GeForce RTX 4070 Ti
12 GB GDDR6X · 504 GB/s · 285 W
RTX 4070 Ti 12 GB · 64 GB RAM → Qwen3 14B
GPU · NVIDIA
RTX 4070 Ti for local LLMs: which models run?
With 12 GB of VRAM and 504 GB/s of bandwidth, the RTX 4070 Ti runs models up to about Qwen3 14B entirely on the GPU.
- VRAM12 GB GDDR6X
- Bandwidth504 GB/s
- Board power285 W
- ArchitectureAda Lovelace
Specifications
- VRAM
- 12 GB
- Memory type
- GDDR6X
- Memory bus
- 192-bit
- Bandwidth
- 504 GB/s
- Board power
- 285 W
- Backends
- CUDA, Vulkan
- Interface
- PCIe 4.0 x16
- Released
- January 2023
- Segment
- Consumer
- FP16 compute
- Not available
Official specData updated 24 days agoSource: NVIDIA specifications
What the RTX 4070 Ti runs
| Qwen2.5 7B InstructQ8_0 · 8K~9.2 GB · 43–58 tok/s estimate | Excellent |
|---|---|
| Mistral 7B Instruct v0.3Q8_0 · 8K~9.2 GB · 43–58 tok/s estimate | Excellent |
| Gemma 3 4B ITQ8_0 · 8K~5.4 GB · 77–108 tok/s estimate | Excellent |
| Qwen3 4BQ8_0 · 8K~6.2 GB · 68–95 tok/s estimate | Excellent |
| Phi-4 Mini Instruct 3.8BQ8_0 · 8K~6 GB · 73–101 tok/s estimate | Excellent |
| Llama 3.2 3B InstructQ8_0 · 8K~5.1 GB · 86–119 tok/s estimate | Excellent |
| Qwen3 1.7BQ8_0 · 8K~3.7 GB · 132–193 tok/s estimate | Excellent |
| Llama 3.2 1B InstructQ8_0 · 8K~2.5 GB · 213–308 tok/s estimate | Excellent |
| Qwen3 0.6BQ8_0 · 8K~2.5 GB · 221–359 tok/s estimate | Excellent |
| Qwen3 14BQ4_K_M · 8K~11 GB · 37–49 tok/s estimate | Compatible |
| DeepSeek R1 Distill Qwen 14BIQ4_XS · 8K~10.4 GB · 39–53 tok/s estimate | Compatible |
| Qwen2.5 14B InstructIQ4_XS · 8K~10.4 GB · 39–53 tok/s estimate | Compatible |
| Phi-4 14BIQ4_XS · 8K~10.2 GB · 40–54 tok/s estimate | Compatible |
| Mistral Nemo 12B InstructQ5_K_M · 8K~10.7 GB · 38–51 tok/s estimate | Compatible |
| Gemma 3 12B ITQ5_K_M · 8K~10.2 GB · 39–53 tok/s estimate | Compatible |
| Gemma 2 9B ITQ6_K · 8K~10.5 GB · 40–55 tok/s estimate | Compatible |
| Qwen3 8BQ8_0 · 8K~10.5 GB · 38–51 tok/s estimate | Compatible |
| Llama 3.1 8B InstructQ8_0 · 8K~10.2 GB · 40–53 tok/s estimate | Compatible |
| DeepSeek R1 Distill Llama 8BQ8_0 · 8K~10.2 GB · 40–53 tok/s estimate | Compatible |
| Mixtral 8x7B InstructQ3_K_M · 8K~21.6 GB · 6.5–12 tok/s estimate | CPU/GPU offload |
| Qwen3 32BQ3_K_M · 8K~18.6 GB · 3–5.5 tok/s estimate | CPU/GPU offload |
| DeepSeek R1 Distill Qwen 32BQ3_K_M · 8K~18.5 GB · 3–5.5 tok/s estimate | CPU/GPU offload |
| Qwen2.5 32B InstructQ3_K_M · 8K~18.5 GB · 3–5.5 tok/s estimate | CPU/GPU offload |
| Qwen2.5 Coder 32B InstructQ3_K_M · 8K~18.5 GB · 3–5.5 tok/s estimate | CPU/GPU offload |
| QwQ 32BQ3_K_M · 8K~18.5 GB · 3–5.5 tok/s estimate | CPU/GPU offload |
| Qwen3 30B-A3BQ3_K_M · 8K~15.4 GB · 29–57 tok/s estimate | CPU/GPU offload |
| Qwen3 Coder 30B-A3B InstructQ3_K_M · 8K~16 GB · 26–51 tok/s estimate | CPU/GPU offload |
| Gemma 3 27B ITQ3_K_M · 8K~15.3 GB · 5–9 tok/s estimate | CPU/GPU offload |
| Gemma 2 27B ITQ3_K_M · 8K~16.5 GB · 4–7.5 tok/s estimate | CPU/GPU offload |
| Mistral Small 3.2 24B InstructQ3_K_M · 8K~13.3 GB · 8.5–16 tok/s estimate | CPU/GPU offload |
| gpt-oss-20bMXFP4 · 8K~13 GB · 46–88 tok/s estimate | CPU/GPU offload |
| Llama 4 Scout 17B-16E InstructQ3_K_M · 8K~55.7 GB · 2.5–5 tok/s estimate | Possible but slow |
| Qwen2.5 72B InstructQ3_K_M · 8K~38.2 GB · 0.7–1.5 tok/s estimate | Not recommended |
| Llama 3.3 70B InstructQ3_K_M · 8K~37 GB · 0.8–1.5 tok/s estimate | Not recommended |
| DeepSeek R1 Distill Llama 70BQ3_K_M · 8K~37 GB · 0.8–1.5 tok/s estimate | Not recommended |
| DeepSeek R1 (671B)Q8_0 · 8K~699 GB · No data | Incompatible |
| Qwen3 235B-A22BQ8_0 · 8K~247 GB · No data | Incompatible |
| gpt-oss-120bMXFP4 · 8K~63.2 GB · No data | Incompatible |
For each model, the highest-precision quantization that runs entirely in fast memory; otherwise the one with the best result. Speeds are ranges; "estimate" unless a benchmark exists for this exact hardware.
EstimateData updated 24 days agoSource: Run LLM Locally engine
Limits in plain words
12 GB of VRAM holds the weights and KV cache of models up to about Qwen3 14B at Q4_K_M. Bigger models spill into system RAM and slow down sharply: generation speed is then limited by RAM bandwidth, not by the GPU.
Where to buy a RTX 4070 Ti
Prices and links follow the market you choose.
No current offer imported for this market: the link opens a marketplace search.
Frequently asked questions
Which LLMs can the RTX 4070 Ti run?
Entirely on the GPU: Qwen3 14B (Q4_K_M), DeepSeek R1 Distill Qwen 14B (IQ4_XS), Qwen2.5 14B Instruct (IQ4_XS), Phi-4 14B (IQ4_XS), Mistral Nemo 12B Instruct (Q5_K_M), Gemma 3 12B IT (Q5_K_M), Gemma 2 9B IT (Q6_K), Qwen3 8B (Q8_0). Larger models need CPU/GPU offload.
Can the RTX 4070 Ti run a 70B model?
Not recommended: a 70B model at Q4_K_M needs about 45.1 GB; the RTX 4070 Ti has 12 GB.
How is the speed estimated?
Token generation reads every active weight once per token, so it is limited by memory bandwidth (504 GB/s here). We apply an efficiency range per backend and label the result as an estimate until real benchmarks exist.