From the RTX 4060 8 GB to the RTX 3090 24 GB: what changes for local AI
+16 GB of VRAM (8 → 24 GB). Of 49 models, 35 newly run well, 8 get a better file or stop offloading, and 6 stay the same.
35newly run well
8better file
6no change
+16GB more VRAM
01Newly runs well
| Model | On the RTX 4060 8 GB | On the RTX 3090 24 GB | Needed |
|---|---|---|---|
| FLUX.1 [dev] | Tight Q3_K_S | Runs well Q8_0 | 15.0 GB |
| FLUX.1 [schnell] | Tight Q3_K_S | Runs well Q8_0 | 15.0 GB |
| FLUX.1 Kontext [dev] | Tight Q3_K_M | Runs well Q8_0 | 15.3 GB |
| FLUX.1 Krea [dev] | Tight Q3_K_M | Runs well Q8_0 | 15.0 GB |
| FLUX.1 Fill [dev] | Tight Q3_K_S | Runs well Q8_0 | 15.3 GB |
| FLUX.2 [klein] 9B | Tight Q3_K_M | Runs well 16-bit | 20.5 GB |
| Krea 2 (Turbo) | Tight Q2_K | Runs well Q8_0 | 16.3 GB |
| Qwen-Image | Offload only Q2_K | Runs well FP8 | 23.2 GB |
| Qwen-Image-Edit (2511) | Offload only Q4_K_M | Runs well FP8 | 23.5 GB |
| Qwen-Image 2.1 | Runs Q5_K_M | Runs well 16-bit | 16.5 GB |
| Z-Image Turbo | Runs Q6_K | Runs well 16-bit | 14.3 GB |
| Z-Image (base) | Runs Q5_K_M | Runs well 16-bit | 14.3 GB |
| Ideogram 4 | Offload only Q4_1 | Runs well Q8_0 | 22.6 GB |
| Boogu-Image (Turbo) | Offload only Q4_1 | Runs well 16-bit | 23.2 GB |
| ERNIE-Image (Turbo) | Runs Q4_K_M | Runs well 16-bit | 18.4 GB |
| HiDream-O1-Image | Offload only FP8 | Runs well 16-bit | 19.2 GB |
| HiDream-I1 (Full) | Offload only Q2_K | Runs well Q8_0 | 21.5 GB |
| HiDream-I1 (Dev) | Offload only Q2_K | Runs well Q8_0 | 21.5 GB |
| Stable Diffusion 3.5 Large | Runs Q4_1 | Runs well 16-bit | 18.8 GB |
| Chroma1-HD | Runs Q4_K_M | Runs well 16-bit | 20.1 GB |
| HunyuanImage 2.1 | Offload only Q4_K_M | Runs well Q8_0 | 23.1 GB |
| Wan 2.1 T2V 14B | Offload only Q4_K_M | Runs well Q8_0 | 20.2 GB |
| Wan 2.1 I2V 14B 480P | Offload only Q4_K_M | Runs well Q8_0 | 22.4 GB |
| Wan 2.1 I2V 14B 720P | Offload only Q4_K_M | Runs well FP8 | 23.2 GB |
| Wan 2.1 VACE 14B | Offload only Q4_K_M | Runs well Q8_0 | 23.5 GB |
| Wan 2.2 T2V A14B | Offload only Q2_K | Runs well Q8_0 | 19.7 GB |
| Wan 2.2 I2V A14B | Offload only Q2_K | Runs well Q8_0 | 19.7 GB |
| Wan 2.2 TI2V 5B | Runs Q5_K_M | Runs well 16-bit | 13.8 GB |
| Wan 2.2 Animate 14B | Offload only Q4_K_M | Runs well FP8 | 22.6 GB |
| Wan Animate 2 (14B) | Offload only Q4_K_M | Runs well Q8_0 | 23.4 GB |
| Wan 2.2 S2V 14B | Offload only Q4_K_M | Runs well FP8 | 21.2 GB |
| SCAIL-2 (character animation) | Offload only Q4_K_M | Runs well Q8_0 | 23.4 GB |
| HunyuanVideo (13B, original) | Offload only Q4_K_M | Runs well Q8_0 | 18.3 GB |
| LTX-Video 13B (0.9.8) | Offload only Q2_K | Runs well Q8_0 | 18.3 GB |
| HunyuanVideo 1.5 | Offload only Q4_K_M | Runs well 16-bit | 21.0 GB |
“Runs well” means a 16-bit or 8-bit file fits entirely in VRAM.
02A better file, or less offloading
| Model | On the RTX 4060 8 GB | On the RTX 3090 24 GB | Needed |
|---|---|---|---|
| FLUX.2 [dev] | Not practical Q2_K | Runs Q4_K_M | 23.4 GB |
| LTX-2 (19B) | Offload only Q4_K_M | Runs Q6_K | 20.8 GB |
| LTX-2.3 (22B) | Offload only Q4_K_M | Runs Q6_K | 22.6 GB |
| LTX-2.5 (22B) | Offload only Q4_K_M | Runs Q6_K | 23.5 GB |
| MiniMax H3 (33B) | Not practical Q3_K_M | Tight Q3_K_M | 21.4 GB |
| MiniMax H3 Pruned | Offload only Q4_K_M | Runs Q6_K | 22.5 GB |
| FLUX.2 [klein] 4B | Runs well FP8 | Runs well 16-bit | 9.6 GB |
| Mage-Flow (Microsoft) | Runs well INT8 | Runs well 16-bit | 10.0 GB |
03No change in what fits
Lumina 2.0 Runs wellSD 3.5 Medium Runs wellSDXL Runs wellIllustrious / Pony Runs wellSD 1.5 Runs wellWan 2.1 1.3B Runs well
Same verdict and same file on both cards. Speed can still differ.
04Speed and features
Memory bandwidth goes from 272 to 936 GB/s (×3.44). For models that fit on both cards, bandwidth and compute decide the speed; this ratio is a rough first guide, not a benchmark.