Best overallExcellent fit
Qwen2.5 14B Instruct
The sweet spot for 24 GB cards. Meaningfully stronger reasoning than 7B while still fine-tunable locally with QLoRA.
- Params
- 14.8B
- Quant
- Q6_K
- Memory
- 14 GB
Reference
Best models for GeForce RTX 3090
The value pick: the same 24 GB as a 4090 at a fraction of the price, roughly 60–70% of the throughput.
29 models fit23 GB usableEstimated
Top picks
Best on GeForce RTX 3090
Runner-upGood fit
Mistral Small 24B Instruct
Built deliberately for low latency on a single card — fewer layers, wider FFN. Apache 2.0 at a size that usually is not.
- Params
- 23.6B
- Quant
- Q4_K_M
- Memory
- 16 GB
ThirdGood fit
Phi-4 14B
Trained largely on curated synthetic data. Punches far above its size on reasoning and maths; the short context limits what you can do with it.
- Params
- 14.7B
- Quant
- Q8_0
- Memory
- 17 GB
Full results
29 models
Estimated
| # | Model | Params | Licence | Quantization | Memory | ~tok/s | Fit | Fine-tune |
|---|---|---|---|---|---|---|---|---|
| 1 | Qwen2.5 14B Instruct | 14.8B | Apache 2.0 | Q6_K | 14 GB | 60 | Excellent fit | Yes |
| 2 | Mistral Small 24B Instruct | 23.6B | Apache 2.0 | Q4_K_M | 16 GB | 51 | Good fit | Yes |
| 3 | Phi-4 14B | 14.7B | MIT | Q8_0 | 17 GB | 46 | Good fit | Yes |
| 4 | Mistral Nemo 12B Instruct | 12.2B | Apache 2.0 | Q6_K | 12 GB | 72 | Excellent fit | Yes |
| 5 | Qwen3 14B | 14.8B | Apache 2.0 | Q6_K | 13 GB | 60 | Excellent fit | Yes |
| 6 | DeepSeek-R1-Distill-Qwen-14B | 14.8B | MIT | Q6_K | 14 GB | 60 | Excellent fit | Yes |
| 7 | Qwen2.5 Coder 32B Instruct | 32.8B | Apache 2.0 | Q4_K_M | 21 GB | 37 | Tight fit | Yes |
| 8 | Gemma 2 9B Instruct | 9.2B | Gemma Terms of Use | Q8_0 | 12 GB | 74 | Excellent fit | Yes |
| 9 | DeepSeek-Coder-V2-Lite Instruct | 15.7B | DeepSeek License | Q5_K_M | 13 GB | 425 | Excellent fit | Yes |
| 10 | StarCoder2 15B | 16B | BigCode OpenRAIL-M | Q8_0 | 17 GB | 43 | Good fit | Yes |
| 11 | Qwen2.5 32B Instruct | 32.8B | Apache 2.0 | Q4_K_M | 21 GB | 37 | Tight fit | Yes |
| 12 | Qwen2.5 Coder 7B Instruct | 7.6B | Apache 2.0 | F16 | 15 GB | 48 | Good fit | Yes |
| 13 | Llama 3.1 8B Instruct | 8B | Llama 3.1 Community License | Q8_0 | 9.5 GB | 85 | Excellent fit | Yes |
| 14 | Gemma 3 12B Instruct | 12.2B | Gemma Terms of Use | Q8_0 | 16 GB | 56 | Good fit | Yes |
| 15 | Qwen3 8B | 8.2B | Apache 2.0 | Q8_0 | 9.8 GB | 83 | Excellent fit | Yes |
| 16 | Qwen2.5 7B Instruct | 7.6B | Apache 2.0 | F16 | 15 GB | 48 | Good fit | Yes |
| 17 | DeepSeek-R1-Distill-Qwen-32B | 32.8B | MIT | Q4_K_M | 21 GB | 37 | Tight fit | Yes |
| 18 | Code Llama 7B Instruct | 6.7B | Llama 2 Community License | Q8_0 | 11 GB | 102 | Excellent fit | Yes |
| 19 | Gemma 3 27B Instruct | 27.4B | Gemma Terms of Use | Q4_K_M | 21 GB | 44 | Tight fit | Yes |
| 20 | Qwen3 30B-A3B | 30.5B | Apache 2.0 | Q4_K_M | 19 GB | 363 | Tight fit | Yes |
| 21 | Mistral 7B Instruct v0.3 | 7.25B | Apache 2.0 | F16 | 15 GB | 50 | Good fit | Yes |
| 22 | Phi-3.5 Mini Instruct | 3.8B | MIT | Q8_0 | 7.3 GB | 179 | Excellent fit | Yes |
| 23 | Gemma 3 4B Instruct | 4.3B | Gemma Terms of Use | F16 | 9.9 GB | 84 | Excellent fit | Yes |
| 24 | Llama 3.2 3B Instruct | 3.2B | Llama 3.2 Community License | F16 | 7.3 GB | 113 | Excellent fit | Yes |
| 25 | Qwen2.5 72B Instruct | 72.7B | Qwen License | Q4_K_M | 45 GB | 16 | Runs with CPU offload | No |
| 26 | Llama 3.3 70B Instruct | 70.6B | Llama 3.3 Community License | Q4_K_M | 44 GB | 17 | Runs with CPU offload | No |
| 27 | SmolLM2 1.7B Instruct | 1.7B | Apache 2.0 | F16 | 5.2 GB | 213 | Excellent fit | Yes |
| 28 | Llama 3.2 1B Instruct | 1.24B | Llama 3.2 Community License | F16 | 3.0 GB | 292 | Excellent fit | Yes |
| 29 | Mixtral 8x7B Instruct | 46.7B | Apache 2.0 | Q4_K_M | 29 GB | 93 | Runs with CPU offload | No |
How this page is produced
Memory is calculated from each model’s published geometry against this machine’s usable capacity. Where the vendor publishes a memory bandwidth figure, an order-of-magnitude token rate is derived from it and labelled as an estimate.
- · 26 of these can also be fine-tuned on this hardware.
Check your own machine
ModelLM can read your actual GPU, VRAM and RAM and size every model against it.
Detect my hardwareRelated