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

Update Rome single node models (#277)

Browse Source
This commit is contained in:
Wenkai Du
2020-10-13 13:33:09 -07:00
committed by GitHub
parent 2b8184808d
commit 33babcb5e2
6 changed files with 227 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files
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..37}
for i in {0..38}
do
$DIR/../topo_expl/topo_expl -m $i > "topo_m$i.log"
$DIR/../TopoVisual/topo_visual.sh -i "topo_m$i.log"
tools/topo_expl/models/topo_8p_rome_n2_1.xml
+52 -47
View File
@@ -1,31 +1,72 @@
<system version="2">
<cpu numaid="2" affinity="00000000,ffffffff,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:c4:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c6:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="96" gcn="906" arch="38911" rank="0" gdr="1">
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<cpu numaid="0" affinity="00000000,00000000,00000000,ffffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="0" sm="90" gcn="906" arch="38911" rank="0" gdr="0">
<xgmi target="0000:26:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:62:00.0" class="0x020000" link_speed="5 GT/s" link_width="4">
<nic>
<net name="enp98s0f0" dev="0" speed="1000" port="0" guid="0x0" maxconn="65536" gdr="0"/>
</nic>
</pci>
</cpu>
<cpu numaid="1" affinity="00000000,00000000,ffffffff,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:21:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:23:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="90" gcn="906" arch="38911" rank="1" gdr="0">
<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="0000:24:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:26:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="90" gcn="906" arch="38911" rank="2" gdr="0">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c6:00.0" 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="90" gcn="906" arch="38911" rank="3" gdr="0">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="2" affinity="00000000,ffffffff,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="1" sm="96" gcn="906" arch="38911" rank="1" gdr="1">
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<gpu dev="4" sm="90" gcn="906" arch="38911" rank="4" gdr="0">
<xgmi target="0000:23:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c4:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c6:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="90" gcn="906" arch="38911" rank="5" gdr="0">
<xgmi target="0000:26:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:e1:00.0" class="0x020700" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0x68ce600003da341c" maxconn="262144" gdr="1"/>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0x20cd600003da341c" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
<cpu numaid="3" affinity="ffffffff,00000000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="2" sm="96" gcn="906" arch="38911" rank="2" gdr="1">
<gpu dev="6" sm="90" gcn="906" arch="38911" rank="6" gdr="0">
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
@@ -33,45 +74,9 @@
</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="3" sm="96" gcn="906" arch="38911" rank="3" gdr="1">
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<gpu dev="7" sm="90" gcn="906" arch="38911" rank="7" gdr="0">
<xgmi target="0000:23:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="1" affinity="00000000,00000000,ffffffff,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="4" sm="96" gcn="906" arch="38911" rank="4" gdr="1">
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:24:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:26:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="96" gcn="906" arch="38911" rank="6" gdr="1">
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:21:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:23:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="96" gcn="906" arch="38911" rank="7" gdr="1">
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="0" affinity="00000000,00000000,00000000,ffffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="5" sm="96" gcn="906" arch="38911" rank="5" gdr="1">
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:26:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
tools/topo_expl/models/topo_8p_rome_n4.xml
+85
View File
@@ -0,0 +1,85 @@
<system version="2">
<cpu numaid="1" affinity="00000000,00000000,00000000,ffff0000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="0" sm="90" gcn="906" arch="38911" rank="0" gdr="0">
<xgmi target="0000:26:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="2" affinity="00000000,00000000,0000ffff,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:21:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:23:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="90" gcn="906" arch="38911" rank="1" gdr="0">
<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="0000:24:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:26:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="90" gcn="906" arch="38911" rank="2" gdr="0">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="3" affinity="00000000,00000000,ffff0000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="90" gcn="906" arch="38911" rank="3" gdr="0">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="5" affinity="00000000,ffff0000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="4" sm="90" gcn="906" arch="38911" rank="4" gdr="0">
<xgmi target="0000:23:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c4:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c6:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="90" gcn="906" arch="38911" rank="5" gdr="0">
<xgmi target="0000:26:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="6" affinity="0000ffff,00000000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="90" gcn="906" arch="38911" rank="6" gdr="0">
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="7" affinity="ffff0000,00000000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<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="90" gcn="906" arch="38911" rank="7" gdr="0">
<xgmi target="0000:23:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="4" affinity="00000000,0000ffff,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:e1:00.0" class="0x020700" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0x68ce600003da341c" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
tools/topo_expl/topo_expl.cpp
+2 -1
View File
@@ -92,7 +92,7 @@ NodeModelDesc model_descs[] = {
{4, "topo_8p6l_6nic.xml", "4 nodes gfx908 8P6L 6 NICs"},
{1, "topo_8p_rome_n2.xml", "single node 8 VEGA20 Rome NPS=2"},
{4, "topo_8p_rome_n2.xml", "4 nodes 8 VEGA20 Rome NPS=2"},
{4, "topo_8p_rome_n2_1.xml", "4 nodes 8 VEGA20 Rome NPS=2 Alt. Model"},
{1, "topo_8p_rome_n2_1.xml", "single node 8 VEGA20 Rome NPS=2 Alt. Model"},
{1, "topo_8p_ts1.xml", "single node 8 VEGA20 TS1"},
{4, "topo_8p_ts1.xml", "4 nodes 8 VEGA20 TS1"},
{1, "topo_8p_ts1_1.xml", "single node 8 VEGA20 TS1 Alt. Model"},
@@ -107,6 +107,7 @@ NodeModelDesc model_descs[] = {
{4, "topo_4p3l_ia.xml", "4 nodes 8 gfx908"},
{4, "topo_8p_rome_n2_2.xml", "4 nodes 8 VEGA20 Rome NPS=2 Alt. Model 2 NET/IF"},
{4, "topo_8p_ts1_n4_2.xml", "4 nodes 8 VEGA20 TS1 NPS=4 3 NET/IF"},
{1, "topo_8p_rome_n4.xml", "single node 8 VEGA20 Rome NPS=4"},
};
int main(int argc,char* argv[])