Dear BlueBEAR users,
We have taken feedback from our user base regarding the time it takes for GPU jobs to start and have made changes with the aim of getting your jobs to start quicker.
The changes we have made are as below:
- For batch BlueBEAR GPU jobs:
- Maximum 2 days run time (“walltime”)
- Maximum 2 GPUs in use simultaneously by a user
- For BEAR Portal jobs:
- Maximum 1 GPU per user
- Maximum 12 hours run time (“walltime”) for CPU and GPU jobs
- BlueBEAR Priority GPU batch jobs:
- Maximum 2 days run time (“walltime”)
- Maximum (2 + priority level) GPUs in use simultaneously by a user
Our reasoning in making these changes are:
- This will reduce the resources that are allocated but not actually being used
- This will result in a more equal and equitable sharing of resources
- This change will reduce queue times
- GPU Priority Access will have a better impact on queue times than at present
BEAR Portal is intended for interactive jobs and the maximum job lengths should reflect this.
If you do need to run your job for more than 2 days, then please review our documentation on how to checkpoint your jobs.
We would also like to report that:
- Extra GPUs will be added to BlueBEAR as soon as we are able to
- Baskerville NCR will open for pilot users later this year
- We have added an additional 2,000 CPU cores to BlueBEAR
We will continue to review the impact of these changes and make necessary adjustments.
If you feel that you still need advice, then please do not hesitate in getting in contact via the service desk or Get Help.
We would like to thank you again for your support.