

32 GB VRAM (vs your 24 GB) plus a bandwidth jump from ~1008 GB/s to ~896 GB/s.

Unlocks: 80 new comfortable

• Qwen 3 0.6B
• Qwen 3 30B-A3B
• Qwen 2.5 Coder 32B Instruct
• Qwen3.6 27B

Shop this upgrade↗

Add a second NVIDIA GeForce RTX 4090

~$1899

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: 93 new comfortable

• Qwen 3 0.6B
• Qwen 3 30B-A3B
• Qwen 2.5 Coder 32B Instruct
• Qwen3.6 27B

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.

DeepSeek V4 Pro (1.6T MoE)

1600B

deepseek

Commercial OK