Skip to content

NVIDIA RTX 4000 Ada Generation

NVIDIA · 20GB GDDR6 · Can run 50 models

Buy Amazon
Manufacturer NVIDIA
VRAM 20 GB
Memory Type GDDR6
Architecture Ada Lovelace
CUDA Cores 6,144
Tensor Cores 192
Bandwidth 360 GB/s
TDP 130W
MSRP $1,250
Released Aug 8, 2023

AI Notes

The RTX 4000 Ada is a compact single-slot workstation GPU with 20GB of GDDR6 VRAM. It can run 13B-parameter models at high quantizations and 30B models at Q4, offering a good balance of AI capability and power efficiency. Its low 130W TDP makes it suitable for workstations with limited cooling capacity.

Compatible Models

Model Parameters Best Quant VRAM Used Fit Est. Speed
Qwen 3 0.6B 600M Q4_K_M 2.5 GB Runs ~144 tok/s
Gemma 3 1B 1B Q8_0 2 GB Runs ~180 tok/s
Llama 3.2 1B 1B Q8_0 3 GB Runs ~120 tok/s
DeepSeek R1 1.5B 1.5B Q8_0 3 GB Runs ~120 tok/s
Gemma 2 2B 2B Q8_0 4 GB Runs ~90 tok/s
Gemma 3n E2B 2B Q4_K_M 3.3 GB Runs ~109 tok/s
Llama 3.2 3B 3B Q8_0 5 GB Runs ~72 tok/s
Phi-3 Mini 3.8B 3.8B Q8_0 5.8 GB Runs ~62 tok/s
Phi-4 Mini 3.8B 3.8B Q4_K_M 4.5 GB Runs ~80 tok/s
Gemma 3 4B 4B Q4_K_M 5 GB Runs ~72 tok/s
Gemma 3n E4B 4B Q4_K_M 4.5 GB Runs ~80 tok/s
Qwen 3 4B 4B Q4_K_M 4.5 GB Runs ~80 tok/s
DeepSeek R1 7B 7B Q8_0 9 GB Runs ~40 tok/s
Falcon 3 7B 7B Q4_K_M 6.8 GB Runs ~53 tok/s
Mistral 7B 7B Q8_0 9 GB Runs ~40 tok/s
Qwen 2.5 7B 7B Q8_0 9 GB Runs ~40 tok/s
Qwen 2.5 Coder 7B 7B Q8_0 9 GB Runs ~40 tok/s
Qwen 2.5 VL 7B 7B Q4_K_M 7 GB Runs ~51 tok/s
Cogito 8B 8B Q4_K_M 7.5 GB Runs ~48 tok/s
DeepSeek R1 8B 8B Q4_K_M 7.5 GB Runs ~48 tok/s
Llama 3.1 8B 8B Q8_0 10 GB Runs ~36 tok/s
Nemotron 3 Nano 8B 8B Q4_K_M 7.5 GB Runs ~48 tok/s
Qwen 3 8B 8B Q4_K_M 7.5 GB Runs ~48 tok/s
Gemma 2 9B 9B Q8_0 11 GB Runs ~33 tok/s
Falcon 3 10B 10B Q4_K_M 8.5 GB Runs ~42 tok/s
Llama 3.2 Vision 11B 11B Q4_K_M 8.5 GB Runs ~42 tok/s
Gemma 3 12B 12B Q4_K_M 10.5 GB Runs ~34 tok/s
Mistral Nemo 12B 12B Q4_K_M 9.5 GB Runs ~38 tok/s
DeepSeek R1 14B 14B Q4_K_M 9.9 GB Runs ~36 tok/s
Phi-4 14B 14B Q4_K_M 9.9 GB Runs ~36 tok/s
Phi-4 Reasoning 14B 14B Q4_K_M 11 GB Runs ~33 tok/s
Qwen 2.5 14B 14B Q4_K_M 9.9 GB Runs ~36 tok/s
Qwen 2.5 Coder 14B 14B Q4_K_M 12 GB Runs ~30 tok/s
Qwen 3 14B 14B Q4_K_M 12 GB Runs ~30 tok/s
StarCoder2 15B 15B Q8_0 17 GB Runs ~21 tok/s
Codestral 22B 22B Q4_K_M 14.7 GB Runs ~24 tok/s
Devstral 24B 24B Q4_K_M 17 GB Runs ~21 tok/s
Magistral Small 24B 24B Q4_K_M 17 GB Runs ~21 tok/s
Mistral Small 3.1 24B 24B Q4_K_M 18 GB Runs (tight) ~20 tok/s
Gemma 2 27B 27B Q4_K_M 17.7 GB Runs (tight) ~20 tok/s
Gemma 3 27B 27B Q4_K_M 20 GB CPU Offload ~5 tok/s
Qwen 3 30B-A3B (MoE) 30B Q4_K_M 22 GB CPU Offload ~5 tok/s
Cogito 32B 32B Q4_K_M 21.5 GB CPU Offload ~5 tok/s
DeepSeek R1 32B 32B Q4_K_M 20.7 GB CPU Offload ~5 tok/s
Qwen 2.5 32B 32B Q4_K_M 20.7 GB CPU Offload ~5 tok/s
Qwen 2.5 Coder 32B 32B Q4_K_M 23 GB CPU Offload ~5 tok/s
Qwen 3 32B 32B Q4_K_M 23 GB CPU Offload ~5 tok/s
QwQ 32B 32B Q4_K_M 21.5 GB CPU Offload ~5 tok/s
Command R 35B 35B Q4_K_M 22.5 GB CPU Offload ~5 tok/s
Mixtral 8x7B 47B Q4_K_M 29.7 GB CPU Offload ~4 tok/s
19 model(s) are too large for this hardware.