Best local AI models for AMD HD 7870M

2 GB GDDR5. At a 4k context, 56 of the 233 models in our catalog with verified parameter counts fit fully, up to Allegro at 2.8B parameters.

Check your own machine against every model →

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.

ModelParametersBest quant that fitsMemory used at 4k
Allegro2.8BQ4_K_M2 GB
Open-Sora Plan2.7BQ4_K_M2 GB
LFM2 1.2B / 2.6B2.6BQ4_K_M1.9 GB
Playground v2.52.6BQ4_K_M1.9 GB
Stable Diffusion 3.5 Medium2.5BQ4_K_M1.8 GB
Canary 1B / Qwen-2.5B2.5BQ4_K_M1.8 GB
SeamlessM4T v22.3BQ5_K_M2 GB
Parler-TTS2.2BQ5_K_M1.9 GB
Kimi K3 DSpark2.2BQ5_K_M2 GB
SmolVLM 256M / 500M / 2B2BQ6_K2 GB
Stable Diffusion 3 Medium2BQ6_K2 GB
Pyramid Flow2BQ6_K2 GB
Wav2Vec2 / XLS-R2BQ6_K2 GB
Moondream 21.9BQ6_K1.9 GB
Qwen3 1.7B1.7BQ6_K1.7 GB
SmolLM2 135M / 360M / 1.7B1.7BQ6_K1.7 GB
StableLM 2 1.6B1.6BQ8_02 GB
Sana 0.6B / 1.6B1.6BQ8_02 GB
Zonos 0.11.6BQ8_02 GB
Dia 1.6B1.6BQ8_02 GB
Whisper Large v31.55BQ8_02 GB
ControlNet / T2I-Adapter / IP-Adapter1.5BQ8_01.9 GB
Hunyuan-DiT1.5BQ8_01.9 GB
Stable Video Diffusion1.5BQ8_01.9 GB
Whisper Large v2 / turbo1.5BQ8_01.9 GB
AudioGen1.5BQ8_01.9 GB
AudioLDM 21.5BQ8_01.9 GB
Tango 21.4BQ8_01.8 GB
TinyLlama 1.1B1.1BQ8_01.4 GB
SantaCoder 1.1B1.1BQ8_01.4 GB

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.

ModelParametersMemory at Q4_K_MSystem RAM at 4k
SmolLM3 3B3B2.2 GB needed4.2 GB
Replit Code v1.5 3B3B2.2 GB needed4.2 GB
Kandinsky 3.13B2.2 GB needed4.2 GB
Voxtral Mini / Small3B2.2 GB needed4.2 GB
Orpheus TTS3B2.2 GB needed4.2 GB
Higgs Audio v23B2.2 GB needed4.2 GB
MusicGen small/medium/large3.3B2.4 GB needed4.4 GB
Stable Diffusion XL3.417B4.1 GB needed6.1 GB
SDXL Turbo3.5B2.6 GB needed4.6 GB
SDXL Lightning3.5B2.6 GB needed4.6 GB

How to read this

The AMD HD 7870M is a mobile graphics card equipped with 2 GB of GDDR5 video memory. This hardware memory size is the absolute limit for running local AI models entirely on the graphics processor. When a model fits inside this 2 GB boundary, the GPU handles all computations quickly. If a model exceeds this limit, the system must split the workload.

To fit models into this limited space, developers use quantization. The quant column shows the specific compression level used to shrink the model. For example, the 2.8B Allegro and 2.7B Open-Sora Plan models use the Q4_K_M quant to fit exactly into 2 GB of video memory. Smaller models like the 1.7B SmolLM2 or 1.7B Qwen3 can run at a higher quality Q6_K quant while using 1.7 GB of video memory. The 1.1B TinyLlama uses the Q8_0 quant and requires only 1.4 GB of video memory.

When you want to run larger models, you must use CPU offloading. This process shares the workload between your GPU and your system RAM. For instance, running the 3B SmolLM3 or 3B Replit Code v1.5 at Q4_K_M requires 2.2 GB of video memory and 4.2 GB of system RAM. The 3.3B MusicGen requires 2.4 GB of video memory and 4.4 GB of system RAM. This setup works well if your computer has 32 GB of system RAM.

Offloading comes with a performance cost. Moving data between the graphics card and system memory is much slower than keeping everything on the GPU. While offloading allows you to run larger options like the 3.5B SDXL Turbo or 3.5B SDXL Lightning, your generation speeds will drop significantly compared to running smaller models fully on the GPU.

You must also consider the context window when running text models. The memory numbers listed here assume a standard 4k context window. If you increase the context length to process longer documents, the model will require more video memory. This extra memory demand can push a model over the 2 GB limit and trigger unexpected CPU offloading.