RUNLOCALAIv38
→WILL IT RUNBEST GPUCOMPARETROUBLESHOOTSTARTPULSEMODELSHARDWARETOOLSBENCH
RUNLOCALAI

Independently operated catalog for local-AI hardware and software. Hand-written verdicts. Source-cited claims. Reproducible commands when we have them.

OP·Fredoline Eruo
DIR
  • Models
  • Hardware
  • Tools
  • Benchmarks
TOOLS
  • Will it run?
  • Compare hardware
  • Cost vs cloud
  • Choose my GPU
  • Quick answers
REF
  • All buyer guides
  • Methodology
  • Glossary
  • Errors KB
  • Trust
EDITOR
  • About
  • Author
  • How we make money
  • Editorial policy
  • Contact
LEGAL
  • Privacy
  • Terms
  • Sitemap
MAIL · MONTHLY DIGEST
Get monthly local AI changes
Monthly recap. No spam.
DISCLOSURE

Some links on this site are affiliate links (Amazon Associates and other first-class retailers). When you buy through them, we earn a small commission at no extra cost to you. Affiliate links do not influence our verdicts — there are cards we rate highly that we don't have affiliate relationships with, and cards that sell well that we refuse to recommend. Read more →

© 2026 runlocalai.coIndependently operated
RUNLOCALAI · v38
Will it run? / NVIDIA GeForce RTX 4090 Mobile / vision

What can NVIDIA GeForce RTX 4090 Mobile run for vision?

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

Memory: 16 GB VRAM + 32 GB system RAM
Runner: llama.cpp / Ollama (CUDA)
AnyChatCodingAgentsReasoningVisionLong contextCreative

Runs comfortably
9 models

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

#1Moondream 2
1.9B
other
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 5.3 GBHeadroom: 10.7 GB
326
tok/s
E
Weights
1.15 GB
KV cache
0.24 GB
Activations
2.11 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
#2LLaVA 1.6 Mistral 7B
7B
other
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 9.2 GBHeadroom: 6.8 GB
89
tok/s
E
Weights
4.23 GB
KV cache
0.88 GB
Activations
2.26 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
#3Qwen 2.5-VL 7B
7B
qwen
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 9.2 GBHeadroom: 6.8 GB
89
tok/s
E
Weights
4.23 GB
KV cache
0.88 GB
Activations
2.26 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
#4Qwen 2-VL 7B
7B
qwen
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 9.2 GBHeadroom: 6.8 GB
89
tok/s
E
Weights
4.23 GB
KV cache
0.88 GB
Activations
2.26 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
#5LLaVA-OneVision 7B
7B
other
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 9.2 GBHeadroom: 6.8 GB
89
tok/s
E
Weights
4.23 GB
KV cache
0.88 GB
Activations
2.26 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
#6MiniCPM-V 2.6 8B
8B
minicpm
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 9.9 GBHeadroom: 6.1 GB
78
tok/s
E
Weights
4.83 GB
KV cache
1.00 GB
Activations
2.29 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
#7MiniCPM-V 3 8B
8B
minicpm
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 9.9 GBHeadroom: 6.1 GB
78
tok/s
E
Weights
4.83 GB
KV cache
1.00 GB
Activations
2.29 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
#8PaliGemma 2 3B
3B
gemma
Commercial OK
Quant: BF16Context: 2,048VRAM: 10.5 GBHeadroom: 5.5 GB
62
tok/s
E
Weights
6.00 GB
KV cache
0.38 GB
Activations
2.35 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
#9Whisper Large v3
1.55B
other
Commercial OK
Quant: FP16Context: 0VRAM: 5.1 GBHeadroom: 10.9 GB
121
tok/s
E
Weights
3.10 GB
KV cache
0.00 GB
Activations
0.16 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →

Runs with tradeoffs
19 models

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

Gemma 4 E4B (Effective 4B)
4B
gemma
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 14.5 GBHeadroom: 1.5 GB
  • • Tight VRAM fit — only 1.5 GB headroom left for context growth
ollama run gemma4:e4b
155
tok/s
E
Weights
2.42 GB
KV cache
2.00 GB
Activations
8.31 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
Gemma 3 4B
4B
gemma
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 14.5 GBHeadroom: 1.5 GB
  • • Tight VRAM fit — only 1.5 GB headroom left for context growth
ollama run gemma3:4b
155
tok/s
E
Weights
2.42 GB
KV cache
2.00 GB
Activations
8.31 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
Gemma 4 E2B (Effective 2B)
2B
gemma
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 13.2 GBHeadroom: 2.8 GB
  • • Tight VRAM fit — only 2.8 GB headroom left for context growth
ollama run gemma4:e2b
176
tok/s
E
Weights
2.13 GB
KV cache
1.00 GB
Activations
8.30 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
Phi-3.5 Vision
4.2B
phi
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 14.8 GBHeadroom: 1.2 GB
  • • Tight VRAM fit — only 1.2 GB headroom left for context growth
148
tok/s
E
Weights
2.54 GB
KV cache
2.10 GB
Activations
8.32 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
Qwen 2.5-VL 3B
3B
qwen
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 13.4 GBHeadroom: 2.6 GB
  • • Tight VRAM fit — only 2.6 GB headroom left for context growth
207
tok/s
E
Weights
1.81 GB
KV cache
1.50 GB
Activations
8.28 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
Janus-Pro 7B
7B
janus
Commercial OK
Quant: Q4_K_MContext: 4,096VRAM: 12.1 GBHeadroom: 3.9 GB
  • • Tight VRAM fit — only 3.9 GB headroom left for context growth
89
tok/s
E
Weights
4.23 GB
KV cache
1.75 GB
Activations
4.31 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
Molmo 7B-D
8B
other
Commercial OK
Quant: Q4_K_MContext: 4,096VRAM: 13.0 GBHeadroom: 3.0 GB
  • • Tight VRAM fit — only 3.0 GB headroom left for context growth
78
tok/s
E
Weights
4.83 GB
KV cache
2.00 GB
Activations
4.34 GB
Runtime
1.80 GB
Model details →Run-on benchmark page →
Llama 3.2 11B Vision Instruct
11B
llama
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 12.2 GBHeadroom: 3.8 GB
  • • Tight VRAM fit — only 3.8 GB headroom left for context growth
ollama run llama3.2-vision:11b
56
tok/s
E
Weights
6.64 GB
KV cache
1.38 GB
Activations
2.38 GB
Runtime
1.80 GB
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: 44 new comfortable, 82 new tradeoff

  • • Gemma 3 1B
  • • Llama 3.2 1B Instruct
  • • DeepSeek R1 Distill Qwen 7B
  • • Llama 3.1 8B Instruct
Shop this upgrade↗

Upgrade to NVIDIA RTX 2080 Ti 22GB (China-mod)

~$350

22 GB VRAM (vs your 16 GB) plus a bandwidth jump from ~576 GB/s to ~616 GB/s.

Unlocks: 65 new comfortable

  • • Gemma 3 1B
  • • Llama 3.2 1B Instruct
  • • Gemma 4 E2B (Effective 2B)
  • • Llama 3.2 3B Instruct
Shop this upgrade↗

Add a second NVIDIA GeForce RTX 4090 Mobile

see current pricing

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: 100 new comfortable

  • • Gemma 3 1B
  • • Llama 3.2 1B Instruct
  • • Gemma 4 E2B (Effective 2B)
  • • Llama 3.2 3B 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.

DeepSeek V4 Pro (1.6T MoE)
1600B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (16 GB) + 60% of system RAM (19 GB) combined.

—
Qwen 3.5 235B-A17B (MoE)
397B
qwen
Commercial OK

Even with CPU offload, needs more memory than your VRAM (16 GB) + 60% of system RAM (19 GB) combined.

—
Qwen 3 235B-A22B
235B
qwen
Commercial OK

Even with CPU offload, needs more memory than your VRAM (16 GB) + 60% of system RAM (19 GB) combined.

—
Llama 4 Scout
109B
llama
Commercial OK

Even with CPU offload, needs more memory than your VRAM (16 GB) + 60% of system RAM (19 GB) combined.

—
DeepSeek R1 (671B reasoning)
671B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (16 GB) + 60% of system RAM (19 GB) combined.

—

How to read these numbers

M
Measured — we ran this exact combo on owner hardware.

~
Extrapolated — predicted from a measured benchmark on similar-bandwidth hardware.

E
Estimated — pure formula based on VRAM bandwidth and model architecture.

Full methodology →

Want a specific benchmark we don't have? Email support@runlocalai.co and we'll prioritize it.