Azure vs. 3CX hosted and AI engine

[egarciapf]

We currently have a setup on Azure and would like to use the AI engine (local) for calls transcriptions. I have 2 questions:

1. Is there any Azure VM size recommended to build a Transcription Engine Server?
2. If we were to use 3CX Hosted, will it include a local Transcription Engine Server in 3CX cloud infrastructure? Or do we need to build regardless our own if we want to use the local engine instead of the Cloud one?

Thanks,
Ernesto

 

[KyriacosS_3CX]

Hello,
EDIT:
Not sure why I read AWS instead of Azure - I will leave that info here too, but for Azure:
You are probably looking to use an NCCads_H100_v5 series machine.
or for bigger oompf an NC RTX PRO 6000 Blackwell Server Edition v6 series.
At what should be a lower entry point perhaps a beefy Standard_NV72ads_A10_v5 will also work.
(
https://learn.microsoft.com/en-us/a...rated/nvadsa10v5-series?tabs=sizeaccelerators
https://learn.microsoft.com/en-us/a...ccelerated/nccadsh100v5-series?tabs=sizebasic
https://learn.microsoft.com/en-us/a...00-bse-v6-series?tabs=sizebasicgp,sizebasicco
)
You need to confirm that you can deploy the OS needed and make sure the cards take RTX drivers. - This is a guideline from published machine specs, not based on personal testing.

1. Will vary on needs but given our general recommendation the AWS equivalent would be in the g5 - g6 categories and up. You need to confirm that you can deploy the OS needed and make sure the cards take RTX drivers. - This is a guideline from published machine specs from AWS, not based on personal testing. (https://aws.amazon.com/ec2/instance-types/g5/ https://aws.amazon.com/ec2/instance-types/g6/)

2. The Local Transcription Engine Server is always self deploy in your cloud/on-prem architecture. The Enterprise Plus (ENT+) package includes metered access to our (hosted/managed by 3CX) Transcription Servers, if you prefer not to deploy yourself.

 

[egarciapf]

Hello,
EDIT:
Not sure why I read AWS instead of Azure - I will leave that info here too, but for Azure:
You are probably looking to use an NCCads_H100_v5 series machine.
or for bigger oompf an NC RTX PRO 6000 Blackwell Server Edition v6 series.
At what should be a lower entry point perhaps a beefy Standard_NV72ads_A10_v5 will also work.
(
https://learn.microsoft.com/en-us/a...rated/nvadsa10v5-series?tabs=sizeaccelerators
https://learn.microsoft.com/en-us/a...ccelerated/nccadsh100v5-series?tabs=sizebasic
https://learn.microsoft.com/en-us/azure/virtual-machines/sizes/gpu-accelerated/nc-rtxpro6000-bse-v6-series?tabs=sizebasicgp,sizebasicco
)
You need to confirm that you can deploy the OS needed and make sure the cards take RTX drivers. - This is a guideline from published machine specs, not based on personal testing.

1. Will vary on needs but given our general recommendation the AWS equivalent would be in the g5 - g6 categories and up. You need to confirm that you can deploy the OS needed and make sure the cards take RTX drivers. - This is a guideline from published machine specs from AWS, not based on personal testing. (https://aws.amazon.com/ec2/instance-types/g5/ https://aws.amazon.com/ec2/instance-types/g6/)

2. The Local Transcription Engine Server is always self deploy in your cloud/on-prem architecture. The Enterprise Plus (ENT+) package includes metered access to our (hosted/managed by 3CX) Transcription Servers, if you prefer not to deploy yourself.

Thanks. Loud and clear!

 

[nikolascx]

@egarciapf
Great questions - some additional tips for you .

One additional consideration is that since transcription jobs are queued, sizing is less about peak concurrent live calls and more about overall processing throughput.
What matters most is how quickly the GPU can clear the transcription queue, especially during busy periods. If call volume spikes, you’ll want enough GPU capacity to avoid backlog delays.
I would spend a month or 2 on azure and check the cost. You can share your costs and your usage with me on PM. We can then discuss how to proceed after the first bills.
Finally, don’t forget Azure GPU quota limits, as these can delay deployment if not requested in advance.

Cheers!