NVIDIAConsumerRTX 50

RTX 5070 voor lokale AI

RTX 5070 biedt 12 GB VRAM voor lokale AI. In de LocalIA-catalogus draaien 168 van 242 modellen comfortabel op één kaart.

VRAM
12GB
Categorie
Consumer
Serie
RTX 50
Vendor
NVIDIA

Modellen die comfortabel draaien

Deze modellen passen in 12 GB met marge voor context en stabiele inference.

DeepSeek V2 Litedeepseek10.1 GBcomfortabelQ4 · / 12 GB
DeepSeek Coder V2 Litedeepseek10.1 GBcomfortabelQ4 · / 12 GB
StarCoder 2 15Bstarcoder9.4 GBcomfortabelQ4 · / 12 GB
Phi-4 Reasoning Vision 15Bphi9.4 GBcomfortabelQ4 · / 12 GB
Qwen 2.5 14Bqwen8.8 GBcomfortabelQ4 · / 12 GB
Qwen 2.5 Coder 14Bqwen8.8 GBcomfortabelQ4 · / 12 GB
Qwen 3 14Bqwen8.8 GBcomfortabelQ4 · / 12 GB
DeepSeek R1 Distill 14Bdeepseek8.8 GBcomfortabelQ4 · / 12 GB
Phi-3 Medium 14Bphi8.8 GBcomfortabelQ4 · / 12 GB
Phi-4 14Bphi8.8 GBcomfortabelQ4 · / 12 GB
GLM-4.5 Airglm8.8 GBcomfortabelQ4 · / 12 GB
Qwen2.5 14B Instructqwen8.8 GBcomfortabelQ4 · / 12 GB
Qwen3 14Bqwen8.8 GBcomfortabelQ4 · / 12 GB
Qwen2.5 Coder 14B Instructqwen8.8 GBcomfortabelQ4 · / 12 GB
DeepSeek R1 Distill Qwen 14Bqwen8.8 GBcomfortabelQ4 · / 12 GB
Llama 2 13Bllama10.0 GBcomfortabelQ5 · / 12 GB
CodeLlama 13Bcodellama10.0 GBcomfortabelQ5 · / 12 GB
OLMo 2 13Bolmo10.0 GBcomfortabelQ5 · / 12 GB
Vicuna 13Bvicuna10.0 GBcomfortabelQ5 · / 12 GB
Mistral Nemo 12Bmistral9.2 GBcomfortabelQ5 · / 12 GB
Gemma 3 12Bgemma9.2 GBcomfortabelQ5 · / 12 GB
StableLM 2 12Bstable9.2 GBcomfortabelQ5 · / 12 GB
Solar 10.7Bsolar8.2 GBcomfortabelQ5 · / 12 GB
Falcon 3 10Bfalcon7.7 GBcomfortabelQ5 · / 12 GB
Gemma 2 9Bgemma10.1 GBcomfortabelQ8 · / 12 GB
Yi 1.5 9Byi10.1 GBcomfortabelQ8 · / 12 GB
Qwen 3.5 9Bqwen10.1 GBcomfortabelQ8 · / 12 GB
GLM-4 9Bglm10.1 GBcomfortabelQ8 · / 12 GB
GLM-4.7 Flashglm10.1 GBcomfortabelQ8 · / 12 GB
GLM-4.1V 9B Thinkingglm10.1 GBcomfortabelQ8 · / 12 GB

Vrijgespeeld in 2x-rig

Met twee kaarten parallel (24 GB totaal) worden grotere modellen bereikbaar.

Command R 35Bcommand22.0 GBkrapQ4 · / 24 GB
Aya 23 35Baya22.0 GBkrapQ4 · / 24 GB
CodeLlama 34Bcodellama21.4 GBkrapQ4 · / 24 GB
Yi 1.5 34Byi21.4 GBkrapQ4 · / 24 GB
dolphin 2.9.1 yi 1.5 34byi21.4 GBkrapQ4 · / 24 GB
Qwen 2.5 32Bqwen20.1 GBcomfortabelQ4 · / 24 GB
Qwen 2.5 Coder 32Bqwen20.1 GBcomfortabelQ4 · / 24 GB
Qwen 3 32Bqwen20.1 GBcomfortabelQ4 · / 24 GB
QwQ 32Bqwq20.1 GBcomfortabelQ4 · / 24 GB
DeepSeek R1 Distill 32Bdeepseek20.1 GBcomfortabelQ4 · / 24 GB
Qwen 2.5 VL 32Bqwen20.1 GBcomfortabelQ4 · / 24 GB
Granite 4 H-Small 32B-A9Bgranite20.1 GBcomfortabelQ4 · / 24 GB
GLM-4.6glm20.1 GBcomfortabelQ4 · / 24 GB
GLM-4.7glm20.1 GBcomfortabelQ4 · / 24 GB
GLM-5glm20.1 GBcomfortabelQ4 · / 24 GB

Vrijgespeeld in 4x-rig

Serverconfiguratie (48 GB totaal) voor de grootste open-weight modellen.

Qwen 2.5 72Bqwen45.3 GBkrapQ4 · / 48 GB
Qwen 2.5 VL 72Bqwen45.3 GBkrapQ4 · / 48 GB
Qwen2.5 72B Instructqwen45.3 GBkrapQ4 · / 48 GB
Llama 2 70Bllama44.0 GBkrapQ4 · / 48 GB
Llama 3 70Bllama44.0 GBkrapQ4 · / 48 GB
Llama 3.1 70Bllama44.0 GBkrapQ4 · / 48 GB
Llama 3.3 70Bllama44.0 GBkrapQ4 · / 48 GB
CodeLlama 70Bcodellama44.0 GBkrapQ4 · / 48 GB
DeepSeek R1 Distill 70Bdeepseek44.0 GBkrapQ4 · / 48 GB
Hermes 3 70Bhermes44.0 GBkrapQ4 · / 48 GB

Vergelijkbare GPUs

VRAM-schattingen bijgewerkt 2026-05-12.