NVIDIA RTX 5060 Ti 16GB
The NVIDIA RTX 5060 Ti 16GB has 16 GB VRAM and 448 GB/s memory bandwidth. It can run 40 of our 70 tracked models natively in VRAM at 8k context.
The NVIDIA RTX 5060 Ti 16GB is an unusual SKU — it has more VRAM than the RTX 5070 (16GB vs 12GB) but on a narrower 128-bit bus (448 GB/s). With 4,608 CUDA cores, it can hold larger models in memory than the 5070 but runs them slower due to bandwidth constraints. A practical option for budget local LLM experimentation with 14B models.
The NVIDIA RTX 5060 Ti 16GB is a consumer-grade NVIDIA GPU based on the Blackwell architecture. Released in 2025. It features 16 GB of GDDR7 at 448 GB/s memory bandwidth. Full llama.cpp and Ollama support out of the box. CUDA 12.x recommended; driver ≥ 525 required.
For local LLM inference, this GPU runs 40 of the 70 models we track natively in VRAM at 8K context. The largest model it handles in VRAM is Qwen 3.5 35B-A3B (MoE) (499.3 t/s at Q2_K). It handles smaller models up to ~7-14B at reasonable precision, with some 27-32B models fitting at lower quantization. On Qwen 3.6 27B, it achieves approximately 38.6 tokens per second at Q3_K_M quantization. An additional 9 models fit with CPU offload — slower but usable.
NVIDIA's CUDA ecosystem provides broad out-of-the-box support across llama.cpp, Ollama, vLLM, and TensorRT-LLM. Among consumer GPUs, it sits above NVIDIA RTX 5060 and NVIDIA RTX 3060 12GB in performance, but below AMD Radeon RX 6800 XT.
| Vendor | NVIDIA |
| Architecture | Blackwell |
| VRAM | 16 GB |
| Memory type | GDDR7 |
| Memory bandwidth | 448 GB/s |
| Compute backend | CUDA |
| Tier | Consumer |
| Released | 2025 |
| Models (native) | 40 / 70 |
| Models (offload) | 9 / 70 |
Models this GPU runs natively in VRAM (40)
- Qwen 3.5 35B-A3B (MoE)35B · MMLU-Pro —Q2_K · ~499.3 t/s
- Yi 1.5 34B Chat34.4B · MMLU-Pro 37.0Q2_K · ~39.6 t/s
- Qwen3 32B32.8B · MMLU-Pro —Q2_K · ~41.5 t/s
- Qwen 2.5 32B Instruct32.5B · MMLU-Pro 55.1Q2_K · ~41.9 t/s
- Qwen 2.5 Coder 32B Instruct32.5B · MMLU-Pro 50.4Q2_K · ~41.9 t/s
- DeepSeek R1 Distill Qwen 32B32.5B · MMLU-Pro 65.0Q2_K · ~41.9 t/s
- Nemotron 3 Nano 30B32B · MMLU-Pro —Q2_K · ~499.3 t/s
- Gemma 4 31B31B · MMLU-Pro —Q2_K · ~43.9 t/s
- Qwen3 30B-A3B (MoE)30B · MMLU-Pro —Q2_K · ~499.3 t/s
- Gemma 2 27B Instruct27.2B · MMLU-Pro 38.0Q2_K · ~50.1 t/s
- Gemma 3 27B Instruct27B · MMLU-Pro —Q3_K_M · ~38.6 t/s
- Qwen 3.6 27B27B · MMLU-Pro —Q3_K_M · ~38.6 t/s
- Gemma 4 26B (MoE)26B · MMLU-Pro —Q3_K_M · ~301.6 t/s
- Mistral Small 3.1 24B Instruct24B · MMLU-Pro —NVFP4 · ~37.3 t/s
- Mistral Small 22B22.2B · MMLU-Pro 49.2NVFP4 · ~40.4 t/s
- GPT-OSS 20B21B · MMLU-Pro —NVFP4 · ~246.4 t/s
- Qwen3 14B14.8B · MMLU-Pro —NVFP4 · ~60.5 t/s
- Qwen 2.5 14B Instruct14.7B · MMLU-Pro 51.2NVFP4 · ~61 t/s
- Phi-4 14B Instruct14B · MMLU-Pro 56.1NVFP4 · ~64 t/s
- Mistral Nemo 12B Instruct12.2B · MMLU-Pro 35.6NVFP4 · ~73.4 t/s
- Gemma 3 12B Instruct12.2B · MMLU-Pro —NVFP4 · ~73.4 t/s
- Gemma 2 9B Instruct9.2B · MMLU-Pro 32.0NVFP4 · ~97.4 t/s
- Llama 3.1 8B Instruct8B · MMLU-Pro 37.5NVFP4 · ~112 t/s
- DeepSeek R1 Distill Llama 8B8B · MMLU-Pro 41.0NVFP4 · ~112 t/s
- Qwen3 8B8B · MMLU-Pro —NVFP4 · ~112 t/s
- Qwen 2.5 7B Instruct7.6B · MMLU-Pro 36.5NVFP4 · ~117.9 t/s
- Mistral 7B Instruct v0.37.25B · MMLU-Pro 30.0NVFP4 · ~123.6 t/s
- Gemma 3 4B Instruct4B · MMLU-Pro —BF16 · ~56 t/s
- Gemma 4 E4B4B · MMLU-Pro —BF16 · ~56 t/s
- Phi-3.5 Mini Instruct3.8B · MMLU-Pro 35.6BF16 · ~58.9 t/s
- Llama 3.2 3B Instruct3.2B · MMLU-Pro 24.0BF16 · ~70 t/s
- Qwen 2.5 3B Instruct3.1B · MMLU-Pro 32.4FP32 · ~36.1 t/s
- Gemma 2 2B Instruct2.6B · MMLU-Pro 17.8FP32 · ~43.1 t/s
- Gemma 4 E2B2B · MMLU-Pro —FP32 · ~56 t/s
- SmolLM2 1.7B Instruct1.7B · MMLU-Pro 19.0FP32 · ~65.9 t/s
- Qwen 2.5 1.5B Instruct1.5B · MMLU-Pro 16.8FP32 · ~74.7 t/s
- Llama 3.2 1B Instruct1.24B · MMLU-Pro 12.5FP32 · ~90.3 t/s
- Gemma 3 1B Instruct1B · MMLU-Pro —FP32 · ~112 t/s
- Qwen 2.5 0.5B Instruct0.5B · MMLU-Pro 10.0FP32 · ~224 t/s
- SmolLM2 360M Instruct0.36B · MMLU-Pro 8.0FP32 · ~311.1 t/s
Models that fit with CPU offload (9)
These use system RAM for layers that don't fit in VRAM — expect much slower inference.
- GLM-4.5 Air 106B106B · MMLU-Pro —Q2_K · ~28.4 t/s
- GLM-4.6V 106B106B · MMLU-Pro —Q2_K · ~28.4 t/s
- Qwen 2.5 72B Instruct72B · MMLU-Pro 58.1Q3_K_M · ~3.6 t/s
- Llama 3.3 70B Instruct70B · MMLU-Pro 68.9Q3_K_M · ~3.7 t/s
- DeepSeek R1 Distill Llama 70B70B · MMLU-Pro 70.0Q3_K_M · ~3.7 t/s
- Llama 3.1 70B Instruct70B · MMLU-Pro 66.4Q3_K_M · ~3.7 t/s
- Mixtral 8x7B Instruct v0.146.7B · MMLU-Pro 29.7NVFP4 · ~17.4 t/s
- Command-R 35B35B · MMLU-Pro 33.0NVFP4 · ~6.4 t/s
- Qwen 3.6 35B35B · MMLU-Pro —NVFP4 · ~6.4 t/s
Too large for this GPU (21)
- Mixtral 8x22B Instruct v0.1
- Llama 3.1 405B Instruct
- DeepSeek V3 671B
- DeepSeek R1 671B
- Llama 4 Scout 109B
- Llama 4 Maverick 400B
- Qwen3 235B-A22B (MoE)
- MiniMax M1 456B
- GPT-OSS 120B
- GLM-4.5 355B
- GLM-4.6 355B
- GLM-4.7 358B
- Qwen 3.5 122B-A10B (MoE)
- MiniMax M2.5 229B
- GLM-5 744B
- MiniMax M2.7 229B
- Nemotron 3 Super 120B
- Kimi K2.6
- GLM-5.1 754B
- DeepSeek V4 Pro 1.6T
- DeepSeek V4 Flash 284B
Frequently asked questions
- How much VRAM does the NVIDIA RTX 5060 Ti 16GB have?
- The NVIDIA RTX 5060 Ti 16GB has 16 GB of GDDR7 with 448 GB/s memory bandwidth.
- What LLMs can the NVIDIA RTX 5060 Ti 16GB run locally?
- The NVIDIA RTX 5060 Ti 16GB can run 40 of the 70 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.1 8B Instruct at NVFP4, Llama 3.2 3B Instruct at BF16, Llama 3.2 1B Instruct at FP32.
- Can the NVIDIA RTX 5060 Ti 16GB run Llama 3.3 70B Instruct?
- The NVIDIA RTX 5060 Ti 16GB can run Llama 3.3 70B Instruct with CPU offload at Q3_K_M quantization, but inference will be slower than native VRAM execution.
- Can the NVIDIA RTX 5060 Ti 16GB run Qwen 3.6 27B?
- Yes. The NVIDIA RTX 5060 Ti 16GB runs Qwen 3.6 27B natively in VRAM at Q3_K_M quantization, achieving approximately 38.6 tokens per second.
- Can the NVIDIA RTX 5060 Ti 16GB run Llama 3.1 8B Instruct?
- Yes. The NVIDIA RTX 5060 Ti 16GB runs Llama 3.1 8B Instruct natively in VRAM at NVFP4 quantization, achieving approximately 112 tokens per second.