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 RTX 4090 48GB (China-mod) / vision

What can NVIDIA RTX 4090 48GB (China-mod) run for vision?

Build: NVIDIA RTX 4090 48GB (China-mod) + — + 32 GB RAM (windows)

Memory: 48 GB VRAM + 32 GB system RAM
Runner: llama.cpp / Ollama (CUDA)
AnyChatCodingAgentsReasoningVisionLong contextCreative

Runs comfortably
27 models

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

#1Gemma 4 E2B (Effective 2B)
2B
gemma
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 13.2 GBHeadroom: 34.8 GBTTFT: instant
ollama run gemma4:e2b
308
tok/s
E
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): ~62 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: 33.2 GBTTFT: fast
258
tok/s
E
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): ~130 ms (fast)
Model details →Run-on benchmark page →
#3Gemma 4 E4B (Effective 4B)
4B
gemma
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 16.5 GBHeadroom: 31.5 GBTTFT: fast
ollama run gemma4:e4b
154
tok/s
E
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): ~124 ms (fast)
Model details →Run-on benchmark page →
#4Gemma 3 4B
4B
gemma
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 16.5 GBHeadroom: 31.5 GBTTFT: fast
ollama run gemma3:4b
154
tok/s
E
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): ~124 ms (fast)
Model details →Run-on benchmark page →
#5LLaVA 1.6 Mistral 7B
7B
other
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 17.9 GBHeadroom: 30.1 GBTTFT: fast
155
tok/s
E
Weights
4.23 GB
KV cache
3.50 GB
Activations
8.40 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~217 ms (fast)
Model details →Run-on benchmark page →
#6Qwen 2.5-VL 3B
3B
qwen
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 13.4 GBHeadroom: 34.6 GBTTFT: instant
362
tok/s
E
Weights
1.81 GB
KV cache
1.50 GB
Activations
8.28 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~93 ms (instant)
Model details →Run-on benchmark page →
#7Qwen 2.5-VL 7B
7B
qwen
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 17.9 GBHeadroom: 30.1 GBTTFT: fast
155
tok/s
E
Weights
4.23 GB
KV cache
3.50 GB
Activations
8.40 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~217 ms (fast)
Model details →Run-on benchmark page →
#8Molmo 7B-D
8B
other
Commercial OK
Quant: Q4_K_MContext: 4,096VRAM: 13.0 GBHeadroom: 35.0 GBTTFT: fast
136
tok/s
E
Weights
4.83 GB
KV cache
2.00 GB
Activations
4.34 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~248 ms (fast)
Model details →Run-on benchmark page →
#9MiniCPM-V 2.6 8B
8B
minicpm
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 19.1 GBHeadroom: 28.9 GBTTFT: fast
136
tok/s
E
Weights
4.83 GB
KV cache
4.00 GB
Activations
8.43 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~248 ms (fast)
Model details →Run-on benchmark page →
#10Janus-Pro 7B
7B
janus
Commercial OK
Quant: Q4_K_MContext: 4,096VRAM: 12.1 GBHeadroom: 35.9 GBTTFT: fast
155
tok/s
E
Weights
4.23 GB
KV cache
1.75 GB
Activations
4.31 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~217 ms (fast)
Model details →Run-on benchmark page →
#11Qwen 2-VL 7B
7B
qwen
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 17.9 GBHeadroom: 30.1 GBTTFT: fast
155
tok/s
E
Weights
4.23 GB
KV cache
3.50 GB
Activations
8.40 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~217 ms (fast)
Model details →Run-on benchmark page →
#12MiniCPM-V 3 8B
8B
minicpm
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 19.1 GBHeadroom: 28.9 GBTTFT: fast
136
tok/s
E
Weights
4.83 GB
KV cache
4.00 GB
Activations
8.43 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~248 ms (fast)
Model details →Run-on benchmark page →

Runs with tradeoffs
3 models

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

Gemma 4 31B Dense
31B
gemma
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 45.1 GBHeadroom: 2.9 GBTTFT: noticeable
  • • Tight VRAM fit — only 2.9 GB headroom left for context growth
ollama run gemma4:31b
35
tok/s
E
Weights
18.72 GB
KV cache
15.50 GB
Activations
9.13 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~961 ms (noticeable)
Model details →Run-on benchmark page →
Molmo 72B
72B
other
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 58.5 GBHeadroom: 8.7 GBTTFT: slow
  • • Partial CPU offload: ~18% of layers run on CPU
15
tok/s
E
Weights
43.47 GB
KV cache
9.00 GB
Activations
4.22 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~2231 ms (slow)
Model details →Run-on benchmark page →
InternVL 2.5 78B
78B
other
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 63.0 GBHeadroom: 4.2 GBTTFT: slow
  • • Partial CPU offload: ~24% of layers run on CPU
14
tok/s
E
Weights
47.09 GB
KV cache
9.75 GB
Activations
4.40 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~2417 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: 97 new comfortable, 28 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 H100 SXM

see current pricing

80 GB VRAM (vs your 48 GB) plus a bandwidth jump from ~1008 GB/s to ~3350 GB/s.

Unlocks: 120 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 RTX 4090 48GB (China-mod)

~$2400

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: 126 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.

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

Even with CPU offload, needs more memory than your VRAM (48 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 (48 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 (48 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 (48 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 (48 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.