Does Azure ML runtime on compute cluster takes longer than compute instance with same specifications?

Question

Hi,

I have a problem with the runtime of my Azure ML jobs. When I use a stronger configuration for a compute cluster than for a compute instance, than it takes way more time until the job finishes, when I compare it to the runtime with a compute instance. How is this possible? I did not find anything related to that issue and assumed, that the runtime should be nearly identical, if I use the same specifications, or even better ones on a compute cluster

Exemplary I get the following results for a specific Azure ML Job:

• Compute Instance Standard_DS11_v2 (2 cores, 14GB RAM, 28GB storage): Runtime 38min
• Compute Cluster Standard_A8m_v2 (8 cores, 64 GB RAM, 80 GB disk, 0-10 nodes): 2h 14min

This is currently a problem, because we switched due to costs from Compute Instances to Compute Cluster and assumed, that this should not affect the runtime, if I´d use at least the same specifications.

Answer 1

Hi @Yannic V,

I'm glad to hear that you were able to identify the issue and resolve it by switching to D-models. Regarding your question about the best machines for ML jobs that require powerful CPUs but not a lot of RAM, the D-series machines are a good option. They are optimized for compute-intensive workloads and offer a good balance of CPU and memory resources. However, if you need even more CPU power, you may want to consider the F-series machines, which are optimized for compute-intensive workloads that require high CPU performance.

To find the most suitable VM for your workload, you can use the Azure VM Selector tool, which provides recommendations based on your workload requirements.

For more info see: Sizes for virtual machines in Azure and Azure VM documentation.

Hope this helps. Do let us know if you any further queries.

---

If this answers your query, do click Accept Answer and Yes for was this answer helpful. This might be beneficial to other community members reading this thread as a solution.