l40s "no ressources available"
Hello everytime i try to choose a l40s i keep getting a "no ressource avilable" message. There are many l40s available in community cloud, but i cant choose them, no matter how little disk space i assign to it.
56 Replies
Same
Send this to team will let you know if I get something.
Still not possible for me to get a l40s, any news about whats going on?
@deathawaits4 are you trying to use spot pods?
no, l40s dont allow spot anyway. Im trying to use a normal on deman pod in community cloud
no filters, no region, nothing set
Odd, have you contacted support on our site yet, if not are you able to?
i will, but i still dont understand this issue because other people can get a pod
Likely an issue with you're specific configuration, reach out to our support team and they can investigate further
i have contaced support, but even then nobody knows why this is happening. I am not running any configuration, im basically just choosing the base pytorch runpond template and it doesnt work
alright, so i found out the issue i think. The l40s cards dont have cuda installed? When i add a filter for the cuda version, it states there is no card available for it. This would be easy to fix.. by installing cuda on them
not sure if there is even a template that doesnt use cuda since its basically a gpu pod which is useless without cuda
what template you are using?
It doesn't matter which template. The issue is with every template. I tried basic pytorch, stable diffusion, runpod Desktop. The problem is, once a cuda version is selected, all l40s become unavailable. It doesn't matter which cuda version. I keep getting asked the same questions by support, but the template is not part of the issue.
intresting let me see if I can replicate it
secure or community?
also any specific region?
Community, but I get the same on secure with the l40s
No region filter
how do you check cuda?
You go to filters
I mean from inside pod
I don't know, I can't open the pod haha
so it's issue with specific pod?
No with every l40s
I can't get a pod with them
I have just deployed new pod and cuda works fine
Yes that's what support also says, but I showcased my issue above
It seems to only hit a few users
If you set a cuda filter, do the l40s stay in Status available?
yes I checked all filters and it works fine
Yea if I set the filter 12.3 cuda they all become unavailable
this one I did is secure cloud
on community cloud on free L40s
though intresting thing is that if I deselect all cuda I get also "There are no longer any instances available with the requested specifications. Please refresh and try again."
On community cloud there seams to be a bug, on secure cloud works fine
Yea I didn't try secure
Lately
But a few days ago I had the same issue
Will forward that bug to team as now I can replicate it
Perfect, thanks!!!
hei, is there any news on this? Yesterday i was able to take a single l40s with 5gig disc space and increase disc space after i rented it
today its back to square one
do not have info though on secure cloud it should be fine. On community cloud you would need check with cuda filter though only 1 cuda selected as it's broken
yea, with any cuda selected it shows 0 availability
sadly secure cloud is 3x the price of community, so thats not an option
why is there such a huge difference anyway
as secure cloud is hosted on dedicated datacenters
yea but the l40s performance is like 10% better than the rtx 6000 a.. which is half the price xD
nvm tho, hope the community card cuda issue gets resolved at some point
I mean its not cuda issue it's cause most L40s are taken
i heartfelty disagree
it lets me choose 6x l40s
which means that atleast 6 cards are free
maximum availability is 8
which means 2 cards are taken
the rest is unusable
nope there is not
how so?
it's visual bug
so why is it only on the l40s?
it just states that there are no instances with the requested resources
but the resources in question is cuda
because once you filter a cuda version the availability becomes 0
its also extreeeemely strange that once you get an l40s that you can only assign it with 5gigabyte disc space
and after you rented, you can choose how much ever you want
i rented the card with 5 gigabyte disc space and then edited the pod to 250
there is clearly more going on than just a visual bug
It would be really great if this could be transfered to someone from the development team
it also doesnt make sense that the visual bug would magically vanish once you choose 7 cards
as i said, there are 8 cards available and 6 of them seem to have issues
2 are rented
@deathawaits4 checked via api and no L40S free on community cloud
does the api do a cuda check?
even if its just a visual bug, this should get looked at
because its extremely strange behaviour for a visual bug
it's not cuda issue it's cause all machines with L40s are taken
Alright, but there is still strange behaviour and this visual bug is massively misleading
the fact that i can only choose a free l40s with 5gigabyte disc space and afterwards assign as much as i want once i rented it, is also weird
btw it's not like we have single machine with 8 L40s 😄
yea, but this visual bug still needs to be looked at
whats the point of a GUI if it displays misleading information? 😄
i gave notice to support, but its just beeing ignored as a non issue