Configure a rig →
AMDDatacenterInstinct CDNA 3+

Instinct MI325X for local AI

Instinct MI325X provides 256 GB of VRAM for local AI. In the LocalIA catalog, 234 out of 242 models run comfortably on a single card.

View all compatible models →Rig around the MI325X ↗
VRAM
256GB
Category
Datacenter
Series
Instinct CDNA 3+
Vendor
AMD

Models that run comfortably

These models fit in 256 GB with room for context and stable inference.

Nemotron 340Bnemotron213.7 GBcomfortableQ4 · / 256 GB
DeepSeek V2deepseek181.3 GBcomfortableQ5 · / 256 GB
DeepSeek Coder V2deepseek181.3 GBcomfortableQ5 · / 256 GB
★Qwen 3 235B A22Bqwen180.6 GBcomfortableQ5 · / 256 GB
★Qwen3 235B A22Bqwen180.6 GBcomfortableQ5 · / 256 GB
Falcon 180Bfalcon201.2 GBcomfortableQ8 · / 256 GB
Mixtral 8x22Bmistral157.6 GBcomfortableQ8 · / 256 GB
★Mistral Large 123Bmistral137.5 GBcomfortableQ8 · / 256 GB
★NVIDIA Nemotron 3 Super 120B A12B BF16nemotron134.1 GBcomfortableQ8 · / 256 GB
★Llama 4 Scout 17Bx16llama121.8 GBcomfortableQ8 · / 256 GB
★Command R+ 104Bcommand116.2 GBcomfortableQ8 · / 256 GB
★Qwen3 Next 80B A3B Instructqwen178.8 GBcomfortableFP16 · / 256 GB
★Qwen 2.5 72Bqwen160.9 GBcomfortableFP16 · / 256 GB
Qwen 2.5 VL 72Bqwen160.9 GBcomfortableFP16 · / 256 GB
★Qwen2.5 72B Instructqwen160.9 GBcomfortableFP16 · / 256 GB
Llama 2 70Bllama156.5 GBcomfortableFP16 · / 256 GB
Llama 3 70Bllama156.5 GBcomfortableFP16 · / 256 GB
Llama 3.1 70Bllama156.5 GBcomfortableFP16 · / 256 GB
★Llama 3.3 70Bllama156.5 GBcomfortableFP16 · / 256 GB
CodeLlama 70Bcodellama156.5 GBcomfortableFP16 · / 256 GB
★DeepSeek R1 Distill 70Bdeepseek156.5 GBcomfortableFP16 · / 256 GB
Hermes 3 70Bhermes156.5 GBcomfortableFP16 · / 256 GB
★Llama 3.1 Nemotron 70Bnemotron156.5 GBcomfortableFP16 · / 256 GB
Athene 70Bathene156.5 GBcomfortableFP16 · / 256 GB
★Llama 3.3 70B Instructllama156.5 GBcomfortableFP16 · / 256 GB
★Llama 3.1 70B Instructllama156.5 GBcomfortableFP16 · / 256 GB
★Mixtral 8x7Bmistral105.1 GBcomfortableFP16 · / 256 GB
Falcon 40Bfalcon89.4 GBcomfortableFP16 · / 256 GB
Command R 35Bcommand78.2 GBcomfortableFP16 · / 256 GB
Aya 23 35Baya78.2 GBcomfortableFP16 · / 256 GB

Tight models

These models barely fit. They can run, but context and speed will be limited.

★Llama 3.1 405Bllama254.6 GBtightQ4 · / 256 GB
Hermes 3 405Bhermes254.6 GBtightQ4 · / 256 GB
★Llama 4 Maverick 17Bx128llama251.5 GBtightQ4 · / 256 GB

Unlocked in a 2x rig

With two cards in parallel (512 GB total), larger models become reachable.

★DeepSeek V3.2deepseek430.6 GBcomfortableQ4 · / 512 GB
★DeepSeek V4 Prodeepseek430.6 GBcomfortableQ4 · / 512 GB
★DeepSeek R1deepseek421.8 GBcomfortableQ4 · / 512 GB
★DeepSeek V3deepseek421.8 GBcomfortableQ4 · / 512 GB
★DeepSeek R1 (0528 snapshot)deepseek421.8 GBcomfortableQ4 · / 512 GB

Similar GPUs

VRAM estimates updated 2026-05-12.