Graphics cards compatible with Stable Diffusion


  • FrançaisFrançais
  • EnglishEnglish

  • Stable Diffusion requires certain graphics to work properly. Discover the models that will allow you to launch it locally.


    Suivez-nous sur notre page Facebook et notre canal Telegram

    Stable Diffusion nécessite certaines graphiques pour fonctionner correctement. Découvrez les modèles qui vous permettront de le lancer en local.

    Installing Stable Diffusion locally is increasingly simple, whether via Automatic1111, Invoke AI or Easy Diffusion. However, we always wonder if our current graphics cards will allow it to run. You need 3 conditions, the card must be Nvidia, its architecture must date from at least 2016 (with Pascal) and it must have 8 GB of Vram.

    AMD cards are having a hard time working with Stable Diffusion and it’s a shame that Team Red isn’t moving its ass because it could grab market share from Nvidia as demand for GPUs, compatible with AIs d images will explode in the future.

    Graphics cards compatible with Stable Diffusion

    • Pascal: GeForce GTX 1070, 1070 Ti, 1080 and 1080 Ti; GeForce Titan X and Titan XP
    • Volta: GeForce Titan V
    • Turing: GeForce RTX 2060 Super, 2070, 2070 Super and 2080; GeForce RTX Titan
    • Ampere: GeForce RTX 3060 Ti and LHR; GeForce RTX3070 and LHR; GeForce RTX3080 and LHR; GeForce RTX3090
    • Ada Lovelace : All 40 Series Graphics Cards

    The 4000 series is obviously the most compatible, but also the one that costs the skin of your buttocks and that of your family :

    • GeForce RTX 4090: the most powerful graphics card on the market, with 18,432 CUDA cores and 24 GB of GDDR6X memory
    • GeForce RTX 4080 Ti: A high-end graphics card, with 15,360 CUDA cores and 16 GB of GDDR6X memory
    • GeForce RTX 4070 Ti: an upper mid-range graphics card, with 9,216 CUDA cores and 10 GB or more of GDDR6X memory

    For prices, it starts at 300 dollars for a GTX 1070 (if you can find any) up to 2500 bucks for the 4090. Knowing that for a card that still has good performance, it is better to aim for a budget between 600 and 900 dollars to be safe. The 8 GB of Vram is indicated for the generation of images. But if you want to use DreamBooth or Lora to train your own models, then you need between 10 to 12 GB of Vram.

    What is terrible is that we are in a world with multi-core processors and the AIs have not been optimized for these architectures. When you consider that AMD’s Threadripper 3990x has 64 cores and 128 threads, you can only imagine what it could do with Stable Diffusion. At the same time, it is a processor that exceeds 5000 dollars…

    Houssen Moshinaly

    Actualité Houssenia Writing's Editor. Copywriter since 2009.

    Blogger and essayist, I have written 9 books on different subjects such as corruption in science, technological singularities or even fictions. I propose political and geopolitical analyzes on the incoming new world. I have a training in web writing and a long career as a proletarian.

    To contact me personally:

    Leave a Reply

    Your email address will not be published. Required fields are marked *