Skip to content

NVIDIA GeForce RTX 4070 Ti

NVIDIA · 12GBGDDR6X · Can run 63 models

BuyAmazon
ManufacturerNVIDIA
VRAM12 GB
Memory TypeGDDR6X
ArchitectureAda Lovelace
CUDA Cores7,680
Tensor Cores240
Bandwidth504 GB/s
TDP285W
MSRP$799
ReleasedJan 5, 2023

AI Notes

The RTX 4070 Ti provides solid AI inference capability with 12GB of GDDR6X VRAM. It can run 7B-parameter models at full precision and 13B models with quantization. The 12GB VRAM limit means larger models require aggressive quantization or offloading to system RAM.

Compatible Models

ModelParametersBest QuantVRAM UsedFitEst. Speed
Qwen 3 0.6B600MQ4_K_M2.5 GBRuns~202 tok/s
Qwen 3.5 0.8B800MQ4_K_M1.5 GBRuns~336 tok/s
Gemma 3 1B1BQ8_02 GBRuns~252 tok/s
Llama 3.2 1B1BQ8_03 GBRuns~168 tok/s
DeepSeek R1 1.5B1.5BQ8_03 GBRuns~168 tok/s
SmolLM2 1.7B1.7BQ8_02.7 GBRuns~187 tok/s
Gemma 2 2B2BQ8_04 GBRuns~126 tok/s
Gemma 3n E2B2BQ4_K_M3.3 GBRuns~153 tok/s
Gemma 4 E2B2BQ4_K_M4 GBRuns~126 tok/s
Qwen 3.5 2B2BQ4_K_M3 GBRuns~168 tok/s
Llama 3.2 3B3BQ8_05 GBRuns~101 tok/s
StarCoder2 3B3BQ4_K_M3.5 GBRuns~144 tok/s
Phi-3 Mini 3.8B3.8BQ8_05.8 GBRuns~87 tok/s
Phi-4 Mini 3.8B3.8BQ4_K_M4.5 GBRuns~112 tok/s
Gemma 3 4B4BQ4_K_M5 GBRuns~101 tok/s
Gemma 3n E4B4BQ4_K_M4.5 GBRuns~112 tok/s
Gemma 4 E4B4BQ4_K_M6 GBRuns~84 tok/s
Qwen 3 4B4BQ4_K_M4.5 GBRuns~112 tok/s
Qwen 3.5 4B4BQ4_K_M4.5 GBRuns~112 tok/s
Yi 1.5 6B6BQ4_K_M5 GBRuns~101 tok/s
Codestral Mamba 7B7BQ4_K_M6.9 GBRuns~73 tok/s
DeepSeek R1 7B7BQ8_09 GBRuns~56 tok/s
Falcon 3 7B7BQ4_K_M6.8 GBRuns~74 tok/s
InternLM 2.5 7B7BQ4_K_M5.5 GBRuns~92 tok/s
Mistral 7B7BQ8_09 GBRuns~56 tok/s
OpenChat 3.5 7B7BQ4_K_M6.9 GBRuns~73 tok/s
Qwen 2.5 7B7BQ8_09 GBRuns~56 tok/s
Qwen 2.5 Coder 7B7BQ8_09 GBRuns~56 tok/s
Qwen 2.5 VL 7B7BQ4_K_M7 GBRuns~72 tok/s
StarCoder2 7B7BQ4_K_M5.5 GBRuns~92 tok/s
WizardLM 2 7B7BQ4_K_M6.9 GBRuns~73 tok/s
Aya Expanse 8B8BQ4_K_M6.5 GBRuns~78 tok/s
Cogito 8B8BQ4_K_M7.5 GBRuns~67 tok/s
DeepSeek R1 8B8BQ4_K_M7.5 GBRuns~67 tok/s
Dolphin 3 8B8BQ4_K_M6 GBRuns~84 tok/s
Granite 3.3 8B8BQ8_010 GBRuns~50 tok/s
Llama 3.1 8B8BQ8_010 GBRuns~50 tok/s
Nemotron 3 Nano 8B8BQ4_K_M7.5 GBRuns~67 tok/s
Nous Hermes 2 8B8BQ4_K_M6 GBRuns~84 tok/s
Qwen 3 8B8BQ4_K_M7.5 GBRuns~67 tok/s
Qwen 3.5 9B9BQ4_K_M7.5 GBRuns~67 tok/s
Yi 1.5 9B9BQ4_K_M6.5 GBRuns~78 tok/s
Yi Coder 9B9BQ4_K_M8 GBRuns~63 tok/s
Falcon 3 10B10BQ4_K_M8.5 GBRuns~59 tok/s
Llama 3.2 Vision 11B11BQ4_K_M8.5 GBRuns~59 tok/s
Mistral Nemo 12B12BQ4_K_M9.5 GBRuns~53 tok/s
DeepSeek R1 14B14BQ4_K_M9.9 GBRuns~51 tok/s
Phi-4 14B14BQ4_K_M9.9 GBRuns~51 tok/s
Qwen 2.5 14B14BQ4_K_M9.9 GBRuns~51 tok/s
Gemma 2 9B9BQ8_011 GBRuns (tight)~46 tok/s
Gemma 3 12B12BQ4_K_M10.5 GBRuns (tight)~48 tok/s
Phi-4 Reasoning 14B14BQ4_K_M11 GBRuns (tight)~46 tok/s
Qwen 2.5 Coder 14B14BQ4_K_M12 GBCPU Offload~13 tok/s
Qwen 3 14B14BQ4_K_M12 GBCPU Offload~13 tok/s
StarCoder2 15B15BQ8_017 GBCPU Offload~9 tok/s
InternLM 2.5 20B20BQ4_K_M12 GBCPU Offload~13 tok/s
gpt-oss 20B21BMXFP415 GBCPU Offload~10 tok/s
Codestral 22B22BQ4_K_M14.7 GBCPU Offload~10 tok/s
Devstral 24B24BQ4_K_M17 GBCPU Offload~9 tok/s
Magistral Small 24B24BQ4_K_M17 GBCPU Offload~9 tok/s
Mistral Small 3.1 24B24BQ4_K_M18 GBCPU Offload~8 tok/s
Gemma 2 27B27BQ4_K_M17.7 GBCPU Offload~8 tok/s
Qwen 3.5 35B A3B35BQ4_K_M12 GBCPU Offload~13 tok/s
51 model(s) are too large for this hardware.