Announcement

Collapse
No announcement yet.

Intel Xeon 6980P SNC3 vs. HEX Clustering Mode Performance

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Intel Xeon 6980P SNC3 vs. HEX Clustering Mode Performance

    Phoronix: Intel Xeon 6980P SNC3 vs. HEX Clustering Mode Performance

    With the Intel Xeon 6900P "Granite Rapids" processors that launched last week there are SNC3 and HEX clustering modes for these new processors. The default Sub-NUMA Clustering 3 (SNC3) mode for the three compute dies while the HEX mode is like SNC1 mode formerly for all three compute dies acting as one NUMA node. Using the flagship 128-core Intel Xeon 6980P processors, I ran some benchmarks looking at the real-world performance difference for SNC3 vs. HEX clustering modes on Granite Rapids.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Interesting results and it looks like SC3 was the right choice for a default, but HEX is good for targeted workloads like NAMD.

    Comment


    • #3
      I didn't even know this was a configurable option. Obviously they chose the right default, but it's pretty cool that you can change this to better suit the workload of a given server.

      Comment


      • #4
        For the benefit of the search engine gods, lscpu from the image:
        Code:
          Address sizes: 52 bits physical, 57 bits virtual
          BIOS model name: Intel(R) Xeon(R) 6980 CPU @2.0GHz
          BIOS CPU family: 179
          CPU family: 6
          Model: 173
          Thread(s) per core: 2
          Core(s) per socket: 128
          Socket(s): 2
        Caches (sum of all):
          L1d: 12 MiB (256 instances)
          L1i: 16 MiB (256 instances)
          L2: 512 MiB (256 instances)
          L3: 1008 MiB (2 instances)
        NUMA:
          NUMA node0 CPU(s): 0-42,256-298 (43+43)
          NUMA node1 CPU(s): 43-85,299-341 (43+43)
          NUMA node2 CPU(s): 86-127,342-383 (42+42)
          NUMA node3 CPU(s): 128-170,384-426 (43+43)
          NUMA node4 CPU(s): 171-213,427-469 (43+43)
          NUMA node5 CPU(s): 214-255,470-511 (42+42)
        Whenever you see a core count that is not a power of two, or even a Rectangular Number, remember that the "leftover" spots on the rectangular die—which is probably a 6x8 grid layout—may either used for other functional blocks (e.g. Emerald Rapids did a 15-core+1 memctl part), or an easy way to deal with wafer yields. 5 dead cores is a lot of good headroom; if those tiles were individually sold, the SKU might have 48/44/40/36 cores or something. It's not unheard of. (Also: 48x3=144, forming perhaps a 6780E?)

        Comment

        Working...
        X