What LLMs Can NVIDIA RTX 4070 Ti Super (16GB) Run?

Hardware capability guide for NVIDIA RTX 4070 Ti Super: 16GB memory with 672.3 GB/s bandwidth across single-card and multi-GPU deployments.

NVIDIA RTX 4070 Ti Super Hardware Specifications

VRAM / Card
16 GB
Mem Bandwidth
672.3 GB/s
FP16 Compute
40.1 TFLOPS
Architecture
Ada Lovelace
Release Year
2024
Available VRAM
16 GB
Total Bandwidth
672.3 GB/s
Fits FP16 / FP8
15
Fits w/ INT4
32 / 85
DeepSeek-R1-Distill-Qwen-14B
14B • DeepSeek
Fits INT4
FP16
32.27 GB
FP8
18 GB
INT4
10.86 GB
~31 tok/s (INT4)Open VRAM Calculator
DeepSeek-R1-Distill-Qwen-7B
7B • DeepSeek
Fits FP8+
FP16
18.27 GB
FP8
11 GB
INT4
7.36 GB
~15 tok/s (FP8)Open VRAM Calculator
DeepSeek-R1-Distill-Qwen-1.5B
1.5B • DeepSeek
Fits FP16+
FP16
7.27 GB
FP8
5.5 GB
INT4
4.61 GB
~8 tok/s (FP16)Open VRAM Calculator
DeepSeek-R1-Distill-Llama-8B
8B • DeepSeek
Fits FP8+
FP16
20.27 GB
FP8
12 GB
INT4
7.86 GB
~15 tok/s (FP8)Open VRAM Calculator
Llama 3 8B
8B • Llama 3
Fits FP8+
FP16
17.81 GB
FP8
9.31 GB
INT4
5.06 GB
~71 tok/s (FP8)Open VRAM Calculator
Llama 3.1 8B
8B • Llama 3
Fits FP8+
FP16
17.81 GB
FP8
9.31 GB
INT4
5.06 GB
~71 tok/s (FP8)Open VRAM Calculator
Llama 3.2 1B
1B • Llama 3
Fits FP16+
FP16
2.55 GB
FP8
1.43 GB
INT4
0.86 GB
~286 tok/s (FP16)Open VRAM Calculator
Llama 3.2 3B
3B • Llama 3
Fits FP16+
FP16
7.23 GB
FP8
3.79 GB
INT4
2.07 GB
~95 tok/s (FP16)Open VRAM Calculator
Llama 3.2 11B
11B • Llama 3
Fits FP8+
FP16
24.11 GB
FP8
12.61 GB
INT4
6.86 GB
~52 tok/s (FP8)Open VRAM Calculator
Qwen3-30B-A3BMoE 3B
30B • Qwen 3
Fits INT4
FP16
61.1 GB
FP8
30.73 GB
INT4
15.54 GB
~381 tok/s (INT4)Open VRAM Calculator
Qwen2.5-14B
14B • Qwen 2
Fits INT4
FP16
30.66 GB
FP8
16.04 GB
INT4
8.72 GB
~82 tok/s (INT4)Open VRAM Calculator
Qwen3-14B
14B • Qwen 3
Fits INT4
FP16
30.66 GB
FP8
16.04 GB
INT4
8.72 GB
~82 tok/s (INT4)Open VRAM Calculator
Qwen2-14B
14B • Qwen 2
Fits INT4
FP16
30.66 GB
FP8
16.04 GB
INT4
8.72 GB
~82 tok/s (INT4)Open VRAM Calculator
Qwen3-8B
8B • Qwen 3
Fits FP8+
FP16
17.93 GB
FP8
9.37 GB
INT4
5.09 GB
~71 tok/s (FP8)Open VRAM Calculator
Qwen2.5-7B
7B • Qwen 2
Fits FP16+
FP16
15.21 GB
FP8
7.96 GB
INT4
4.33 GB
~41 tok/s (FP16)Open VRAM Calculator
Qwen2-7B
7B • Qwen 2
Fits FP16+
FP16
15.21 GB
FP8
7.96 GB
INT4
4.33 GB
~41 tok/s (FP16)Open VRAM Calculator
Qwen3-4B
4B • Qwen 3
Fits FP16+
FP16
9 GB
FP8
4.72 GB
INT4
2.58 GB
~71 tok/s (FP16)Open VRAM Calculator
Qwen2.5-3B
3B • Qwen 2
Fits FP16+
FP16
6.57 GB
FP8
3.46 GB
INT4
1.91 GB
~95 tok/s (FP16)Open VRAM Calculator
Qwen2.5-1.5B
1.5B • Qwen 2
Fits FP16+
FP16
3.49 GB
FP8
1.89 GB
INT4
1.1 GB
~190 tok/s (FP16)Open VRAM Calculator
Qwen2.5-0.5B
0.5B • Qwen 2
Fits FP16+
FP16
1.36 GB
FP8
0.83 GB
INT4
0.57 GB
~571 tok/s (FP16)Open VRAM Calculator
Qwen3-0.6B
0.6B • Qwen 3
Fits FP16+
FP16
1.94 GB
FP8
1.12 GB
INT4
0.71 GB
~476 tok/s (FP16)Open VRAM Calculator
Qwen3-1.7B
1.7B • Qwen 3
Fits FP16+
FP16
4.14 GB
FP8
2.22 GB
INT4
1.26 GB
~168 tok/s (FP16)Open VRAM Calculator
Mistral-Nemo-12B
12B • Mistral
Fits FP8+
FP16
26.46 GB
FP8
13.83 GB
INT4
7.52 GB
~48 tok/s (FP8)Open VRAM Calculator
Mistral-7B-v0.3
7B • Mistral
Fits FP16+
FP16
15.71 GB
FP8
8.21 GB
INT4
4.46 GB
~41 tok/s (FP16)Open VRAM Calculator
Gemma-2-9B
9B • Gemma
Fits FP8+
FP16
21.53 GB
FP8
11.22 GB
INT4
6.06 GB
~63 tok/s (FP8)Open VRAM Calculator
Gemma-2-2B
2B • Gemma
Fits FP16+
FP16
5.11 GB
FP8
2.71 GB
INT4
1.5 GB
~143 tok/s (FP16)Open VRAM Calculator
Phi-3-14B
14B • Phi
Fits INT4
FP16
30.97 GB
FP8
16.19 GB
INT4
8.8 GB
~82 tok/s (INT4)Open VRAM Calculator
Phi-3-7B
7B • Phi
Fits FP16+
FP16
15.71 GB
FP8
8.21 GB
INT4
4.46 GB
~41 tok/s (FP16)Open VRAM Calculator
Phi-3-3.8B
3.8B • Phi
Fits FP16+
FP16
11.02 GB
FP8
5.72 GB
INT4
3.07 GB
~75 tok/s (FP16)Open VRAM Calculator
CodeLlama-13B
13B • CodeLlama
Fits INT4
FP16
33.56 GB
FP8
17.44 GB
INT4
9.37 GB
~88 tok/s (INT4)Open VRAM Calculator
CodeLlama-7B
7B • CodeLlama
Fits FP8+
FP16
18.71 GB
FP8
9.71 GB
INT4
5.21 GB
~82 tok/s (FP8)Open VRAM Calculator
GPT-OSS-20BMoE 3.6B
21B • GPT-OSS
Fits INT4
FP16
43.15 GB
FP8
21.78 GB
INT4
11.09 GB
~317 tok/s (INT4)Open VRAM Calculator

NVIDIA RTX 4070 Ti Super 1x / 2x / 4x Multi-GPU LLM Compatibility Matrix

Precomputed at 8K context, batch size 1 across key benchmark models.

LLM ModelFP8 VRAM Req1× (16GB)2× (32GB)4× (64GB)
DeepSeek-R1(671B)675 GBNeeds 22×Needs 22×Needs 22×
DeepSeek-R1-Distill-Qwen-32B(32B)36 GBNeeds 2×INT4FP8 / INT4
Llama 4 Scout(109B)111.21 GBNeeds 4×Needs 4×INT4
Llama 3.1 8B(8B)9.31 GBFP8 / INT4FP16 / FP8 / INT4FP16 / FP8 / INT4
Llama 3.1 70B(70B)78.31 GBNeeds 3×Needs 3×INT4
Qwen3-235B-A22B(235B)237.95 GBNeeds 8×Needs 8×Needs 8×
Qwen3-32B(32B)36.23 GBNeeds 2×INT4FP8 / INT4
Qwen3-14B(14B)16.04 GBINT4FP16 / FP8 / INT4FP16 / FP8 / INT4
Mixtral-8x7B(46.7B)48.5 GBNeeds 2×INT4FP8 / INT4
GPT-OSS-20B(21B)21.78 GBINT4FP8 / INT4FP16 / FP8 / INT4

Dedicated Compatibility Pages for All 22 GPUs

NVIDIA RTX 4070 Ti Super Local LLM Deployment FAQ

tools.gpuCompatibility.gpuDetail.faq.q1▾
With 16GB of memory on a single NVIDIA RTX 4070 Ti Super, 32 benchmark models in our registry fit cleanly in FP16, FP8, or INT4 at 8K context.
What generation speed can I expect from NVIDIA RTX 4070 Ti Super?▾
With 672.3 GB/s memory bandwidth and 16GB memory, single-stream decode speed scales inversely with active model weight size read per token.
When should I scale to 2x or 4x NVIDIA RTX 4070 Ti Super?▾
Multi-GPU tensor parallelism becomes necessary when model weights plus KV cache exceed 16GB or when serving high-concurrency batches.

Explore Related LLM & GPU Tools

Browse full AI toolkit directory →