Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
tsoc
hg-misc-tools
Commits
520ac826
Commit
520ac826
authored
Mar 08, 2026
by
one
Browse files
[rccl-tests] Add scripts and topo files for BW1000
parent
b142b6fb
Changes
6
Show whitespace changes
Inline
Side-by-side
Showing
6 changed files
with
611 additions
and
0 deletions
+611
-0
projects/rccl-tests/bw1000/graph-16r-allreduce.xml
projects/rccl-tests/bw1000/graph-16r-allreduce.xml
+201
-0
projects/rccl-tests/bw1000/mpirun_rccltest
projects/rccl-tests/bw1000/mpirun_rccltest
+155
-0
projects/rccl-tests/bw1000/run-allreduce.sh
projects/rccl-tests/bw1000/run-allreduce.sh
+42
-0
projects/rccl-tests/bw1000/run-alltoall.sh
projects/rccl-tests/bw1000/run-alltoall.sh
+39
-0
projects/rccl-tests/bw1000/run-p2p.sh
projects/rccl-tests/bw1000/run-p2p.sh
+28
-0
projects/rccl-tests/bw1000/topo-gdr-bw1000.xml
projects/rccl-tests/bw1000/topo-gdr-bw1000.xml
+146
-0
No files found.
projects/rccl-tests/bw1000/graph-16r-allreduce.xml
0 → 100644
View file @
520ac826
<graphs
version=
"1"
>
<graph
id=
"0"
pattern=
"4"
crossnic=
"0"
nchannels=
"8"
speedintra=
"24"
speedinter=
"24"
latencyinter=
"0"
typeintra=
"NVL"
typeinter=
"PXB"
samechannels=
"0"
>
<channel>
<net
dev=
"0"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x1"
/>
<net
dev=
"0"
/>
</channel>
<channel>
<net
dev=
"0x1"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<net
dev=
"0x1"
/>
</channel>
<channel>
<net
dev=
"0x2"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x3"
/>
<net
dev=
"0x2"
/>
</channel>
<channel>
<net
dev=
"0x3"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<net
dev=
"0x3"
/>
</channel>
<channel>
<net
dev=
"0x4"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x5"
/>
<net
dev=
"0x4"
/>
</channel>
<channel>
<net
dev=
"0x5"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<net
dev=
"0x5"
/>
</channel>
<channel>
<net
dev=
"0x6"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x7"
/>
<net
dev=
"0x6"
/>
</channel>
<channel>
<net
dev=
"0x7"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<net
dev=
"0x7"
/>
</channel>
</graph>
<graph
id=
"1"
pattern=
"1"
crossnic=
"0"
nchannels=
"8"
speedintra=
"24"
speedinter=
"24"
latencyinter=
"0"
typeintra=
"NVL"
typeinter=
"PXB"
samechannels=
"0"
>
<channel>
<net
dev=
"0"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<net
dev=
"0"
/>
</channel>
<channel>
<net
dev=
"0x1"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<net
dev=
"0x1"
/>
</channel>
<channel>
<net
dev=
"0x2"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<net
dev=
"0x2"
/>
</channel>
<channel>
<net
dev=
"0x3"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<net
dev=
"0x3"
/>
</channel>
<channel>
<net
dev=
"0x4"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<net
dev=
"0x4"
/>
</channel>
<channel>
<net
dev=
"0x5"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<net
dev=
"0x5"
/>
</channel>
<channel>
<net
dev=
"0x6"
/>
<gpu
dev=
"0x6"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<net
dev=
"0x6"
/>
</channel>
<channel>
<net
dev=
"0x7"
/>
<gpu
dev=
"0x7"
/>
<gpu
dev=
"0"
/>
<gpu
dev=
"0x1"
/>
<gpu
dev=
"0x2"
/>
<gpu
dev=
"0x3"
/>
<gpu
dev=
"0x4"
/>
<gpu
dev=
"0x5"
/>
<gpu
dev=
"0x6"
/>
<net
dev=
"0x7"
/>
</channel>
</graph>
<graph
id=
"2"
pattern=
"6"
crossnic=
"0"
nchannels=
"0"
speedintra=
"0"
speedinter=
"0"
latencyinter=
"0"
typeintra=
"LOC"
typeinter=
"LOC"
samechannels=
"0"
/>
<graph
id=
"2"
pattern=
"3"
crossnic=
"0"
nchannels=
"0"
speedintra=
"0"
speedinter=
"0"
latencyinter=
"0"
typeintra=
"LOC"
typeinter=
"LOC"
samechannels=
"0"
/>
<graph
id=
"3"
pattern=
"5"
crossnic=
"0"
nchannels=
"0"
speedintra=
"0"
speedinter=
"0"
latencyinter=
"0"
typeintra=
"LOC"
typeinter=
"LOC"
samechannels=
"0"
/>
</graphs>
projects/rccl-tests/bw1000/mpirun_rccltest
0 → 100644
View file @
520ac826
#!/bin/bash
set
-e
# =================================================
# Helper functions
# =================================================
help
()
{
cat
<<
EOF
RCCL Tests MPI run helper script
Usage:
$(
basename
"
$0
"
)
[OPTIONS]
OPTIONS:
-h, --help Show this help message and exit
-np Total number of processes (default: sum of per-node counts in --hosts)
-H, --hosts Comma-separated list of nodes with optional process count per node
Format: node01:8,node02:8
If count is omitted, falls back to auto-detected GPU count per node.
--tcp-iface TCP interface to use for communication (default:
${
tcp_iface
}
)
--ssh-port SSH port to use for remote connections (default:
${
ssh_port
}
)
EOF
}
# =================================================
# Global variables
# =================================================
np
=
hosts_raw
=
tcp_iface
=
p14p2
ssh_port
=
3333
rccltest_runscript
=
"
${
PWD
}
/run_rccltest"
rccltest_args
=()
mpi_bin
=
/opt/mpi/bin/mpirun
ompi_prefix
=
/opt/mpi
# Detect the number of GPUs per node (used as fallback when count is not specified in --hosts)
ngpu_per_node
=
$(
hy-smi
--showid
2>/dev/null |
grep
-ic
"Device ID"
)
if
[[
-z
"
${
ngpu_per_node
}
"
||
"
${
ngpu_per_node
}
"
-eq
0
]]
;
then
echo
"[WRAPPER] Failed to get the number of GPUs per node via hy-smi. Defaulting to 8."
ngpu_per_node
=
8
else
echo
"[WRAPPER] Detected
${
ngpu_per_node
}
GPUs per node."
fi
# =================================================
# Parameter parsing
# =================================================
while
[[
$#
-gt
0
]]
;
do
case
"
${
1
}
"
in
-h
|
--help
)
help
;
exit
0
;;
-np
)
np
=
${
2
}
;
shift
2
;;
-H
|
--hosts
)
hosts_raw
=
${
2
}
;
shift
2
;;
--tcp-iface
)
tcp_iface
=
${
2
}
;
shift
2
;;
--ssh-port
)
ssh_port
=
${
2
}
;
shift
2
;;
--
)
shift
;
rccltest_args+
=(
"
$@
"
)
;
break
;;
*
)
rccltest_args+
=(
"
${
1
}
"
)
;
shift
;;
esac
done
# =================================================
# Parse hosts into parallel arrays: node_names[], node_slots[]
# Input format: node01:8,node02:8 (count optional, falls back to ngpu_per_node)
# =================================================
parse_hosts
()
{
node_names
=()
node_slots
=()
IFS
=
','
read
-ra
entries
<<<
"
${
hosts_raw
}
"
for
entry
in
"
${
entries
[@]
}
"
;
do
local
name
=
"
${
entry
%%
:
*
}
"
local
slots
=
"
${
entry
##*
:
}
"
# If no ':' was present, entry == name == slots
if
[[
"
${
entry
}
"
!=
*
:
*
]]
;
then
slots
=
"
${
ngpu_per_node
}
"
fi
node_names+
=(
"
${
name
}
"
)
node_slots+
=(
"
${
slots
}
"
)
done
}
# =================================================
# Run rccl test script
# =================================================
if
[
-z
"
${
hosts_raw
}
"
]
;
then
# Run single-node test if --hosts is not set
echo
"[WRAPPER] No compute nodes specified. Running in single-node mode."
# Default np to ngpu_per_node when not set
np
=
"
${
np
:-${
ngpu_per_node
}}
"
echo
"Using np=
${
np
}
"
${
mpi_bin
}
--allow-run-as-root
\
--bind-to
none
\
--mca
pml ucx
\
--mca
osc ucx
\
--mca
btl ^vader,tcp,openib,uct
\
--mca
coll ^hcoll
\
$(
env
|
grep
-E
'^(NCCL|RCCL|UCX|HSA)_'
|
cut
-d
=
-f1
|
awk
'{print "-x", $1}'
)
\
-np
${
np
}
\
${
rccltest_runscript
}
"
${
rccltest_args
[@]
}
"
else
# Multi-node mode
echo
"[WRAPPER] Running in multi-node mode."
parse_hosts
# Build MPI -H string and auto-sum np
hosts_string
=
""
np_sum
=
0
for
i
in
"
${
!node_names[@]
}
"
;
do
hosts_string+
=
"
${
node_names
[
$i
]
}
:
${
node_slots
[
$i
]
}
,"
((
np_sum +
=
node_slots[
$i
]
))
done
hosts_string
=
"
${
hosts_string
%,
}
"
# -np overrides auto-sum if explicitly provided
np
=
"
${
np
:-${
np_sum
}}
"
echo
"[WRAPPER] MPI hosts:
${
hosts_string
}
"
echo
"[WRAPPER] Total processes (np):
${
np
}
"
echo
"[WRAPPER] Using TCP interface:
${
tcp_iface
}
"
echo
"[WRAPPER] Using SSH port:
${
ssh_port
}
"
# Copy files to remote nodes (skip current node)
current_node
=
$(
hostname
)
copyto_hosts
=()
for
name
in
"
${
node_names
[@]
}
"
;
do
if
[[
"
${
name
}
"
!=
"
${
current_node
}
"
]]
;
then
copyto_hosts+
=(
"
${
name
}
"
)
fi
done
if
[
${#
copyto_hosts
[@]
}
-gt
0
]
;
then
echo
"[WRAPPER] Copying files to remote nodes in parallel:
${
copyto_hosts
[*]
}
"
for
node
in
"
${
copyto_hosts
[@]
}
"
;
do
rsync
-az
-e
"ssh -p
${
ssh_port
}
"
${
PWD
}
/build
${
PWD
}
/scripts
${
rccltest_runscript
}
${
NCCL_TOPO_FILE
}
${
NCCL_GRAPH_FILE
}
"
${
node
}
:
${
PWD
}
/"
&
rsync
-az
-e
"ssh -p
${
ssh_port
}
"
/opt/dtk/rccl/lib
${
node
}
:/opt/dtk/rccl/ &
rsync
-az
-e
"ssh -p
${
ssh_port
}
"
/opt/mpi /opt/ucx
${
node
}
:/opt/ &
done
wait
echo
"[WRAPPER] Files synchronized successfully."
fi
${
mpi_bin
}
--allow-run-as-root
\
--prefix
${
ompi_prefix
}
\
--bind-to
none
\
--mca
pml ucx
\
--mca
btl_tcp_if_include
${
tcp_iface
}
\
--mca
plm_rsh_args
"-p
${
ssh_port
}
"
\
$(
env
|
grep
-E
'^(NCCL|RCCL|UCX|HSA|HIP)_'
|
cut
-d
=
-f1
|
awk
'{print "-x", $1}'
)
\
-x
PATH
-x
LD_LIBRARY_PATH
\
-np
${
np
}
\
-H
${
hosts_string
}
\
"
${
rccltest_args
[@]
}
"
fi
projects/rccl-tests/bw1000/run-allreduce.sh
0 → 100644
View file @
520ac826
#!/bin/bash
set
-e
SSH_PORT
=
3333
unset
UCX_HOME
# export UCX_LOG_LEVEL=fatal
export
NCCL_TOPO_DUMP_FILE
=
${
PWD
}
/topo-generated.xml
export
NCCL_GRAPH_DUMP_FILE
=
${
PWD
}
/graph-generated.xml
# export NCCL_DEBUG=INFO
# export NCCL_DEBUG_SUBSYS=ALL
#export RCCL_SDMA_COPY_ENABLE=1
#export RCCL_SDMA_LINK_MODE=0
# PCIe混合链路
# export NCCL_SIMPLE_CHANNELS=32
# export RCCL_P2P_XHCL_CHANNEL_NUM=31
# export RCCL_COLL_XHCL_CHANNEL_NUM=28
export
HSA_FORCE_FINE_GRAIN_PCIE
=
1
export
NCCL_SOCKET_IFNAME
=
p14p2
export
NCCL_IB_HCA
=
"=mlx5_1,mlx5_2,mlx5_3,mlx5_4,mlx5_7,mlx5_8,mlx5_9,mlx5_10"
export
NCCL_NET_GDR_LEVEL
=
SYS
export
NCCL_NET_GDR_READ
=
1
# export NCCL_ALGO=Ring
# export NCCL_PROTO=Simple
export
NCCL_SIMPLE_CHANNELS
=
32
unset
NCCL_NCHANNELS_PER_PEER
export
NCCL_TOPO_FILE
=
${
PWD
}
/topo-gdr-bw1000.xml
# export NCCL_GRAPH_FILE=${PWD}/graph-16r-allreduce.xml
./mpirun_rccltest
-np
2
\
./build/all_reduce_perf
-b
4
-e
16G
-f
2
-w
3
-g
1
./mpirun_rccltest
-np
4
\
./build/all_reduce_perf
-b
4
-e
16G
-f
2
-w
3
-g
1
./mpirun_rccltest
-np
8
\
./build/all_reduce_perf
-b
4
-e
16G
-f
2
-w
3
-g
1
./mpirun_rccltest
-np
16
-H
node01,node02
--ssh-port
${
SSH_PORT
}
\
./build/all_reduce_perf
-b
4
-e
16G
-f
2
-w
3
-g
1
./mpirun_rccltest
-np
32
-H
node01,node02,node03,node04
--ssh-port
${
SSH_PORT
}
\
./build/all_reduce_perf
-b
4
-e
16G
-f
2
-w
3
-g
1
projects/rccl-tests/bw1000/run-alltoall.sh
0 → 100644
View file @
520ac826
#!/bin/bash
set
-e
SSH_PORT
=
3333
unset
UCX_HOME
# export UCX_LOG_LEVEL=fatal
export
NCCL_TOPO_DUMP_FILE
=
${
PWD
}
/topo-generated.xml
export
NCCL_GRAPH_DUMP_FILE
=
${
PWD
}
/graph-generated.xml
# export NCCL_DEBUG=INFO
# export NCCL_DEBUG_SUBSYS=ALL
#export RCCL_SDMA_COPY_ENABLE=1
#export RCCL_SDMA_LINK_MODE=0
# PCIe混合链路
# export NCCL_SIMPLE_CHANNELS=32
# export RCCL_P2P_XHCL_CHANNEL_NUM=31
# export RCCL_COLL_XHCL_CHANNEL_NUM=28
export
HSA_FORCE_FINE_GRAIN_PCIE
=
1
export
NCCL_SOCKET_IFNAME
=
p14p2
export
NCCL_IB_HCA
=
"=mlx5_1,mlx5_2,mlx5_3,mlx5_4,mlx5_7,mlx5_8,mlx5_9,mlx5_10"
export
NCCL_P2P_LEVEL
=
SYS
export
NCCL_NET_GDR_LEVEL
=
PHB
export
NCCL_NET_GDR_READ
=
1
unset
NCCL_NCHANNELS_PER_PEER
export
NCCL_TOPO_FILE
=
${
PWD
}
/topo-gdr-bw1000.xml
./mpirun_rccltest
-np
2
\
./build/alltoall_perf
-b
32
-e
16G
-f
2
-w
3
-g
1
./mpirun_rccltest
-np
4
\
./build/alltoall_perf
-b
64
-e
16G
-f
2
-w
3
-g
1
./mpirun_rccltest
-np
8
\
./build/alltoall_perf
-b
128
-e
16G
-f
2
-w
3
-g
1
./mpirun_rccltest
-np
16
-H
node01,node02
--ssh-port
${
SSH_PORT
}
\
./build/alltoall_perf
-b
256
-e
16G
-f
2
-w
3
-g
1
./mpirun_rccltest
-np
32
-H
node01,node02,node03,node04
--ssh-port
${
SSH_PORT
}
\
./build/alltoall_perf
-b
512
-e
16G
-f
2
-w
3
-g
1
projects/rccl-tests/bw1000/run-p2p.sh
0 → 100644
View file @
520ac826
#!/bin/bash
set
-e
SSH_PORT
=
3333
unset
UCX_HOME
export
NCCL_TOPO_DUMP_FILE
=
${
PWD
}
/topo-generated.xml
export
NCCL_GRAPH_DUMP_FILE
=
${
PWD
}
/graph-generated.xml
# export NCCL_DEBUG=INFO
# export NCCL_DEBUG_SUBSYS=ALL
export
HSA_FORCE_FINE_GRAIN_PCIE
=
1
export
NCCL_SOCKET_IFNAME
=
p14p2
export
NCCL_IB_HCA
=
"=mlx5_1,mlx5_2,mlx5_3,mlx5_4,mlx5_7,mlx5_8,mlx5_9,mlx5_10"
export
NCCL_P2P_LEVEL
=
SYS
export
NCCL_NET_GDR_LEVEL
=
PHB
export
NCCL_NET_GDR_READ
=
1
unset
NCCL_NCHANNELS_PER_PEER
export
NCCL_TOPO_FILE
=
${
PWD
}
/topo-gdr-bw1000.xml
for
g
in
{
0..7
}
;
do
echo
echo
"Running with GPU
${
g
}
"
export
HIP_VISIBLE_DEVICES
=
${
g
}
./mpirun_rccltest
-np
2
-H
node01:1,node02:1
--ssh-port
${
SSH_PORT
}
\
./build/sendrecv_perf
-b
2G
-e
2G
-f
2
-w
3
-g
1
echo
done
projects/rccl-tests/bw1000/topo-gdr-bw1000.xml
0 → 100644
View file @
520ac826
<system
version=
"2"
>
<cpu
numaid=
"3"
>
<pci
busid=
"0000:99:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:9b:00.0"
class=
"0x020700"
>
<nic>
<net
name=
"mlx5_1"
dev=
"0"
gdr=
"1"
/>
<net
name=
"mlx5_2"
dev=
"1"
gdr=
"1"
/>
</nic>
</pci>
<pci
busid=
"0000:9d:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:9f:00.0"
class=
"0x0b4000"
>
<gpu
dev=
"0"
sm=
"80"
gcn=
"gfx936"
arch=
"169983"
rank=
"0"
gdr=
"1"
>
<xgmi
target=
"0000:20:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:10:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:13:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:16:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:19:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1c:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1f:00.0"
count=
"1"
tclass=
"0x068000"
/>
</gpu>
</pci>
</pci>
<pci
busid=
"0000:54:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:56:00.0"
class=
"0x0b4000"
>
<gpu
dev=
"1"
sm=
"80"
gcn=
"gfx936"
arch=
"169983"
rank=
"1"
gdr=
"1"
>
<xgmi
target=
"0000:20:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:10:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:13:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:16:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:19:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1c:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1f:00.0"
count=
"1"
tclass=
"0x068000"
/>
</gpu>
</pci>
</pci>
</pci>
</cpu>
<cpu
numaid=
"0"
>
<pci
busid=
"0000:01:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:06:00.0"
class=
"0x020700"
>
<nic>
<net
name=
"mlx5_3"
dev=
"2"
gdr=
"1"
/>
<net
name=
"mlx5_4"
dev=
"3"
gdr=
"1"
/>
</nic>
</pci>
<pci
busid=
"0000:03:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:05:00.0"
class=
"0x0b4000"
>
<gpu
dev=
"3"
sm=
"80"
gcn=
"gfx936"
arch=
"169983"
rank=
"3"
gdr=
"1"
>
<xgmi
target=
"0000:20:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:10:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:13:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:16:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:19:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1c:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1f:00.0"
count=
"1"
tclass=
"0x068000"
/>
</gpu>
</pci>
</pci>
<pci
busid=
"0000:5b:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:5d:00.0"
class=
"0x0b4000"
>
<gpu
dev=
"2"
sm=
"80"
gcn=
"gfx936"
arch=
"169983"
rank=
"2"
gdr=
"1"
>
<xgmi
target=
"0000:20:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:10:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:13:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:16:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:19:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1c:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1f:00.0"
count=
"1"
tclass=
"0x068000"
/>
</gpu>
</pci>
</pci>
</pci>
</cpu>
<cpu
numaid=
"7"
>
<pci
busid=
"0000:e4:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:e9:00.0"
class=
"0x020700"
>
<nic>
<net
name=
"mlx5_7"
dev=
"4"
gdr=
"1"
/>
<net
name=
"mlx5_8"
dev=
"5"
gdr=
"1"
/>
</nic>
</pci>
<pci
busid=
"0000:e6:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:e8:00.0"
class=
"0x0b4000"
>
<gpu
dev=
"4"
sm=
"80"
gcn=
"gfx936"
arch=
"169983"
rank=
"4"
gdr=
"1"
>
<xgmi
target=
"0000:20:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:10:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:13:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:16:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:19:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1c:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1f:00.0"
count=
"1"
tclass=
"0x068000"
/>
</gpu>
</pci>
</pci>
<pci
busid=
"0000:bf:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:c1:00.0"
class=
"0x0b4000"
>
<gpu
dev=
"5"
sm=
"80"
gcn=
"gfx936"
arch=
"169983"
rank=
"5"
gdr=
"1"
>
<xgmi
target=
"0000:20:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:10:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:13:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:16:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:19:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1c:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1f:00.0"
count=
"1"
tclass=
"0x068000"
/>
</gpu>
</pci>
</pci>
</pci>
</cpu>
<cpu
numaid=
"4"
>
<pci
busid=
"0000:ab:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:ad:00.0"
class=
"0x020700"
>
<nic>
<net
name=
"mlx5_9"
dev=
"6"
gdr=
"1"
/>
<net
name=
"mlx5_10"
dev=
"7"
gdr=
"1"
/>
</nic>
</pci>
<pci
busid=
"0000:af:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:b1:00.0"
class=
"0x0b4000"
>
<gpu
dev=
"7"
sm=
"80"
gcn=
"gfx936"
arch=
"169983"
rank=
"7"
gdr=
"1"
>
<xgmi
target=
"0000:20:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:10:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:13:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:16:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:19:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1c:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1f:00.0"
count=
"1"
tclass=
"0x068000"
/>
</gpu>
</pci>
</pci>
<pci
busid=
"0000:c8:00.0"
class=
"0x060400"
>
<pci
busid=
"0000:ca:00.0"
class=
"0x0b4000"
>
<gpu
dev=
"6"
sm=
"80"
gcn=
"gfx936"
arch=
"169983"
rank=
"6"
gdr=
"1"
>
<xgmi
target=
"0000:20:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:10:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:13:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:16:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:19:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1c:00.0"
count=
"1"
tclass=
"0x068000"
/>
<xgmi
target=
"0000:1f:00.0"
count=
"1"
tclass=
"0x068000"
/>
</gpu>
</pci>
</pci>
</pci>
</cpu>
</system>
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment