Google Compute Engine: Unable to attach GPU to instance

google-cloud-platformgoogle-compute-engine

Here is what I am trying to do:
I configured an instance on Google Compute Engine to use NVIDIA K80 GPU. You can see the hardware specs below.

First time I also opened "Quotas" page and requested a GPU for europe-west1 zone. I had "0/0". Then after some bureaucracy proceedings with Google support they gave me one GPU. So now it looks like this:

Now,the problem is, when I launch the instance, I am always getting:

The zone 'projects/eyeframeserver/zones/europe-west1-b' does not have
enough resources available to fulfill the request. Try a different
zone, or try again later.

This is already second day in row that I am trying it,and nothing changes. The dudes from Google to whom I submit quota requests say it is not their problem. And Google groups are silent.

I also tried to re-create instance from scratch,using different hardware params. it only start-ups ok,if the GPU is not attached.

Best Answer

If the CPU platform you selected was either Automatic or Skylake, currently, Intel Skylake or later + your chosen GPU type NVDIA Tesla K80 is not a supported combination. It is recommended that you should try other combinations like other CPU platform with GPU type or Skylake and P100 or other.

Related Solutions

Ssh – Unable to connect Google Compute Engine instance via SSH in browser

It looks like you've added AllowUsers in /etc/ssh/sshd_config configuration file.

To resolve this issue, you'll need to attach the boot disk of your VM instance to a healthy instance as the second disk. Mount it, edit the configuration file and fix the issue.

Here are the steps you can take to resolve the issue:

First of all, take a snapshot of your instance’s disk, in case if a loss or corruption happens you can recover your disk.
In the Developers Console, click on your instance. Uncheck Delete boot disk when instance is deleted and then delete the instance. The boot disk will remain under “Disks”, and now you can attach the disk to another instance. You can also do this step using gcloud command:
```
$ gcloud compute instances delete NAME --keep-disks all
```
Now attach the disk to a healthy instance as an additional disk. You can do this through the Developers Console or using the gcloud command:
```
$ gcloud compute instances attach-disk EXAMPLE-INSTANCE --disk DISK --zone ZONE
```
SSH into your healthy instance.
Determine where the secondary disk lives:
```
$ ls -l /dev/disk/by-id/google-*
```

Mount the disk:

$ sudo mkdir /mnt/tmp
$ sudo mount /dev/disk/by-id/google-persistent-disk-1-part1 /mnt/tmp

Where google-persistent-disk-1 is the name of the disk

Edit sshd_config configuration file and remove AllowUsers line and save it.
```
$ sudo nano /mnt/tmp/etc/ssh/sshd_config
```
Now unmout the disk:
```
$ sudo umount /mnt/tmp
```
Detach it from the VM instance. This can be done through the Developers Console or using the command below:
```
$ gcloud compute instances detach-disk EXAMPLE-INSTANCE --disk DISK
```
Now create a new instance using your fixed boot disk.

google-cloud-platform – Google Cloud Instance Hostname Configuration (Compute Engine)

AFAIK, GCP CE doesn't give such a facility.

If you want to access them from your local machine through a name / alias instead of IP, you can always bind the IP and the hostname you want into your hosts file. So, you will be able to connect to your instance by the name you bound to its public IP address. if your OS is a GNU/Linux or Unix-like,you can use the following command: echo "111.111.111.111 my.whateveriwant-hostname.tld" | sudo tee -a /etc/hosts ( Assuming that 111.111.111.111 is your instance's public IP address )

If your need is to access them from one of the instance inside of their local network, you can simply use their hostname.

I hope this will help!

Best Answer

Related Solutions

Ssh – Unable to connect Google Compute Engine instance via SSH in browser

google-cloud-platform – Google Cloud Instance Hostname Configuration (Compute Engine)

Related Topic