Implement NIC identification and remapping (#420)

* Add 1H16P GPU model

* Implement NIC identification and remapping

* Revert "Sort IB devices based on device name (#413)"

This reverts commit 2d0ed8dff6.

* Fix permute and check order

* Correction on IB speed reporting

* Revert "Allow user to link layer with RCCL_IB_HCA_SKIP_LINK_LAYER (#361)"

This reverts commit caf5c9992a.
This commit is contained in:
Wenkai Du
2021-08-24 09:42:04 -07:00
committed by GitHub
parent 5f15ed6e3e
commit 5c8380ff5b
11 changed files with 623 additions and 47 deletions
+1 -1
View File
@@ -21,7 +21,7 @@
DIR="$(cd -P "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
for i in {0..55}
for i in {0..57}
do
if [[ $i -eq 50 ]] || [[ $i -eq 51 ]]
then
+191
View File
@@ -0,0 +1,191 @@
<system version="2">
<cpu numaid="0" affinity="00000000,00000000,00000000,ffffffff,00000000,00000000,00000000,ffffffff" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:41:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:49:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:4c:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:4e:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="0" sm="90" gcn="910" arch="38911" rank="0" gdr="0">
<xgmi target="0000:51:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:56:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0e:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:4f:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:51:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="1" sm="90" gcn="910" arch="38911" rank="1" gdr="0">
<xgmi target="0000:4e:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:56:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:59:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:11:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:52:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:54:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:56:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="2" sm="90" gcn="910" arch="38911" rank="2" gdr="0">
<xgmi target="0000:4e:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:51:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:59:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:16:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:57:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:59:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="3" sm="90" gcn="910" arch="38911" rank="3" gdr="0">
<xgmi target="0000:51:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:56:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:19:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
</pci>
</cpu>
<cpu numaid="1" affinity="00000000,00000000,ffffffff,00000000,00000000,00000000,ffffffff,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:01:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:09:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:0c:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:0e:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="4" sm="90" gcn="910" arch="38911" rank="4" gdr="0">
<xgmi target="0000:4e:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:11:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8f:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:0f:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:11:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="5" sm="90" gcn="910" arch="38911" rank="5" gdr="0">
<xgmi target="0000:51:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0e:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:92:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:97:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:0b:00.0" class="0x020000" vendor="0x8086" device="0x10d3" subsystem_vendor="0x8086" subsystem_device="0xa01f" link_speed="2.5 GT/s" link_width="1">
<nic>
<net name="enp11s0" dev="0" speed="1000" port="0" guid="0x0" maxconn="65536" gdr="0"/>
</nic>
</pci>
</pci>
<pci busid="0000:12:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:14:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:16:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="6" sm="90" gcn="910" arch="38911" rank="6" gdr="0">
<xgmi target="0000:56:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:19:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:92:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:97:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:17:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:19:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="7" sm="90" gcn="910" arch="38911" rank="7" gdr="0">
<xgmi target="0000:59:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:16:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:9a:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
</pci>
</cpu>
<cpu numaid="2" affinity="00000000,ffffffff,00000000,00000000,00000000,ffffffff,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:c2:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:ca:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:cd:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:cf:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="8" sm="90" gcn="910" arch="38911" rank="8" gdr="0">
<xgmi target="0000:d2:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d7:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8f:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:d0:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:d2:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="9" sm="90" gcn="910" arch="38911" rank="9" gdr="0">
<xgmi target="0000:cf:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d7:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:da:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:92:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:d3:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:d5:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:d7:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="10" sm="90" gcn="910" arch="38911" rank="10" gdr="0">
<xgmi target="0000:cf:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d2:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:da:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:97:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:d8:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:da:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="11" sm="90" gcn="910" arch="38911" rank="11" gdr="0">
<xgmi target="0000:d2:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d7:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:9a:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
</pci>
</cpu>
<cpu numaid="3" affinity="ffffffff,00000000,00000000,00000000,ffffffff,00000000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:82:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:8a:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:8d:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:8f:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="12" sm="90" gcn="910" arch="38911" rank="12" gdr="0">
<xgmi target="0000:0e:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:cf:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:92:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:90:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:92:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="13" sm="90" gcn="910" arch="38911" rank="13" gdr="0">
<xgmi target="0000:11:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:16:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d2:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8f:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:93:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16 GT/s" link_width="16">
<pci busid="0000:95:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:97:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="14" sm="90" gcn="910" arch="38911" rank="14" gdr="0">
<xgmi target="0000:11:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:16:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d7:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:9a:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:98:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="Unknown speed" link_width="8">
<pci busid="0000:9a:00.0" class="0x038000" vendor="0x1002" device="0x740c" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="Unknown speed" link_width="16">
<gpu dev="15" sm="90" gcn="910" arch="38911" rank="15" gdr="0">
<xgmi target="0000:19:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:da:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:97:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
</pci>
</cpu>
</system>
@@ -0,0 +1,126 @@
<system version="2">
<cpu numaid="1" affinity="00000000,00000000,00000000,ffff0000,00000000,00000000,00000000,ffff0000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:43:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:46:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:48:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:4a:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="90" gcn="908" arch="38911" rank="0" gdr="1">
<xgmi target="0000:50:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0f:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:4c:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:4e:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:50:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="90" gcn="908" arch="38911" rank="1" gdr="1">
<xgmi target="0000:4a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0f:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:45:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x1590" subsystem_device="0x0345" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_1" dev="3" speed="100000" port="1" guid="0x181515ffffa4e988" maxconn="262144" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="3" affinity="00000000,00000000,ffff0000,00000000,00000000,00000000,ffff0000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:03:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:05:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:08:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:0a:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="90" gcn="908" arch="38911" rank="2" gdr="1">
<xgmi target="0000:4a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:50:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0f:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:0b:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:0d:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:0f:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="90" gcn="908" arch="38911" rank="3" gdr="1">
<xgmi target="0000:4a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:50:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0a:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:13:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x1590" subsystem_device="0x0345" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_3" dev="2" speed="100000" port="1" guid="0x301515ffffa4e988" maxconn="262144" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="5" affinity="00000000,ffff0000,00000000,00000000,00000000,ffff0000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:c3:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c6:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c8:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:ca:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="90" gcn="908" arch="38911" rank="4" gdr="1">
<xgmi target="0000:d0:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:90:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:cc:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:ce:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:d0:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="90" gcn="908" arch="38911" rank="5" gdr="1">
<xgmi target="0000:ca:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:90:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:c5:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x1590" subsystem_device="0x0345" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_5" dev="1" speed="100000" port="1" guid="0x601515ffffa4e988" maxconn="262144" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="7" affinity="ffff0000,00000000,00000000,00000000,ffff0000,00000000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:83:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:86:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:88:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:8a:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="90" gcn="908" arch="38911" rank="6" gdr="1">
<xgmi target="0000:ca:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d0:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:90:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:8c:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:8e:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:90:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="90" gcn="908" arch="38911" rank="7" gdr="1">
<xgmi target="0000:ca:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d0:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8a:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:85:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x1590" subsystem_device="0x0345" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_7" dev="0" speed="100000" port="1" guid="0x64780affffa4e988" maxconn="262144" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
</system>
+3 -1
View File
@@ -124,7 +124,9 @@ NodeModelDesc model_descs[] = {
{1, "topo_8p_90a.xml", "single node gfx90a"},
{4, "topo_8p_rome_4n_1.xml", "4 nodes node 8 gfx908 Rome 4 NICs NPS=4"},
{4, "topo_collnet_n4.xml", "4 nodes collnet 4 NICs for multiple SAT"},
{1, "topo_8p_rome_vm1.xml", "single node 8 gfx908 Rome VM"},
{1, "topo_8p_rome_vm1.xml", "single node 8 gfx908 Rome VM"},
{1, "topo_16p1h.xml", "single node 16P1H"},
{4, "topo_8p_rome_4n_2.xml", "4 nodes 8 gfx908 Rome 4 NICs NPS=4 Alt. Model"},
};
int main(int argc,char* argv[])
+1 -1
View File
@@ -489,7 +489,7 @@ ncclResult_t initTransportsRank_1(struct ncclComm* comm, struct allGather1Data_t
if (comm->topo->nodes[GPU].count == comm->topo->nRanks && (comm->topo->type & RCCL_TOPO_CR8G))
allGather3Data[rank].nc = 4;
if (comm->topo->nodes[GPU].count == comm->topo->nRanks && comm->topo->nodes[GPU].nodes[idx].gpu.gcn == 910)
allGather3Data[rank].nc = 6;
allGather3Data[rank].nc = 4;
allGather3Data[rank].tree.pattern = treeGraph.pattern;
allGather3Data[rank].tree.nChannels = treeGraph.nChannels;
allGather3Data[rank].tree.sameChannels = treeGraph.sameChannels;