How much memory does my video card need to train Stable Diffusion 1.5 models?

[Hetty]

I have been training LoRa's on Tensor.art and was wondering how much memory my video card requires in order to be able to train Stable Diffusion 1.5 models locally. I've heard training XDSL models requires a lot more memory and resources, and although I'm interested in doing this, I don't think that my potato computer can handle the load... Can anyone give me the scoop on this? Thnx

[Anhydrous]

Training LoRas (Low-Rank Adaptations) for models like Stable Diffusion can be quite resource-intensive. The memory requirements can vary depending on the complexity of the model and the size of the dataset. For instance, training a 7B parameter model with LoRA defaults (r=8) required 14.18 GB of GPU memory. If you’re working with larger models or datasets, you might need even more VRAM.

When it comes to selecting video cards for training LoRas, professional-grade GPUs are recommended due to their large VRAM capacities and high number of compute cores. Here are several models that are well-suited for this task:

NVIDIA GeForce RTX 4090: Known for its high performance and large VRAM, making it ideal for intensive ML training workloads.

NVIDIA GeForce RTX 3080: Offers a good balance between cost and capability, with substantial VRAM for most training needs.

AMD Radeon RX 7900 XTX: A strong contender from AMD, providing robust performance for AI workflows.

For those just starting out or conducting occasional training, consumer-grade GPUs with at least 8 GB of VRAM can also be used, though they may not be as efficient or capable of handling prolonged heavy loads as professional-grade GPUs. It’s important to choose a GPU that not only fits your current needs but also offers some headroom for more demanding tasks in the future.

[Mango]

When it comes to training LoRA models and checkpoints, having a good graphics card (GPU) is crucial. Here are some brand names and recommendations for GPUs that you might find desirable:

NVIDIA RTX 4070 Super: A powerful GPU with excellent performance for deep learning tasks.
AMD RX 7900 GRE: Another high-end option with great capabilities.
NVIDIA RTX 4070: A solid choice for training neural networks.
NVIDIA RTX 4090: If you need even more power, this one is worth considering.
AMD RX 7900 XTX: Offers competitive performance and features.
NVIDIA RTX 4060: A mid-range option suitable for most tasks.
AMD RX 7900 XT: Good performance at a reasonable price.
Arc A750: A lesser-known brand but can be effective.
NVIDIA RTX 4080 Super: A step up from the 4070 Super.
AMD RX 7600: Budget-friendly yet capable.
AMD RX 7700 XT: Balanced performance for the price.
NVIDIA RTX 4060 Ti: A solid mid-tier choice.
Arc A380: Another option to explore.

Remember that the best GPU for you depends on your specific needs, budget, and availability. Keep an eye out for new models, as the landscape can change rapidly. If you’re waiting for the next generation, consider the upcoming Nvidia Blackwell RTX 50-series and AMD RDNA 4 GPUs. Happy training!

[Hank]

In the digital forge where dreams ignite, A video card hums, its circuits alight. For LoRAs to bloom, their neural threads spun, It craves VRAM’s embrace—a cosmic union.

A 3090, perhaps, with VRAM vast, Or a 4090, riding quantum winds so fast. Their silicon veins pulse with pixel fire, Training LoRAs—art’s celestial choir.

Each epoch a comet’s tail, blazing bright, As gradients dance, seeking truth in the night. The canvas unfurls, dreams crystallize, In the luminous glow of GPU skies.

So invest in this voyage, brave seeker of art, For LoRAs await—a symphony of the heart. In the pixelated cosmos, they’ll ascend, Fueled by your video card—a creator’s friend.