PCI (I/O) partitioning on Sub-Numa Clustering

[gkgabel]

I understand to a level how SNC in Intel processors partitions CPU, memory and LLC into clusters. However it is not clear to me how PCI root complexes are divided among these sub-numa clusters. We have a Granite Rapids CPU in our lab and though we can see different PCI root complexes mapped to different sub-numa cluster presented as NUMA nodes but I want to understand 1) logically what is the scheme of this partitioning the PCI root complexes 2) Also like physical NUMA nodes are there penalties for accessing a PCI device shown on a remote Sub-NUMA cluster within a single physical NUMA and if so are by how much?

Thanks

[IntelSupport · Accepted solution]

Hello gkgabel,

Greetings from Intel!  

Thank you for contacting Intel. To proceed further with your query, I kindly request that you share the below details: 

1. Product details 
2. System details  
3. Elaborate the issue that you are facing. 

Please feel free to respond to this email at your earliest convenience. 

Regards, 

Shankith K P

Intel Customer Support Technician

[IntelSupport]

Hello gkgabel,

Greetings for the day!

We would like to inform you that we are closing this request as no response has been received from our previous follow-ups. 

You can get back to us anytime. We are happy to assist you.

Regards,

Shankith K P

Intel Customer Support Technician