Skip to content

NVIDIA GeForce RTX 5090

NVIDIA · 32GB GDDR7 · Can run 56 models

Buy Amazon
Manufacturer NVIDIA
VRAM 32 GB
Memory Type GDDR7
Architecture Blackwell
CUDA Cores 21,760
Tensor Cores 680
Bandwidth 1792 GB/s
TDP 575W
MSRP $1,999
Released Jan 30, 2025

AI Notes

The RTX 5090 is the ultimate consumer GPU for local AI. With 32GB of GDDR7 VRAM, it can run most 30B-parameter models at full precision and 70B models with quantization (Q4). The massive CUDA and tensor core count delivers exceptional inference throughput for real-time AI workloads.

Compatible Models

Model Parameters Best Quant VRAM Used Fit Est. Speed
Qwen 3 0.6B 600M Q4_K_M 2.5 GB Runs ~717 tok/s
Gemma 3 1B 1B Q8_0 2 GB Runs ~896 tok/s
Llama 3.2 1B 1B Q8_0 3 GB Runs ~597 tok/s
DeepSeek R1 1.5B 1.5B Q8_0 3 GB Runs ~597 tok/s
Gemma 2 2B 2B Q8_0 4 GB Runs ~448 tok/s
Gemma 3n E2B 2B Q4_K_M 3.3 GB Runs ~543 tok/s
Llama 3.2 3B 3B Q8_0 5 GB Runs ~358 tok/s
Phi-3 Mini 3.8B 3.8B Q8_0 5.8 GB Runs ~309 tok/s
Phi-4 Mini 3.8B 3.8B Q4_K_M 4.5 GB Runs ~398 tok/s
Gemma 3 4B 4B Q4_K_M 5 GB Runs ~358 tok/s
Gemma 3n E4B 4B Q4_K_M 4.5 GB Runs ~398 tok/s
Qwen 3 4B 4B Q4_K_M 4.5 GB Runs ~398 tok/s
DeepSeek R1 7B 7B Q8_0 9 GB Runs ~199 tok/s
Falcon 3 7B 7B Q4_K_M 6.8 GB Runs ~264 tok/s
Mistral 7B 7B Q8_0 9 GB Runs ~199 tok/s
Qwen 2.5 7B 7B Q8_0 9 GB Runs ~199 tok/s
Qwen 2.5 Coder 7B 7B Q8_0 9 GB Runs ~199 tok/s
Qwen 2.5 VL 7B 7B Q4_K_M 7 GB Runs ~256 tok/s
Cogito 8B 8B Q4_K_M 7.5 GB Runs ~239 tok/s
DeepSeek R1 8B 8B Q4_K_M 7.5 GB Runs ~239 tok/s
Llama 3.1 8B 8B Q8_0 10 GB Runs ~179 tok/s
Nemotron 3 Nano 8B 8B Q4_K_M 7.5 GB Runs ~239 tok/s
Qwen 3 8B 8B Q4_K_M 7.5 GB Runs ~239 tok/s
Gemma 2 9B 9B Q8_0 11 GB Runs ~163 tok/s
Falcon 3 10B 10B Q4_K_M 8.5 GB Runs ~211 tok/s
Llama 3.2 Vision 11B 11B Q4_K_M 8.5 GB Runs ~211 tok/s
Gemma 3 12B 12B Q4_K_M 10.5 GB Runs ~171 tok/s
Mistral Nemo 12B 12B Q4_K_M 9.5 GB Runs ~189 tok/s
DeepSeek R1 14B 14B Q4_K_M 9.9 GB Runs ~181 tok/s
Phi-4 14B 14B Q4_K_M 9.9 GB Runs ~181 tok/s
Phi-4 Reasoning 14B 14B Q4_K_M 11 GB Runs ~163 tok/s
Qwen 2.5 14B 14B Q4_K_M 9.9 GB Runs ~181 tok/s
Qwen 2.5 Coder 14B 14B Q4_K_M 12 GB Runs ~149 tok/s
Qwen 3 14B 14B Q4_K_M 12 GB Runs ~149 tok/s
StarCoder2 15B 15B Q8_0 17 GB Runs ~105 tok/s
Codestral 22B 22B Q4_K_M 14.7 GB Runs ~122 tok/s
Devstral 24B 24B Q4_K_M 17 GB Runs ~105 tok/s
Magistral Small 24B 24B Q4_K_M 17 GB Runs ~105 tok/s
Mistral Small 3.1 24B 24B Q4_K_M 18 GB Runs ~100 tok/s
Gemma 2 27B 27B Q4_K_M 17.7 GB Runs ~101 tok/s
Gemma 3 27B 27B Q4_K_M 20 GB Runs ~90 tok/s
Qwen 3 30B-A3B (MoE) 30B Q4_K_M 22 GB Runs ~81 tok/s
Cogito 32B 32B Q4_K_M 21.5 GB Runs ~83 tok/s
DeepSeek R1 32B 32B Q4_K_M 20.7 GB Runs ~87 tok/s
Qwen 2.5 32B 32B Q4_K_M 20.7 GB Runs ~87 tok/s
Qwen 2.5 Coder 32B 32B Q4_K_M 23 GB Runs ~78 tok/s
Qwen 3 32B 32B Q4_K_M 23 GB Runs ~78 tok/s
QwQ 32B 32B Q4_K_M 21.5 GB Runs ~83 tok/s
Command R 35B 35B Q4_K_M 22.5 GB Runs ~80 tok/s
Mixtral 8x7B 47B Q4_K_M 29.7 GB Runs (tight) ~60 tok/s
Cogito 70B 70B Q4_K_M 43 GB CPU Offload ~13 tok/s
DeepSeek R1 70B 70B Q4_K_M 43.5 GB CPU Offload ~12 tok/s
Llama 3.1 70B 70B Q4_K_M 43.5 GB CPU Offload ~12 tok/s
Llama 3.3 70B 70B Q4_K_M 43.5 GB CPU Offload ~12 tok/s
Qwen 2.5 72B 72B Q4_K_M 44.7 GB CPU Offload ~12 tok/s
Qwen 2.5 VL 72B 72B Q4_K_M 41 GB CPU Offload ~13 tok/s
13 model(s) are too large for this hardware.