For instance Intel Xeon Gold 6154 has S4S scalability, but the second generation Intel Xeon Gold 6254 has S4 scalability. As I know this is about the number of UPIs, but i'm not sure. If so, what i'm trying to understand is, while using a CPU pinning design, with 4 socket architecture, for throughput-sensitive systems, can this still cause a performance problem?
Thank you for your concern.
About the performance issue, I see that the second generation Intel Xeon Gold 6254 processor has a 4S scalability. So this means that CPU has 2 UPIs. So in case we use 4 socket architecture, there will be a square cpu design instead of mesh, which means not all the cpu's are directly connected. I was considering if this may cause performance problems.
We use these processors in our NFV infrastructure, to virtualise our tput-sensitive services. We normally dedicate sockets for virtual services, so we do not use UPI links at this level.
But considering the Host Operating System, CPUs are used randomly, so it can cause a performance loss. I think this is more relevant if i discuss this with the OS vendor, but I just wanted to take your opinion in CPU perspective. Because this seems like a handicap for the second generation Intel Xeon Gold 6254 compared to the older Gold 6154 version.