Much lower memory bandwidth with NUMA disabled

[111alan]

Since some applications could not effectively utilize more than one NUMA node, I'm testing the situation when NUMA was disabled with a 2S 12-channel DDR4-2666 cascadelake-sp system. 

The memory latency impact was expected, but the bandwidth was also severely lowered to about 73GB/s, that is 1/3 the bandwidth of the same system with NUMA enabled, or almost half that of a 1S system.

Why disabling NUMA has such huge effect on memory bandwidth, even more than on latency? Are there anything I can do to improve on this? Thanks.

[Maria_R_Intel]

Thank you for posting on the Intel* Community.

To better assist you, please provide us with the below information:

• Are you using a server system? (provide details)
• Provide the specific CPU model on your system.

Best regards, 

Maria R.  

Intel Customer Support Technician 

[Maria_R_Intel]

Hello 111alan,

Were you able to check the previous post? 

Let me know if you need more assistance. 

Best regards, 

Maria R.  

Intel Customer Support Technician 

[111alan]

Currently waiting Emeth to check out the situation. Thank you for the assistance.

[Maria_R_Intel]

Hello 111alan,

Thank you for let us know, we're glad to hear that.

In this case, we will proceed to close this thread to avoid duplication of effort and the support will continue through the current case with Emeth.

Best regards,

Maria R.

Intel Customer Support Technician