I get an error "Instance Launch Failed. Max spot instance count exceeded" When I try to provision g3.16xlarge or g3.8xlarge. However I can provision g3s.xlarge and g3.4xlarge instance. The differece in the two types is the number of GPUs. The ones I can provision have only one GPU. I have checked my quota, I have a quota of 16 for all G and VT spot instances. I do not have any current instance running at the time of making the spot instance request. Can someone help mr with this ? Thanks
Asked
Active
Viewed 46 times
0
-
Have you checked all regions in your AWS account? – Allan Chua Jun 30 '23 at 04:48
-
What is your quota for "Running On-Demand G and VT instances" and "All G and VT Spot Instance Requests" ? – Anon Coward Jun 30 '23 at 05:03
-
All G and VT Spot Instance Requests quota is 16 and Running On-Demand G and VT instances quota is 0. I have asked for additional quota – arc Jun 30 '23 at 09:56
3 Answers
1
It looks like those specific GPU instances in the region of your choice are super constrained for that given point in time. What you can try is to spawn Spot instances in another region or just reserve 1 instance for your company to gain access to those instances.
Reserving 1 instance of your desired type gives you a slightly more expensive price compared to spot instances but stil cheaper than on-demand instances.
You may also want to try to reach out to AWS support or your company's account manager to see if they could help you get a spot instance for this family type.
Allan Chua
- 9,305
- 9
- 41
- 61
-
Have tried spawning an instance from another region. However I get the same error. – arc Jun 30 '23 at 10:02
-
If I have to reserver an instance, the minimum period I need to commit is for 1 month. That is too high for me. I need these instances, off and on. – arc Jun 30 '23 at 10:04
-
Then spot instances may not be the right provisioning model for you as AWS can take it away anytime even if you are processing something. Maybe consider the on-demand EC2 and stop the instance while its not being used. – Allan Chua Jun 30 '23 at 10:08
0
From what I head GPU instances are not readily available for spot instances.
Spot instances are surplus AWS resources, Given the buzz around AI, most GPU instances are taken in non-spot uses cases.
chrishollinworth
- 170
- 1
- 8
0
Dropped the idea of using spot instance for multi GPU instance type. Provisioned an on-demand multi GPU instance by increasing the limit of vCPUs.
arc
- 71
- 8