Cedar: Difference between revisions

Cedar (view source)

109 bytes added , 3 years ago

no edit summary

cc_staff

28

edits

@@ Line 107: / Line 107: @@
 = Performance = <!--T:17-->
-Theoretical peak double precision performance of Cedar is 6547 teraflops for CPUs, plus 7434 for GPUs, yielding almost 14 petaflops of theoretical peak double precision performance. 22 fully connected "islands" of 32 base or large nodes each have 1024 cores in a fully non-blocking topology (Omni-Path fabric), with each island designed to yield over 30 teraflops of double-precision performance (measured with high performance LINPACK). There is a 2:1 blocking factor between the 1024 core islands. Similarly the Skylake and Cascade Lake nodes span 44 non-blocking islands of 32 nodes each, forming islands of 1536 cores.
+Theoretical peak double precision performance of Cedar is 6547 teraflops for CPUs, plus 7434 for GPUs, yielding almost 14 petaflops of theoretical peak double precision performance.
+Cedar's network topology is mad up of "islands". Most islands contain 32 nodes: there are 18 islands with 32 Broadwell nodes, each with 32 cores (i.e., 1024 cores per island), 44 islands with 32 Skylake and Cascade Lake nodes, each with 48 cores (i.e., 1536 cores per island), 4 islands with 32 P100 GPU nodes and 6 islands with 32 V100 GPU nodes. There are also 2 islands with 32 large memory Broadwell nodes. Nodes within an island are connected in a fully non-blocking topology (Omni-Path fabric). There is a 2:1 blocking factor between islands.
 <!--T:16-->