RTX 5090 32 GB vs RTX 4090 24 GB for local AI
The RTX 5090 32 GB has 8 GB more memory, and that decides it for local AI: it runs a better file (or runs at all) for 13 of 49 models, and 47 run at 8-bit or better against 43.
RTX 5090 32 GB
- VRAM
- 32 GB
- Memory
- GDDR7
- Bandwidth
- 1792 GB/s
- Architecture
- Blackwell
- FP8 compute
- Yes
- Launch price
- $1,999
VS
RTX 4090 24 GB
- VRAM
- 24 GB
- Memory
- GDDR6X
- Bandwidth
- 1008 GB/s
- Architecture
- Ada Lovelace
- FP8 compute
- Yes
- Launch price
- $1,599
01Model by model
| Model | RTX 5090 32 GB | RTX 4090 24 GB | ||
|---|---|---|---|---|
| image models | ||||
| FLUX.1 dev | Runs well 16-bit | Runs well FP8 | ||
| FLUX.1 schnell | Runs well 16-bit | Runs well FP8 | ||
| FLUX.1 Kontext | Runs well 16-bit | Runs well FP8 | ||
| FLUX.1 Krea | Runs well 16-bit | Runs well FP8 | ||
| FLUX.1 Fill | Runs well 16-bit | Runs well Q8_0 | ||
| FLUX.2 dev | Runs Q6_K | Runs Q4_K_M | ||
| FLUX.2 klein 9B | Runs well 16-bit | Runs well 16-bit | ||
| FLUX.2 klein 4B | Runs well 16-bit | Runs well 16-bit | ||
| Krea 2 | Runs well 16-bit | Runs well FP8 | ||
| Qwen-Image | Runs well FP8 | Runs well FP8 | ||
| Qwen-Image-Edit | Runs well FP8 | Runs well FP8 | ||
| Qwen-Image 2.1 | Runs well 16-bit | Runs well 16-bit | ||
| Z-Image Turbo | Runs well 16-bit | Runs well 16-bit | ||
| Z-Image | Runs well 16-bit | Runs well 16-bit | ||
| Ideogram 4 | Runs well FP8 | Runs well FP8 | ||
| Boogu-Image | Runs well 16-bit | Runs well 16-bit | ||
| ERNIE-Image | Runs well 16-bit | Runs well 16-bit | ||
| HiDream-O1 | Runs well 16-bit | Runs well 16-bit | ||
| Mage-Flow | Runs well 16-bit | Runs well 16-bit | ||
| Lumina 2.0 | Runs well 16-bit | Runs well 16-bit | ||
| HiDream-I1 Full | Runs well FP8 | Runs well FP8 | ||
| HiDream-I1 | Runs well FP8 | Runs well FP8 | ||
| SD 3.5 Large | Runs well 16-bit | Runs well 16-bit | ||
| SD 3.5 Medium | Runs well 16-bit | Runs well 16-bit | ||
| Chroma1-HD | Runs well 16-bit | Runs well 16-bit | ||
| SDXL | Runs well 16-bit | Runs well 16-bit | ||
| Illustrious / Pony | Runs well 16-bit | Runs well 16-bit | ||
| SD 1.5 | Runs well 16-bit | Runs well 16-bit | ||
| HunyuanImage 2.1 | Runs well FP8 | Runs well FP8 | ||
| video models | ||||
| Wan 2.1 14B | Runs well FP8 | Runs well FP8 | ||
| Wan 2.1 1.3B | Runs well 16-bit | Runs well 16-bit | ||
| Wan 2.1 I2V 480P | Runs well FP8 | Runs well FP8 | ||
| Wan 2.1 I2V 720P | Runs well FP8 | Runs well FP8 | ||
| Wan VACE 14B | Runs well Q8_0 | Runs well Q8_0 | ||
| Wan 2.2 T2V | Runs well FP8 | Runs well FP8 | ||
| Wan 2.2 I2V | Runs well FP8 | Runs well FP8 | ||
| Wan 2.2 5B | Runs well 16-bit | Runs well 16-bit | ||
| Wan 2.2 Animate | Runs well FP8 | Runs well FP8 | ||
| Wan Animate 2 | Runs well Q8_0 | Runs well Q8_0 | ||
| Wan 2.2 S2V | Runs well FP8 | Runs well FP8 | ||
| SCAIL-2 | Runs well FP8 | Runs well FP8 | ||
| HunyuanVideo 13B | Runs well 16-bit | Runs well FP8 | ||
| LTX-Video 13B | Runs well FP8 | Runs well FP8 | ||
| HunyuanVideo 1.5 | Runs well 16-bit | Runs well 16-bit | ||
| LTX-2 | Runs well Q8_0 | Runs Q6_K | ||
| LTX-2.3 | Runs well Q8_0 | Runs Q6_K | ||
| LTX-2.5 | Runs well Q8_0 | Runs Q6_K | ||
| MiniMax H3 | Runs Q5_K_M | Tight Q3_K_M | ||
| MiniMax H3 Pruned | Runs well FP8 | Runs Q6_K | ||
Highlighted: the GPU that runs a better file for that model. Launch prices are the maker’s original list prices, not today’s street prices. Calculated from real file sizes; how the numbers work.
02Score
47run well on RTX 5090 32 GB
43run well on RTX 4090 24 GB
13better on RTX 5090 32 GB
0better on RTX 4090 24 GB