Consumer · GDDR7
RTX 5080 16 GB: which LLMs can it run?
On its own, it loads up to gpt-oss-20b in Q4. With several cards, memory adds up and opens bigger models.
Specifications
Memory16 GB
Bandwidth960 GB/s
FP16 compute~225 TF
Power360 W
Indicative price€1,200
Cloud rental€0.35/h
Against cards at the same price
Models and expected speed
| Model | Q4 | Q8 | FP16 |
|---|---|---|---|
| Qwen3 4B | ✓ ~207 tok/s | ✓ ~120 tok/s | ✓ ~65 tok/s |
| Llama 3.1 8B | ✓ ~107 tok/s | ✓ ~61 tok/s | 2 × · ~49 tok/s |
| Qwen3 8B | ✓ ~104 tok/s | ✓ ~60 tok/s | 2 × · ~48 tok/s |
| Gemma 3 12B | ✓ ~72 tok/s | ✓ ~41 tok/s | 2 × · ~33 tok/s |
| Mistral Nemo 12B | ✓ ~71 tok/s | ✓ ~40 tok/s | 2 × · ~32 tok/s |
| Phi-4 14B | ✓ ~59 tok/s | 2 × · ~50 tok/s | 3 × · ~40 tok/s |
| Qwen3 14B | ✓ ~58 tok/s | 2 × · ~50 tok/s | 3 × · ~40 tok/s |
| gpt-oss-20b | ✓ ~229 tok/s | 2 × · ~200 tok/s | 3 × · ~162 tok/s |
| Mistral Small 24B | 2 × · ~54 tok/s | 2 × · ~31 tok/s | 4 × · ~33 tok/s |
| Gemma 3 27B | 2 × · ~48 tok/s | 3 × · ~41 tok/s | 4 × · ~29 tok/s |
| Qwen3.8 27B | 2 × · ~47 tok/s | 3 × · ~40 tok/s | 4 × · ~28 tok/s |
| Qwen3 30B-A3B | 2 × · ~372 tok/s | 3 × · ~326 tok/s | ✗ |
| Qwen3 32B | 2 × · ~40 tok/s | 3 × · ~34 tok/s | ✗ |
| Qwen2.5 Coder 32B | 2 × · ~40 tok/s | 3 × · ~34 tok/s | ✗ |
| DeepSeek R1 Distill 32B | 2 × · ~40 tok/s | 3 × · ~34 tok/s | ✗ |
| Mixtral 8x7B | 2 × · ~100 tok/s | 4 × · ~115 tok/s | ✗ |
| Llama 3.3 70B | 4 × · ~37 tok/s | ✗ | ✗ |
| GLM-4.5-Air 106B | ✗ | ✗ | ✗ |
| gpt-oss-120b | ✗ | ✗ | ✗ |
| Mistral Large 123B | ✗ | ✗ | ✗ |
| Qwen3 235B-A22B | ✗ | ✗ | ✗ |
| GLM-4.5 355B | ✗ | ✗ | ✗ |
| Llama 3.1 405B | ✗ | ✗ | ✗ |
| DeepSeek R1 671B | ✗ | ✗ | ✗ |
| DeepSeek V3.1 671B | ✗ | ✗ | ✗ |
| Kimi K2 1T | ✗ | ✗ | ✗ |
8k tokens, one user. “2 ×” = number of cards needed, speed with tensor parallelism (vLLM). ✗ = doesn’t fit, even with 4 cards.