Best local AI models for AMD R5 M330

The largest models that fit fully

The 30 largest of the 56 models that fit; every smaller model in the catalog fits too. Best quant means the highest quality compression whose weights and 4k context both sit inside the memory.

Close, but only with CPU offload

These need more than the card holds at their smallest practical quant, so part of the model runs from system memory (figures assume 32 GB of it). They work, several times slower.

How to read this

The AMD R5 M330 is an entry level laptop graphics card equipped with 2 GB of DDR3 video memory. This limited VRAM capacity dictates which local AI models you can run directly on the hardware. To fit within this 2 GB boundary, models must use quantization. Quantization is a compression method that reduces the precision of model weights to save space. The quant column shows the optimal format that balances model accuracy with the strict memory limits of your hardware.

For fully local execution, the largest fitting models include Allegro 2.8B and Open-Sora Plan 2.7B using the Q4_K_M quant, which consumes exactly 2 GB of VRAM. Other compatible options are LFM2 2.6B and Playground v2.5, which require 1.9 GB of VRAM at Q4_K_M. You can also run Stable Diffusion 3.5 Medium and Canary 1B / Qwen-2.5B at Q4_K_M, using 1.8 GB of VRAM. These configurations maximize the processing power of your GPU without exceeding its physical memory.

Slightly smaller models can run with higher precision quants. SeamlessM4T v2 2.3B and Kimi K3 DSpark 2.2B both utilize 2 GB of VRAM at the Q5_K_M quant. Parler-TTS 2.2B fits well at Q5_K_M using 1.9 GB of VRAM. If you prefer Q6_K quants for better output quality, SmolVLM 2B, Stable Diffusion 3 Medium 2B, Pyramid Flow 2B, and Wav2Vec2 / XLS-R 2B will all fit exactly into 2 GB of VRAM. Moondream 2 1.9B fits at Q6_K using 1.9 GB of VRAM.

Highly compressed Q8_0 quants are available for smaller models. StableLM 2 1.6B, Sana 1.6B, Zonos 0.1 1.6B, and Whisper Large v3 1.55B all run at Q8_0 using 2 GB of VRAM. ControlNet / T2I-Adapter / IP-Adapter 1.5B, Hunyuan-DiT 1.5B, Stable Video Diffusion 1.5B, Whisper Large v2 / turbo 1.5B, AudioGen 1.5B, and AudioLDM 2 1.5B use 1.9 GB of VRAM at Q8_0. Tango 2 1.4B uses 1.8 GB at Q8_0, while TinyLlama 1.1B and SantaCoder 1.1B require 1.4 GB of VRAM at Q8_0.

When a model exceeds the 2 GB VRAM limit, you must use CPU offloading. This process splits the workload between your GPU and your system RAM. Offloading allows you to run larger models, but it significantly reduces processing speed because DDR3 VRAM and system RAM are much slower than modern graphics memory. For these setups, we assume a standard system configuration with 32 GB of system RAM.

With CPU offloading, you can run SmolLM3 3B, Replit Code v1.5 3B, Kandinsky 3.1 3B, Voxtral Mini / Small 3B, Orpheus TTS 3B, and Higgs Audio v2 3B. These models need 2.2 GB of VRAM at Q4_K_M and 4.2 GB of system RAM. MusicGen 3.3B requires 2.4 GB at Q4_K_M and 4.4 GB of system RAM. Larger image generators like SDXL Turbo 3.5B and SDXL Lightning 3.5B need 2.6 GB at Q4_K_M and 4.6 GB of system RAM. Stable Diffusion XL 3.417B requires 4.1 GB at FP8 / optimized and 6.1 GB of system RAM.

You must monitor your context window size during operation. Running text models with a standard 4k context window increases memory usage. As the conversation grows, the active memory can easily exceed the 2 GB VRAM limit of the AMD R5 M330. If the context memory overflows, the system will automatically offload data to your system RAM, which causes a noticeable drop in generation speed.