[ home / rules / faq ] [ overboard / sfw / alt ] [ leftypol / edu / labor / siberia / lgbt / latam / hobby / tech / games / anime / music / draw / AKM ] [ meta ] [ wiki / shop / tv / tiktok / twitter / patreon ] [ GET / ref / marx / booru ]

/tech/ - Technology

"Technology reveals the active relation of man to nature" - Karl Marx
Name
Options
Subject
Comment
Flag
File
Embed
Password(For file deletion.)

Not reporting is bourgeois


File: 1727047685683.jpg (28.22 KB, 800x533, expert gun handling.jpg)

 

>i dont game or mine coinz
>see someone made a Stable Diffusion LoRa for a rare character I like and decide to give it a try
>there are a few tools and workarounds to stop my ~decade-old budget GPU from out-of-memorying, like tiling, although the quality is still ok enough for basic hobby fun
>decide try to make a new embedding for a rare character
>see a guide for [their formatting kept] "VERY LOW VRAM"
>aw sweet
>mfw "VERY LOW VRAM" is now apparently "[12GB/16GB/20GB]"
bitch i aint payin half a thousand for very low vram

>>26476
They made a what??

>>26476
If the program has sufficiently local access patterns, you can get decent performances with mostly swap.

Mine has 256M.

>>26478
There's swap for video?

>>26484
I thought VRAM meant virtual RAM/memory, my mistake. The answer to your question depends on how the application manages memory. System level memory access will pass through the virtual memory subsystem, but low-level VRAM access will be bounded by physical size. The program might even implement its own virtual memory layer for obvious reasons. Integrated graphics are a special case, because they often map their VRAM onto a physical memory region, whose size is sometimes governed by dynamic bios settings.

I suggest you take a look at the memory settings of your application. Performancewise the worst-case scenario would be running a virtual machine that emulates a video card.

>>26476
Isn't stable diffusion minimum VRAM like 4Gb?

>>26488
For one, I believe you can run it on CPU/RAM only if you have enough, it will just generally be much slower.
I don't have a 4GB VRAM device and I'm running it. But newer higher-res models trained on larger pics (e.g. 1024x1024 pics instead of 512x512 pics) or any upscaling to a larger output requires using tiling (generating e.g. 4 smaller images to make one big image) or just using CPU mode. So 4GB would be the recommended, at least for whenever model they were talking about.


Unique IPs: 5

[Return][Go to top] [Catalog] | [Home][Post a Reply]
Delete Post [ ]
[ home / rules / faq ] [ overboard / sfw / alt ] [ leftypol / edu / labor / siberia / lgbt / latam / hobby / tech / games / anime / music / draw / AKM ] [ meta ] [ wiki / shop / tv / tiktok / twitter / patreon ] [ GET / ref / marx / booru ]