Install Steam
login
|
language
简体中文 (Simplified Chinese)
繁體中文 (Traditional Chinese)
日本語 (Japanese)
한국어 (Korean)
ไทย (Thai)
Български (Bulgarian)
Čeština (Czech)
Dansk (Danish)
Deutsch (German)
Español - España (Spanish - Spain)
Español - Latinoamérica (Spanish - Latin America)
Ελληνικά (Greek)
Français (French)
Italiano (Italian)
Bahasa Indonesia (Indonesian)
Magyar (Hungarian)
Nederlands (Dutch)
Norsk (Norwegian)
Polski (Polish)
Português (Portuguese - Portugal)
Português - Brasil (Portuguese - Brazil)
Română (Romanian)
Русский (Russian)
Suomi (Finnish)
Svenska (Swedish)
Türkçe (Turkish)
Tiếng Việt (Vietnamese)
Українська (Ukrainian)
Report a translation problem
May I know your GPU? And also does it still freeze if you disable image generation?
Unfortunately the AI requires a lot of VRAM and also uses pytorch which only works on NVIDIA. It's listed in the system requirements and the store page, but if you don't meet the GPU requirements you can try to run the 125M model and disable image generation, or you can feel free to refund
1080TI ( 11G ) somehow is so slow for me...
And videocard is loaded only for 60-70% ( https://imgur.com/zUT9bPQ )
Idk why... Using 1.3B model with fast generation and it's taking for about 20-30 ( or even more ) minutes for 1 action...
So only way for me is to use 125M model
Waiting to initialize is the longest part of the process I have seen and every time it was because it crashed when I looked at the log. Even when it worked even just a few minutes ago in a previous world gen. So always check if you are in doubt.
Even with a RTX 3090 it can still take 5-20 seconds per turn.
The symptoms you're describing sounds like a bug where your GPU wasn't actually used; it must be using CPU to take 20-30 minutes which is an extremely long time. I added you in case you want to help me debug this, although since I'm also working on other things it might take me a while to get around to it
You can check your vram usage with something like MSI afterburner. If it's topping out that means VRAM might be the bottleneck. If not then maybe it's not using the GPU at all for some reason.
If VRAM is the issue, you can try setting image generation mode to Wombo
A note to everyone search in google your video card (if its nvidia) the card name and how much VRAM. I looked it up for you and it says
the GTX 1660 has 6 GB VRAM.
If your computer is displaying anything at all like the desktop, any background apps, steam, steam overlay your VRAM will be a bit below 6GB. That means you shouldn't attempt to run a 6GB VRAM requirement AI. Sorry :(
But you can confirm this. Start a new world in the game with the 6GB AI model. press tilda, look at the center column Wait a little bit, it'll show some log stuff and it should stop at some point or just wait about 15 seconds.
Then click the copy to clipboard and then paste it in a new text file on your computer. Near the top it tells you how much VRAM you have. My guess is something around 5GB if you did a fresh computer restart and the only thing you did after that was open Steam then AI rougelite. Thats a good way to ensure you get as much VRAM reserved for the game before other programs reserve it. You can also try to reduce any other background programs that open on computer start.
Back to the log, at the bottom, if you let it run till the log stops displaying new info you'll note it'll say some sort of error. From what I've seen this means there wasn't enough VRAM even if it doesn't say anything about VRAM.
This post inspired me to make a NVIDA VRAM chart, I posted it here
https://steamcommunity.com/app/1889620/discussions/0/3457093950218277869/
RTX 3090, looks like no GPU or CPU usage happens when i consult the AI for text (13GB VRAM model)
It looks like it finishes really fast when i press ~ and look at left side, but it gets stuck on consulting the AI.
https://imgur.com/a/PCkw8a6
and then I have to alt-F4 to quit out. No other programs using VRAM are open and this is using 23.3GB VRAM.
Does it happen consistently or only for a specific save?
Could you press copy to clipboard and paste it in a paste.ee and send it over?