You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Why is this module "Semantic Mapping "not trained? #28
Hi
I want to known what gpu you use to run this code, I tried to reproduce this code on 3090 but report "CUDA out of memory".
I hope your question i solved as well!
This is the mean that your total memory is not enough for running the code. what is the total memory of your 3090 GPU?
My 3090 for running this code is equipped with dual 3090 graphics cards, each having 24GB memory. you can try to reduce the number of threads to run this code successfully.
---- Replied Message ----
| From | ***@***.***> |
| Date | 04/06/2024 18:16 |
| To | ***@***.***> |
| Cc | ***@***.***>***@***.***> |
| Subject | Re: [devendrachaplot/Object-Goal-Navigation] Why is this module "Semantic Mapping "not trained? (Issue #28) |
This is the mean that your total memory is not enough for running the code. what is the total memory of your 3090 GPU?
My 3090 for running this code is equipped with dual 3090 graphics cards, each having 24GB memory. you can try to reduce the number of threads to run this code successfully.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
Activity
[-]About Semantic_Mapping[/-][+] Why is this module "Semantic Mapping "not trained?[/+]wuyukun-tong commentedon Apr 5, 2024
Hi
I want to known what gpu you use to run this code, I tried to reproduce this code on 3090 but report "CUDA out of memory".
I hope your question i solved as well!
zyjzxy9 commentedon Apr 6, 2024
This is the mean that your total memory is not enough for running the code. what is the total memory of your 3090 GPU?
My 3090 for running this code is equipped with dual 3090 graphics cards, each having 24GB memory. you can try to reduce the number of threads to run this code successfully.
wuyukun-tong commentedon Apr 10, 2024