Best local AI models for AMD Pro 455

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 Radeon Pro 455 is a mobile graphics card equipped with 2 GB of GDDR5 video memory. This dedicated memory size dictates the maximum size of the AI models you can run entirely on the hardware. When a model fits completely within this 2 GB limit, it executes with the fastest possible processing speeds. If a model exceeds this capacity, it cannot run solely on the graphics card without utilizing system memory.

To fit larger models into the limited 2 GB video memory, you must use quantized versions. Quantization reduces the numerical precision of model weights to save space. The best quant column indicates the optimal balance between model accuracy and memory usage. For example, the 2.8B Allegro and 2.7B Open-Sora Plan models both require a Q4_K_M quantization to fit exactly into 2 GB of used video memory. Models like the 2B SmolVLM or 2B Stable Diffusion 3 Medium can run at a higher Q6_K quantization while still staying within the 2 GB limit.

Smaller models allow you to use even higher precision levels. The 1.6B StableLM 2, 1.6B Sana, 1.6B Zonos 0.1, and 1.6B Dia models can all run at Q8_0 quantization while using exactly 2 GB of video memory. Similarly, Whisper Large v3 at 1.55B fits into 2 GB at Q8_0 quantization. You can also run the 1.1B TinyLlama or 1.1B SantaCoder at Q8_0 quantization while using only 1.4 GB of video memory, leaving some safety margin.

When a model is too large for the 2 GB video memory, you must use CPU offload. This technique splits the model between your graphics card and your system RAM. We assume your computer has 32 GB of system RAM for these scenarios. For instance, running the 3B SmolLM3, 3B Replit Code v1.5, 3B Kandinsky 3.1, 3B Voxtral, 3B Orpheus TTS, or 3B Higgs Audio v2 at Q4_K_M quantization requires 2.2 GB of video memory and 4.2 GB of system RAM. Offloading allows these models to run, but it significantly reduces processing speed because system RAM is much slower than GDDR5 video memory.

Larger creative models also rely heavily on CPU offload. The 3.3B MusicGen requires 2.4 GB of video memory and 4.4 GB of system RAM at Q4_K_M quantization. The 3.5B SDXL Turbo and 3.5B SDXL Lightning require 2.6 GB of video memory and 4.6 GB of system RAM at Q4_K_M quantization. The 3.417B Stable Diffusion XL requires 4.1 GB of video memory and 6.1 GB of system RAM when running at FP8 or optimized settings.

You must also consider the 4k context caveat when running text models. The memory numbers listed here represent the base model requirements. Generating long responses or processing large prompts increases memory consumption. If you use a full 4k context window, the active memory usage will rise beyond the base figures, which may push a model over the 2 GB limit and trigger slow system RAM offloading.