§01·spec · /gpus
RTX 5090
nvidia50 series32GB VRAM
51 open-weights AI models run on the RTX 5090 — see which fit, how fast they go, and the VRAM each needs.
RTX 5090D, the China-market export version, ships the same 32 GB of GDDR7 on the same 512-bit bus — so every “fits / doesn’t fit” verdict on this page applies to it unchanged. It has the same die and core count, with AI throughput capped in firmware, so the speeds here are an upper bound for it rather than a match. Nothing on the RTX 5090D has been measured yet — send a run if you have one.
§02·models that run on this GPU
51 totalLLM · 16
| Model | Best speed | Min VRAM | Works | Evidence | |
|---|---|---|---|---|---|
| gpt-oss 20B | 298.2tokens/s | 16GB | ✓ | 3benchesrecipe | check ↗ |
| Qwen3 30B-A3B | 226.1tokens/s | 19GB | ✓ | 2benchesrecipe | check ↗ |
| Qwen3-8B | 200.4tokens/s | 6GB | ✓ | 3benchesrecipe | check ↗ |
| Qwen3 14B | 123.8tokens/s | 17GB | ✓ | 3benchesrecipe | check ↗ |
| Qwen3 32B | 61.4tokens/s | 29GB | ✓ | 3benchesrecipe | check ↗ |
| Apodex 1.1 mini | 32GB | ✓ | recipe | check ↗ | |
| DeepSeek R1 Distill 14B | 10GB | ✓ | recipe | check ↗ | |
| Devstral Small 2 (24B) | 32GB | ✓ | recipe | check ↗ | |
| Laguna XS 2.1 | 24GB | ✓ | recipe | check ↗ | |
| Llama 3.1 70B | 32GB | ✓ | recipe | check ↗ | |
| Llama 3.1 8B | 10GB | ✓ | recipe | check ↗ | |
| Mistral Small 3.2 24B | 32GB | ✓ | recipe | check ↗ | |
| Nanbeige4.2 3B | 32GB | ✓ | recipe | check ↗ | |
| North Mini Code 1.0 | 32GB | ✓ | recipe | check ↗ | |
| Ornith 1.0 35B | 32GB | ✓ | recipe | check ↗ | |
| Phi-4 | 32GB | ✓ | recipe | check ↗ |
- Best speed
- 298.2tokens/s
- Min VRAM
- 16GB
- Evidence
- 3benches
- Best speed
- 226.1tokens/s
- Min VRAM
- 19GB
- Evidence
- 2benches
- Best speed
- 200.4tokens/s
- Min VRAM
- 6GB
- Evidence
- 3benches
- Best speed
- 123.8tokens/s
- Min VRAM
- 17GB
- Evidence
- 3benches
- Best speed
- 61.4tokens/s
- Min VRAM
- 29GB
- Evidence
- 3benches
- Best speed
- Min VRAM
- 32GB
- Best speed
- Min VRAM
- 10GB
- Best speed
- Min VRAM
- 32GB
- Best speed
- Min VRAM
- 24GB
- Best speed
- Min VRAM
- 32GB
- Best speed
- Min VRAM
- 10GB
- Best speed
- Min VRAM
- 32GB
- Best speed
- Min VRAM
- 32GB
- Best speed
- Min VRAM
- 32GB
- Best speed
- Min VRAM
- 32GB
- Best speed
- Min VRAM
- 32GB
Multimodal · 10
| Model | Best speed | Min VRAM | Works | Evidence | |
|---|---|---|---|---|---|
| Gemma 4 26B MoE | 180.3tokens/s | 29GB | ✓ | 2benchesrecipe | check ↗ |
| Qwen3.5 35B | 165.2tokens/s | 24GB | ✓ | 2benchesrecipe | check ↗ |
| Gemma4 31B | 61.1tokens/s | 20GB | ✓ | 2benchesrecipe | check ↗ |
| Qwen3.5 27B | 58.8tokens/s | 20GB | ✓ | 2benchesrecipe | check ↗ |
| Fara1.5-27B | 32GB | ✓ | recipe | check ↗ | |
| Fara1.5-4B | 24GB | ✓ | recipe | check ↗ | |
| Fara1.5-9B | 24GB | ✓ | recipe | check ↗ | |
| Gemma 4 E4B-IT | 20GB | ✓ | recipe | check ↗ | |
| Muse Glimmer 30B | 32GB | ✓ | recipe | check ↗ | |
| Qwen3.8 27B | 32GB | ✓ | recipe | check ↗ |
- Best speed
- 180.3tokens/s
- Min VRAM
- 29GB
- Evidence
- 2benches
- Best speed
- 165.2tokens/s
- Min VRAM
- 24GB
- Evidence
- 2benches
- Best speed
- 61.1tokens/s
- Min VRAM
- 20GB
- Evidence
- 2benches
- Best speed
- 58.8tokens/s
- Min VRAM
- 20GB
- Evidence
- 2benches
- Best speed
- Min VRAM
- 32GB
- Best speed
- Min VRAM
- 24GB
- Best speed
- Min VRAM
- 24GB
- Best speed
- Min VRAM
- 20GB
- Best speed
- Min VRAM
- 32GB
- Best speed
- Min VRAM
- 32GB
Image · 9
| Model | Best speed | Min VRAM | Works | Evidence | |
|---|---|---|---|---|---|
| Flux.1 Dev | 2.38it/s | 24GB | ✓ | 4benchesrecipe | check ↗ |
| Chroma V48 | 24GB | ✓ | recipe | check ↗ | |
| ERNIE-Image-Turbo | 24GB | ✓ | recipe | check ↗ | |
| Flux.2-Klein-4B | 9GB | ✓ | recipe | check ↗ | |
| HiDream-O1-Image | 17GB | ✓ | recipe | check ↗ | |
| Krea 2 | 16GB | ✓ | recipe | check ↗ | |
| LongCat Image | 18GB | ✓ | recipe | check ↗ | |
| Qwen-Image | 21GB | ✓ | recipe | check ↗ | |
| Qwen-Image-2.1 | 12GB | ✓ | recipe | check ↗ |
- Best speed
- 2.38it/s
- Min VRAM
- 24GB
- Evidence
- 4benches
- Best speed
- Min VRAM
- 24GB
- Best speed
- Min VRAM
- 24GB
- Best speed
- Min VRAM
- 9GB
- Best speed
- Min VRAM
- 17GB
- Best speed
- Min VRAM
- 16GB
- Best speed
- Min VRAM
- 18GB
- Best speed
- Min VRAM
- 21GB
- Best speed
- Min VRAM
- 12GB
Video · 10
| Model | Best speed | Min VRAM | Works | Evidence | |
|---|---|---|---|---|---|
| CogVideoX 1.5 | 20GB | ✓ | recipe | check ↗ | |
| HunyuanVideo 1.5 | 14GB | ✓ | recipe | check ↗ | |
| LightX2V | 22GB | ✓ | recipe | check ↗ | |
| LTX-2.5 | 24GB | ✓ | recipe | check ↗ | |
| LTX-2.3 | 28GB | ✓ | recipe | check ↗ | |
| MiniMax H3 (Hailuo 3) | 12GB | ✓ | recipe | check ↗ | |
| Mochi 1 | 22GB | ✓ | recipe | check ↗ | |
| Sulphur 2 | 31GB | ✓ | recipe | check ↗ | |
| Wan 2.2 TI2V-5B | 8GB | ✓ | recipe | check ↗ | |
| Wan 2.2 14B | 24GB | ✓ | recipe | check ↗ |
- Best speed
- Min VRAM
- 20GB
- Best speed
- Min VRAM
- 14GB
- Best speed
- Min VRAM
- 22GB
- Best speed
- Min VRAM
- 24GB
- Best speed
- Min VRAM
- 28GB
- Best speed
- Min VRAM
- 12GB
- Best speed
- Min VRAM
- 22GB
- Best speed
- Min VRAM
- 31GB
- Best speed
- Min VRAM
- 8GB
- Best speed
- Min VRAM
- 24GB
3D · 4
| Model | Best speed | Min VRAM | Works | Evidence | |
|---|---|---|---|---|---|
| Hunyuan3D | 29GB | ✓ | recipe | check ↗ | |
| TRELLIS.2-4B | 24GB | ✓ | recipe | check ↗ | |
| TRELLIS image-large | 16GB | ✓ | recipe | check ↗ | |
| Waypoint 1.5 | 12GB | ✓ | recipe | check ↗ |
TTS · 1
| Model | Best speed | Min VRAM | Works | Evidence | |
|---|---|---|---|---|---|
| Kokoro TTS | 3GB | ✓ | recipe | check ↗ |
§03·tested recipes
showing 6 of 51- imageintermediate12GB+recipe
Qwen-Image-2.1 on RTX 5090: what stays resident in 32 GB, int8 vs bf16, and third-party timings
- musicintermediate16GB+recipe
YuE2-3B on RTX 5090: the 32GB card the library caps at 22 GiB
- llmintermediate32GB+recipe
Apodex 1.1 mini on RTX 5090: Q5_K_M Agentic LLM at the Full 262K Context
- multimodalintermediate32GB+recipe
Qwen3.8-27B on RTX 5090: the full 262K context window on one card
- multimodalintermediate32GB+recipe
Muse Glimmer 30B on RTX 5090: K-Quant-Dynamic, vision and DFlash in llama.cpp
- videoadvanced24GB+recipe
LTX-2.5 on RTX 5090: 22B Audio-Video in 32 GB via ComfyUI's INT8 Path