NVIDIA GeForce RTX 4060 Ti 16GB vs NVIDIA GeForce RTX 5090
For running models locally, priced in Germany. At 8k context the first runs 16 models fully in memory, the second 29.
Specifications
| Spec | NVIDIA GeForce RTX 4060 Ti 16GB | NVIDIA GeForce RTX 5090 |
|---|---|---|
| GPU | GeForce RTX 4060 Ti 16GB | GeForce RTX 5090 |
| Memory | 16 GB GDDR6 | 32 GB GDDR7 |
| Memory bus | 128-bit | 512-bit |
| Memory bandwidth | 288 GB/s | 1,792 GB/s |
| FP16 compute | 44 TFLOPS | 209 TFLOPS |
| Software | CUDA | CUDA |
| Board power | 165 W | 575 W |
| PCIe generation | Gen 4 | Gen 5 |
| PCIe lanes | x8 | x16 |
What each runs
Estimated decode speed with the most precise quant that fits each one est.
| Model | NVIDIA GeForce RTX 4060 Ti 16GB | NVIDIA GeForce RTX 5090 |
|---|---|---|
| Qwen 3.8 27B | offload | 37tok/sQ8_0 |
| Qwen 3.6 27B | offload | 38tok/sQ8_0 |
| Gemma 4 31B | offload | 42tok/sQ6_K |
| Qwen 3.6 35B-A3B | offload | 249tok/sUD-Q6_K |
| Qwen 3.5 27B | offload | 38tok/sQ8_0 |
| Gemma 4 26B-A4B | offload | 147tok/sQ8_0 |
| Devstral Small 2 24B | 11tok/sQ4_K_M | 42tok/sQ8_0 |
| Qwen 3.5 35B-A3B | offload | 253tok/sQ6_K |
| GLM 4.7 Flash | offload | 117tok/sQ8_0 |
| gpt-oss 20B | 41tok/sMXFP4 | 253tok/sMXFP4 |
| Qwen3 Coder 30B-A3B | offload | 204tok/sQ6_K |
| Kimi Linear 48B-A3B | offload | 343tok/sQ4_K_M |
| Qwen3 32B | offload | 38tok/sQ6_K |
| Gemma 4 12B | 14tok/sQ8_0 | 85tok/sQ8_0 |
| Qwen 3.5 9B | 18tok/sQ8_0 | 113tok/sQ8_0 |
| Mistral Small 3.2 24B | 11tok/sQ4_K_M | 42tok/sQ8_0 |
| Gemma 3 27B | offload | 37tok/sQ8_0 |
| DeepSeek R1 Distill Qwen 32B | offload | 38tok/sQ6_K |
| Qwen3 14B | 13tok/sQ6_K | 65tok/sQ8_0 |
| Ministral 3 14B | 11tok/sQ8_0 | 71tok/sQ8_0 |
| DeepSeek R1 0528 Qwen3 8B | 18tok/sQ8_0 | 112tok/sQ8_0 |
| Qwen3 8B | 18tok/sQ8_0 | 112tok/sQ8_0 |
| Phi-4 14B | 13tok/sQ6_K | 64tok/sQ8_0 |
| Qwen 3.5 4B | 37tok/sQ8_0 | 231tok/sQ8_0 |
| Ministral 3 8B | 18tok/sQ8_0 | 109tok/sQ8_0 |
| Granite 4.2 8B | 17tok/sQ8_0 | 104tok/sQ8_0 |
| Gemma 4 E4B | 21tok/sQ8_0 | 131tok/sQ8_0 |
| LFM2.5 8B-A1B | 73tok/sQ8_0 | 455tok/sQ8_0 |
| Llama 3.1 8B | 23tok/sQ6_K | 145tok/sQ6_K |
| Qwen 3.5 122B-A10B | offload | offload |
| Nemotron 3 Super 120B-A12B | won't run | offload |
| Qwen3 Coder Next | won't run | offload |
| Mistral Small 4 119B | offload | offload |
| GLM 4.5 Air | won't run | offload |
| Qwen3 Next 80B-A3B | won't run | offload |
| Nemotron 3.5 Lightning 30B-A3B | offload | offload |
| Llama 3.3 70B | won't run | offload |
| DeepSeek V4 Flash | won't run | won't run |
| GLM 5.3 Flash | won't run | won't run |
| Qwen 3.8 Flash Next | won't run | won't run |
| MiniMax M2.7 | won't run | won't run |
| gpt-oss 120B | won't run | won't run |