What can NVIDIA GeForce RTX 5090 run for vision?
Build: RTX 5090 + Ryzen 9 9950X + 64GB DDR5
Memory: 32 GB VRAM + 64 GB system RAM
Runner: llama.cpp / Ollama (CUDA)
Runs comfortably29 models
Ranked by fit for vision use case + predicted speed. Click a row for VRAM breakdown.
Quant: Q4_K_MContext: 8,192VRAM: 15.8 GBHeadroom: 16.2 GBTTFT: fastollama run gemma4:12b161tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 15.8 GBHeadroom: 16.2 GBTTFT: fast
ollama run gemma4:12b161
tok/s
Estimated
Weights
7.60 GB
KV cache
6.00 GB
Activations
0.39 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~246 ms (fast)
Quant: Q4_K_MContext: 8,192VRAM: 13.2 GBHeadroom: 18.8 GBTTFT: fastollama run qwen3.5:9b214tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 13.2 GBHeadroom: 18.8 GBTTFT: fast
ollama run qwen3.5:9b214
tok/s
Estimated
Weights
6.60 GB
KV cache
4.50 GB
Activations
0.34 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~184 ms (fast)
Quant: Q8_0Context: 8,192VRAM: 8.4 GBHeadroom: 23.6 GBTTFT: instantollama run gemma4:e4b274tok/sEstimated
Quant: Q8_0Context: 8,192VRAM: 8.4 GBHeadroom: 23.6 GBTTFT: instant
ollama run gemma4:e4b274
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): ~82 ms (instant)
Quant: Q4_K_MContext: 8,192VRAM: 18.4 GBHeadroom: 13.6 GBTTFT: fastollama run ministral-3:14b138tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 18.4 GBHeadroom: 13.6 GBTTFT: fast
ollama run ministral-3:14b138
tok/s
Estimated
Weights
9.10 GB
KV cache
7.00 GB
Activations
0.46 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~287 ms (fast)
Quant: Q8_0Context: 8,192VRAM: 8.4 GBHeadroom: 23.6 GBTTFT: instantollama run gemma3:4b274tok/sEstimated
Quant: Q8_0Context: 8,192VRAM: 8.4 GBHeadroom: 23.6 GBTTFT: instant
ollama run gemma3:4b274
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): ~82 ms (instant)
Quant: Q8_0Context: 8,192VRAM: 5.1 GBHeadroom: 26.9 GBTTFT: instantollama run gemma4:e2b548tok/sEstimated
Quant: Q8_0Context: 8,192VRAM: 5.1 GBHeadroom: 26.9 GBTTFT: instant
ollama run gemma4:e2b548
tok/s
Estimated
Weights
2.20 GB
KV cache
1.00 GB
Activations
0.12 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~41 ms (instant)
Quant: Q4_K_MContext: 8,192VRAM: 6.5 GBHeadroom: 25.5 GBTTFT: instant459tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 6.5 GBHeadroom: 25.5 GBTTFT: instant
459
tok/s
Estimated
Weights
2.50 GB
KV cache
2.10 GB
Activations
0.13 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~86 ms (instant)
Quant: BF16Context: 8,192VRAM: 9.6 GBHeadroom: 22.4 GBTTFT: instant194tok/sEstimated
Quant: BF16Context: 8,192VRAM: 9.6 GBHeadroom: 22.4 GBTTFT: instant
194
tok/s
Estimated
Weights
6.00 GB
KV cache
1.50 GB
Activations
0.31 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~61 ms (instant)
Quant: Q4_K_MContext: 8,192VRAM: 10.0 GBHeadroom: 22.0 GBTTFT: fast276tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 10.0 GBHeadroom: 22.0 GBTTFT: fast
276
tok/s
Estimated
Weights
4.50 GB
KV cache
3.50 GB
Activations
0.23 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~143 ms (fast)
Quant: Q4_K_MContext: 8,192VRAM: 18.3 GBHeadroom: 13.7 GBTTFT: fast138tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 18.3 GBHeadroom: 13.7 GBTTFT: fast
138
tok/s
Estimated
Weights
9.00 GB
KV cache
7.00 GB
Activations
0.46 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~287 ms (fast)
Quant: Q4_K_MContext: 8,192VRAM: 10.0 GBHeadroom: 22.0 GBTTFT: fast276tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 10.0 GBHeadroom: 22.0 GBTTFT: fast
276
tok/s
Estimated
Weights
4.50 GB
KV cache
3.50 GB
Activations
0.23 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~143 ms (fast)
Quant: Q4_K_MContext: 2,048VRAM: 3.3 GBHeadroom: 28.7 GBTTFT: instant1015tok/sEstimated
Quant: Q4_K_MContext: 2,048VRAM: 3.3 GBHeadroom: 28.7 GBTTFT: instant
1015
tok/s
Estimated
Weights
1.20 GB
KV cache
0.24 GB
Activations
0.06 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~39 ms (instant)
Runs with tradeoffs8 models
Tight VRAM, partial CPU offload, or context-limited.
Quant: Q4_K_MContext: 8,192VRAM: 31.6 GBHeadroom: 0.4 GBTTFT: noticeable- • Tight VRAM fit — only 0.4 GB headroom left for context growth
ollama run gemma4:26b-moe74tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 31.6 GBHeadroom: 0.4 GBTTFT: noticeable
- • Tight VRAM fit — only 0.4 GB headroom left for context growth
ollama run gemma4:26b-moe74
tok/s
Estimated
Weights
16.00 GB
KV cache
13.00 GB
Activations
0.81 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~532 ms (noticeable)
Quant: Q4_K_MContext: 8,192VRAM: 31.6 GBHeadroom: 0.4 GBTTFT: noticeable- • Tight VRAM fit — only 0.4 GB headroom left for context growth
74tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 31.6 GBHeadroom: 0.4 GBTTFT: noticeable
- • Tight VRAM fit — only 0.4 GB headroom left for context growth
74
tok/s
Estimated
Weights
16.00 GB
KV cache
13.00 GB
Activations
0.81 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~532 ms (noticeable)
Quant: Q4_K_MContext: 8,192VRAM: 47.7 GBHeadroom: 22.7 GBTTFT: noticeable- • Partial CPU offload: ~33% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
ollama run nemotron3:33b5tok/sEstimated
Quant: Q4_K_MContext: 8,192VRAM: 47.7 GBHeadroom: 22.7 GBTTFT: noticeable
- • Partial CPU offload: ~33% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
ollama run nemotron3:33b5
tok/s
Estimated
Weights
28.00 GB
KV cache
16.50 GB
Activations
1.41 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~676 ms (noticeable)
Quant: Q4_K_MContext: 4,096VRAM: 62.9 GBHeadroom: 7.5 GBTTFT: noticeable- • Partial CPU offload: ~49% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
2tok/sEstimated
Quant: Q4_K_MContext: 4,096VRAM: 62.9 GBHeadroom: 7.5 GBTTFT: noticeable
- • Partial CPU offload: ~49% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
2
tok/s
Estimated
Weights
41.00 GB
KV cache
18.00 GB
Activations
2.05 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1475 ms (noticeable)
Quant: Q4_K_MContext: 2,048VRAM: 58.8 GBHeadroom: 11.6 GBTTFT: noticeable- • Partial CPU offload: ~46% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
2tok/sEstimated
Quant: Q4_K_MContext: 2,048VRAM: 58.8 GBHeadroom: 11.6 GBTTFT: noticeable
- • Partial CPU offload: ~46% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
2
tok/s
Estimated
Weights
45.00 GB
KV cache
9.75 GB
Activations
2.25 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1597 ms (noticeable)
Quant: Q4_K_MContext: 2,048VRAM: 66.6 GBHeadroom: 3.8 GBTTFT: noticeable- • Partial CPU offload: ~52% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
ollama run llama3.2-vision:90b1tok/sEstimated
Quant: Q4_K_MContext: 2,048VRAM: 66.6 GBHeadroom: 3.8 GBTTFT: noticeable
- • Partial CPU offload: ~52% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
ollama run llama3.2-vision:90b1
tok/s
Estimated
Weights
51.00 GB
KV cache
11.25 GB
Activations
2.55 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1843 ms (noticeable)
Quant: AWQ-INT4Context: 2,048VRAM: 54.9 GBHeadroom: 15.5 GBTTFT: noticeable- • Partial CPU offload: ~42% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
1tok/sEstimated
Quant: AWQ-INT4Context: 2,048VRAM: 54.9 GBHeadroom: 15.5 GBTTFT: noticeable
- • Partial CPU offload: ~42% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
1
tok/s
Estimated
Weights
42.00 GB
KV cache
9.00 GB
Activations
2.10 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1475 ms (noticeable)
Quant: AWQ-INT4Context: 2,048VRAM: 67.7 GBHeadroom: 2.7 GBTTFT: noticeable- • Partial CPU offload: ~53% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
1tok/sEstimated
Quant: AWQ-INT4Context: 2,048VRAM: 67.7 GBHeadroom: 2.7 GBTTFT: noticeable
- • Partial CPU offload: ~53% of layers run on CPU
- • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
1
tok/s
Estimated
Weights
52.00 GB
KV cache
11.25 GB
Activations
2.60 GB
Runtime
1.80 GB
Time to first token (prefill, 512-token prompt): ~1843 ms (noticeable)
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: 187 new comfortable, 48 new tradeoff
- • Qwen 3 0.6B
- • Llama 3.1 8B Instruct
- • Qwen 3 30B-A3B
- • Qwen 2.5 Coder 32B Instruct
Upgrade to NVIDIA A100 40GB
see current pricing
40 GB VRAM (vs your 32 GB) plus a bandwidth jump from ~1792 GB/s to ~1555 GB/s.
Unlocks: 184 new comfortable
- • Qwen 3 0.6B
- • Llama 3.1 8B Instruct
- • Qwen 3 30B-A3B
- • Qwen 2.5 Coder 32B Instruct
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: 226 new comfortable
- • Qwen 3 0.6B
- • Llama 3.1 8B Instruct
- • Qwen 3 30B-A3B
- • Qwen 2.5 Coder 32B Instruct
Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.
Won't runtop 5 popular models
Need more memory than you have. Shown for orientation.
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
Even with CPU offload, needs more memory than your VRAM (32 GB) + 60% of system RAM (38 GB) combined.
—
How to read these numbers
Measured here
Measured here - RunLocalAI ran this exact combo on owner hardware with public evidence.Source-backed
Source-backed / community - a reproduced public source supports the speed, but it is not labeled as owner-measured.Extrapolated
Extrapolated - predicted from a measured benchmark on similar-bandwidth hardware.Estimated
Estimated - formula based on VRAM bandwidth and model architecture; not a benchmark row.Want a specific benchmark we don't have? Email Contact support and we'll prioritize it.