Trigger: ral-roasted
Shoutout to @barlogspam176 for the idea ❤
☕ Buy me a coffee: https://ko-fi.com/ralfingerai
🍺 Join my discord: https://discord.com/invite/pAz4Bt3rqb
Description
FAQ
Comments (20)
OMG I cant keep up with you... lawdy have mercy on my poor lil rtx 2070 super :)))
Haha, sorry! I also had a 2070s great card!
Try Boltning! It's incredibly fast! Check my Pokemon post for settings (and remember to use the VAE!!!) https://civitai.com/images/14751887
@RedAstronaut I'll check it out. I've tried a number of lighting hyper and turbo checkpoints and I do like the speed part of them.
@RedAstronaut O(n second thought, just saw it's over 12gb..... I'll pass but thanks anyways.
@Stable_Confuzion It's worth the 12gb! It's one of the best models. Top 5. Spend some time looking at the gallery on it before deciding!
@RedAstronaut I did decide after looking through the gallery to download it. It was when I went to the download button and saw the file size that I decided a 13gb checkpoint was probably not a good match for my 8gb card. Which was kinda the whole point of this post from the get go.
@Stable_Confuzion Does the checkpoint size matter compared to the card's VRAM? I ran Boltning fine on my GTX 1080 which has 8 GB VRAM too, same as the 2070 Super.
@RedAstronaut months ago me did run XL models way above 8GB on card with only 6GB VRAM that is on internet café using comfy --lowvram, it works using the fact of https://huggingface.co/docs/diffusers/api/loaders/single_file#loading-a-pipeline-using-fromsinglefile which single file ckpt/safetensors loader can split into blocks that sequentially placed on memory instead of all at once, additionally fp32 XL with 12-13GB size can still be used on such card after --force-fp16 which indicates the 16bit pruning cached into RAM, btw. said PC has 16GB RAM
hmm I'll give it a try and report back! I had assume the entire CP was loaded into vram...
I did not do anything special when I used it. I just use Forge UI and saved it in models and clicked refresh and it worked. It worked really fast on my GTX 1080.
@RedAstronaut omg.... thank you for your persistence and patience. I have not changed any settings to optimize it yet, I swears it took 5 days to load :))) Maybe because I'm impatient? 🙄🙄 Anyways, it did take me some time to load but I'm watching Dave Chapelle and a number of other things at the same time so while as of right now I cant speak for the speed, however, my first and only batch is 10 images with hires fix, a prompt known to me and wow... the quality is amazing. Thank you again and I will update on performance soon after I tweak things but... Yes, this 13gb checkpoint loads and functions in forge.... on my lil ole 8gb rtx
@Memeater thank you too as your reply also encouraged me to give it a shot!
I humbly bow my head. Thank you for the lesson.
@Stable_Confuzion no worries m8, just wanna say for comfy things done behind the curtain, user no need to know the nitty gritty, glad forge did achieve it also
looking at this makes me oddly hungry...
sick


















