AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add 8P6L multi-node models (#239)

Browse Source 
[ROCm/rccl commit: d5f90e19b5]
This commit is contained in:
Wenkai Du
2020-07-21 14:10:36 -07:00
committed by GitHub
parent 56d8c7c893
commit f604fc774e
15 changed files with 830 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/tools/scripts/topo_val.sh
+1 -1
View File
@@ -21,7 +21,7 @@
DIR="$(cd -P "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
for i in {0..13}
for i in {0..20}
do
$DIR/../topo_expl/topo_expl -m $i > "topo_m$i.log"
$DIR/../TopoVisual/topo_visual.sh -i "topo_m$i.log"
projects/rccl/tools/topo_expl/models/topo_8p6l.xml
+10 -10
View File
@@ -2,7 +2,7 @@
<cpu numaid="0" affinity="00000000,00000000,00000000,00000000,00000000,000000ff,ffff0000,00ffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:61:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:63:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="30" rank="0" gdr="0">
<gpu dev="0" sm="30" rank="0" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
@@ -14,7 +14,7 @@
</pci>
<pci busid="0000:41:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:43:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="30" rank="1" gdr="0">
<gpu dev="1" sm="30" rank="1" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
@@ -26,7 +26,7 @@
</pci>
<pci busid="0000:22:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:24:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="30" rank="2" gdr="0">
<gpu dev="2" sm="30" rank="2" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
@@ -38,7 +38,7 @@
</pci>
<pci busid="0000:01:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:03:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="30" rank="3" gdr="0">
<gpu dev="3" sm="30" rank="3" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
@@ -52,7 +52,7 @@
<cpu numaid="1" affinity="00000000,00000000,00000000,00000000,00000000,ffffff00,0000ffff,ff000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:e1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:e3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="30" rank="4" gdr="0">
<gpu dev="4" sm="30" rank="4" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
@@ -64,7 +64,7 @@
</pci>
<pci busid="0000:c1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="30" rank="5" gdr="0">
<gpu dev="5" sm="30" rank="5" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
@@ -76,7 +76,7 @@
</pci>
<pci busid="0000:a1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:a3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="30" rank="6" gdr="0">
<gpu dev="6" sm="30" rank="6" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
@@ -88,7 +88,7 @@
</pci>
<pci busid="0000:81:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:83:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="30" rank="7" gdr="0">
<gpu dev="7" sm="30" rank="7" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
@@ -98,9 +98,9 @@
</gpu>
</pci>
</pci>
<pci busid="0000:c4:00.0" class="0x020000" link_speed="2.5 GT/s" link_width="1">
<pci busid="0000:9a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="enp196s0" dev="0" speed="10000" port="0" guid="0x0" maxconn="65536" gdr="0"/>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0x2f3ea3cccb374e8f" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
projects/rccl/tools/topo_expl/models/topo_8p6l_1nic.xml
+107
View File
@@ -0,0 +1,107 @@
<system version="1">
<cpu numaid="0" affinity="00000000,00000000,00000000,00000000,00000000,000000ff,ffff0000,00ffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:61:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:63:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="30" rank="0" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:41:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:43:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="30" rank="1" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:22:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:24:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="30" rank="2" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:01:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:03:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="30" rank="3" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:1a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0xb39bb4cdcd174297" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
<cpu numaid="1" affinity="00000000,00000000,00000000,00000000,00000000,ffffff00,0000ffff,ff000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:e1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:e3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="30" rank="4" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="30" rank="5" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:a1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:a3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="30" rank="6" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:81:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:83:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="30" rank="7" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
</system>
projects/rccl/tools/topo_expl/models/topo_8p6l_2nic.xml
+112
View File
@@ -0,0 +1,112 @@
<system version="1">
<cpu numaid="0" affinity="00000000,00000000,00000000,00000000,00000000,000000ff,ffff0000,00ffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:61:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:63:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="30" rank="0" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:41:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:43:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="30" rank="1" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:22:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:24:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="30" rank="2" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:01:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:03:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="30" rank="3" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:1a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0xb39bb4cdcd174297" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
<cpu numaid="1" affinity="00000000,00000000,00000000,00000000,00000000,ffffff00,0000ffff,ff000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:e1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:e3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="30" rank="4" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="30" rank="5" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:a1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:a3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="30" rank="6" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:81:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:83:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="30" rank="7" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:9a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_1" dev="1" speed="200000" port="1" guid="0x2f3ea3cccb374e8f" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
projects/rccl/tools/topo_expl/models/topo_8p6l_3nic.xml
+117
View File
@@ -0,0 +1,117 @@
<system version="1">
<cpu numaid="0" affinity="00000000,00000000,00000000,00000000,00000000,000000ff,ffff0000,00ffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:61:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:63:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="30" rank="0" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:41:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:43:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="30" rank="1" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:22:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:24:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="30" rank="2" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:01:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:03:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="30" rank="3" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:1a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0xb39bb4cdcd174297" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:2a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_1" dev="1" speed="200000" port="1" guid="0x768caa3b34834210" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
<cpu numaid="1" affinity="00000000,00000000,00000000,00000000,00000000,ffffff00,0000ffff,ff000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:e1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:e3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="30" rank="4" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="30" rank="5" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:a1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:a3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="30" rank="6" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:81:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:83:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="30" rank="7" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:9a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_2" dev="2" speed="200000" port="1" guid="0x2f3ea3cccb374e8f" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
projects/rccl/tools/topo_expl/models/topo_8p6l_4nic.xml
+122
View File
@@ -0,0 +1,122 @@
<system version="1">
<cpu numaid="0" affinity="00000000,00000000,00000000,00000000,00000000,000000ff,ffff0000,00ffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:61:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:63:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="30" rank="0" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:41:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:43:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="30" rank="1" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:22:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:24:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="30" rank="2" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:01:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:03:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="30" rank="3" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:1a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0xb39bb4cdcd174297" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:2a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_1" dev="1" speed="200000" port="1" guid="0x768caa3b34834210" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
<cpu numaid="1" affinity="00000000,00000000,00000000,00000000,00000000,ffffff00,0000ffff,ff000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:e1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:e3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="30" rank="4" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="30" rank="5" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:a1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:a3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="30" rank="6" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:81:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:83:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="30" rank="7" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:9a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_2" dev="2" speed="200000" port="1" guid="0x2f3ea3cccb374e8f" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:ba:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_3" dev="3" speed="200000" port="1" guid="0xf440c4ac4a1b4175" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
projects/rccl/tools/topo_expl/models/topo_8p6l_5nic.xml
+127
View File
@@ -0,0 +1,127 @@
<system version="1">
<cpu numaid="0" affinity="00000000,00000000,00000000,00000000,00000000,000000ff,ffff0000,00ffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:61:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:63:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="30" rank="0" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:41:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:43:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="30" rank="1" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:22:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:24:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="30" rank="2" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:01:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:03:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="30" rank="3" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:1a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0xb39bb4cdcd174297" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:2a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_1" dev="1" speed="200000" port="1" guid="0x768caa3b34834210" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:3a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_2" dev="2" speed="200000" port="1" guid="0x19b2fb0cd9f240ea" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
<cpu numaid="1" affinity="00000000,00000000,00000000,00000000,00000000,ffffff00,0000ffff,ff000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:e1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:e3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="30" rank="4" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="30" rank="5" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:a1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:a3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="30" rank="6" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:81:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:83:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="30" rank="7" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:9a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_3" dev="3" speed="200000" port="1" guid="0x2f3ea3cccb374e8f" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:ba:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_4" dev="4" speed="200000" port="1" guid="0xf440c4ac4a1b4175" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
projects/rccl/tools/topo_expl/models/topo_8p6l_6nic.xml
+132
View File
@@ -0,0 +1,132 @@
<system version="1">
<cpu numaid="0" affinity="00000000,00000000,00000000,00000000,00000000,000000ff,ffff0000,00ffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:61:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:63:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="30" rank="0" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:41:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:43:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="30" rank="1" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:22:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:24:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="30" rank="2" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:01:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:03:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="30" rank="3" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:1a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0xb39bb4cdcd174297" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:2a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_1" dev="1" speed="200000" port="1" guid="0x768caa3b34834210" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:3a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_2" dev="2" speed="200000" port="1" guid="0x19b2fb0cd9f240ea" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
<cpu numaid="1" affinity="00000000,00000000,00000000,00000000,00000000,ffffff00,0000ffff,ff000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:e1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:e3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="30" rank="4" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="30" rank="5" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:a1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:a3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="30" rank="6" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:24:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:81:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:83:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="30" rank="7" gdr="1">
<xgmi target="0000:63:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:e3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:9a:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_3" dev="3" speed="200000" port="1" guid="0x2f3ea3cccb374e8f" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:ba:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_4" dev="4" speed="200000" port="1" guid="0xf440c4ac4a1b4175" maxconn="262144" gdr="1"/>
</nic>
</pci>
<pci busid="0000:da:00.0" class="0x020000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_5" dev="5" speed="200000" port="1" guid="0xbb4c1b2931cd460d" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
projects/rccl/tools/topo_expl/topo_expl.cpp
+48
View File
@@ -77,6 +77,12 @@ const char *model_descriptions[] = {
"4 nodes with 8 VEGA20 GPUs XGMI 4P2H 1 NIC 2nd Hive",
"4 nodes with 8 VEGA20 GPUs XGMI 4P2H 2 NIC",
"single node 8 VEGA20 Rome",
"4 nodes gfx908 8P6L 1 NIC",
"4 nodes gfx908 8P6L 2 NICs",
"4 nodes gfx908 8P6L 3 NICs",
"4 nodes gfx908 8P6L 4 NICs",
"4 nodes gfx908 8P6L 5 NICs",
"4 nodes gfx908 8P6L 6 NICs",
NULL,
};
@@ -171,6 +177,48 @@ int main(int argc,char* argv[])
node = new NodeModel("topo_8p_rome.xml");
network.AddNode(node);
break;
case 14:
for (int i=0; i<4; i++) {
node = new NodeModel("topo_8p6l.xml");
network.AddNode(node);
}
break;
case 15:
for (int i=0; i<4; i++) {
node = new NodeModel("topo_8p6l_1nic.xml");
network.AddNode(node);
}
break;
case 16:
for (int i=0; i<4; i++) {
node = new NodeModel("topo_8p6l_2nic.xml");
network.AddNode(node);
}
break;
case 17:
for (int i=0; i<4; i++) {
node = new NodeModel("topo_8p6l_3nic.xml");
network.AddNode(node);
}
break;
case 18:
for (int i=0; i<4; i++) {
node = new NodeModel("topo_8p6l_4nic.xml");
network.AddNode(node);
}
break;
case 19:
for (int i=0; i<4; i++) {
node = new NodeModel("topo_8p6l_5nic.xml");
network.AddNode(node);
}
break;
case 20:
for (int i=0; i<4; i++) {
node = new NodeModel("topo_8p6l_6nic.xml");
network.AddNode(node);
}
break;
default:
printf("Invalid model_id %d\n", model_id);
exit(0);