What can NVIDIA GeForce RTX 5090 run locally?

9 models fit comfortably on NVIDIA GeForce RTX 5090 with 32 GB system RAM. 2 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 5090 based on windows support and the GPU backend.

What can NVIDIA GeForce RTX 5090 run for vision?

Build: NVIDIA GeForce RTX 5090 + — + 32 GB RAM (windows)

Memory: 32 GB VRAM + 32 GB system RAM

Runner: llama.cpp / Ollama (CUDA)

Any Chat Coding Agents Reasoning Vision Long context Creative

Runs comfortably
9 models

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

#1Gemma 4 E2B (Effective 2B)

gemma

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 13.2 GBHeadroom: 18.8 GBTTFT: instant

ollama run gemma4:e2b

548

tok/s

Weights

2.13 GB

KV cache

1.00 GB

Activations

8.30 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~41 ms (instant)

Model details →Run-on benchmark page →

#2Phi-3.5 Vision

4.2B

phi

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 14.8 GBHeadroom: 17.2 GBTTFT: instant

459

tok/s

Weights

2.54 GB

KV cache

2.10 GB

Activations

8.32 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~86 ms (instant)

Model details →Run-on benchmark page →

#3Gemma 4 E4B (Effective 4B)

gemma

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 16.5 GBHeadroom: 15.5 GBTTFT: instant

ollama run gemma4:e4b

274

tok/s

Weights

4.25 GB

KV cache

2.00 GB

Activations

8.40 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~82 ms (instant)

Model details →Run-on benchmark page →

#4Gemma 3 4B

gemma

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 16.5 GBHeadroom: 15.5 GBTTFT: instant

ollama run gemma3:4b

274

tok/s

Weights

4.25 GB

KV cache

2.00 GB

Activations

8.40 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~82 ms (instant)

Model details →Run-on benchmark page →

#5Llama 3.2 11B Vision Instruct

11B

llama

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 27.8 GBHeadroom: 4.2 GBTTFT: fast

ollama run llama3.2-vision:11b

100

tok/s

Weights

11.69 GB

KV cache

5.50 GB

Activations

8.78 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~225 ms (fast)

Model details →Run-on benchmark page →

#6Gemma 4 26B MoE

26B

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 23.6 GBHeadroom: 8.4 GBTTFT: noticeable

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): ~532 ms (noticeable)

Model details →Run-on benchmark page →

#7Gemma 3 27B

27B

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 24.3 GBHeadroom: 7.7 GBTTFT: noticeable

ollama run gemma3:27b

tok/s

Weights

16.30 GB

KV cache

3.38 GB

Activations

2.86 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~553 ms (noticeable)

Model details →Run-on benchmark page →

#8MedGemma 27B

27B

gemma

Quant: Q4_K_MContext: 2,048VRAM: 24.3 GBHeadroom: 7.7 GBTTFT: noticeable

tok/s

Weights

16.30 GB

KV cache

3.38 GB

Activations

2.86 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~553 ms (noticeable)

Model details →Run-on benchmark page →

#9Gemma 4 31B Dense

31B

gemma

Commercial OK

Quant: Q4_K_MContext: 2,048VRAM: 27.4 GBHeadroom: 4.6 GBTTFT: noticeable

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): ~635 ms (noticeable)

Model details →Run-on benchmark page →

Runs with tradeoffs
2 models

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

Gemma 3 12B

12B

gemma

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 29.4 GBHeadroom: 2.6 GBTTFT: fast

• Tight VRAM fit — only 2.6 GB headroom left for context growth

ollama run gemma3:12b

tok/s

Weights

12.75 GB

KV cache

6.00 GB

Activations

8.83 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~246 ms (fast)

Model details →Run-on benchmark page →

Pixtral 12B

12B

mistral

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 29.4 GBHeadroom: 2.6 GBTTFT: fast

• Tight VRAM fit — only 2.6 GB headroom left for context growth

ollama run pixtral:12b

tok/s

Weights

12.75 GB

KV cache

6.00 GB

Activations

8.83 GB

Runtime

1.80 GB

Time to first token (prefill, 512-token prompt): ~246 ms (fast)

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: 25 new comfortable, 18 new tradeoff

• Gemma 3 1B
• Llama 3.2 1B Instruct
• Llama 3.2 3B Instruct
• Phi-3.5 Mini Instruct

Shop this upgrade↗

Upgrade to NVIDIA A100 40GB

see current pricing

40 GB VRAM (vs your 32 GB) plus a bandwidth jump from ~1792 GB/s to ~? GB/s.

Unlocks: 34 new comfortable

• Gemma 3 1B
• Llama 3.2 1B Instruct
• Llama 3.2 3B Instruct
• Phi-3.5 Mini Instruct

Shop this upgrade↗

Add a second NVIDIA GeForce RTX 5090

~$2499

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: 38 new comfortable

• Gemma 3 1B
• Llama 3.2 1B Instruct
• Llama 3.2 3B Instruct
• Phi-3.5 Mini Instruct

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