AI/rocm-systems
2
0
Форкнуть 0
Код Задачи Запросы на слияние Действия Пакеты Проекты Релизы Вики Активность

Add 8P6L multi-node models (#239)

Просмотр исходного кода
Этот коммит содержится в:
Wenkai Du
2020-07-21 14:10:36 -07:00
коммит произвёл GitHub
родитель 684f3e6af4
Коммит d5f90e19b5
15 изменённых файлов: 830 добавлений и 32 удалений
Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы
src/graph/connect.cc
+3 -3
Просмотреть файл
@@ -9,6 +9,7 @@
#include "graph.h"
#include "trees.h"
#include "rings.h"
#include "topo.h"
/******************************************************************/
/********************* Internode connection ***********************/
@@ -288,9 +289,8 @@ ncclResult_t ncclTopoPostset(struct ncclComm* comm, int* firstRanks, struct nccl
memcpy(ringPrev+nChannels*nranks, ringPrev, nChannels*nranks*sizeof(int));
memcpy(ringNext+nChannels*nranks, ringNext, nChannels*nranks*sizeof(int));
char *str = NULL;
NCCLCHECK(parseChordalRing(comm->topo, &str));
int end = std::min((int)ncclMaxNchannels(), (str ? nChannels*3 : ncclMinNchannels()));
int end = std::min((int)ncclMaxNchannels(),
((comm->topo->nodes[NET].count == 0 && comm->topo->type == RCCL_TOPO_CR8G) ? nChannels*3 : ncclMinNchannels()));
// Duplication should be complete now
nChannels = comm->nChannels = std::min(MAXCHANNELS,nChannels*2);
src/graph/paths.cc
+6
Просмотреть файл
@@ -329,6 +329,12 @@ ncclResult_t ncclTopoCheckGdr(struct ncclTopoSystem* system, int64_t busId, int
// Check if we are close enough that it makes sense to enable GDR
int netGdrLevel = PATH_PXB;
#ifdef TOPO_EXPL
int arch, vendor, model;
NCCLCHECK(ncclTopoCpuType(system, &arch, &vendor, &model));
if (arch == NCCL_TOPO_CPU_ARCH_X86 && vendor == NCCL_TOPO_CPU_VENDOR_AMD && model == NCCL_TOPO_CPU_TYPE_ROME)
netGdrLevel = PATH_PHB;
#endif
NCCLCHECK(ncclGetLevel(&ncclTopoUserGdrLevel, NULL, "NCCL_NET_GDR_LEVEL"));
if (ncclTopoUserGdrLevel != -2) netGdrLevel = ncclTopoUserGdrLevel;
int distance = gpu->paths[NET][n].type;
src/graph/search.cc
+34 -16
Просмотреть файл
@@ -25,6 +25,7 @@ static float getMaxWidth(struct ncclTopoSystem* system, struct ncclTopoNode* gpu
}
ncclResult_t ncclTopoSearchInit(struct ncclTopoSystem* system) {
system->maxWidth = 0.0;
system->type = 0;
int inter = system->nodes[NET].count;
if (inter == 0 && system->nodes[GPU].count == 1) {
system->maxWidth = LOC_WIDTH;
@@ -714,30 +715,30 @@ end:
return ncclSuccess;
}
ncclResult_t parseChordalRing(struct ncclTopoSystem* system, char **str) {
static const char *ringBase = "0 6 7 4 5 3 2 1|0 5 6 3 7 1 4 2|0 4 6 2 7 5 1 3|0 1 2 3 5 4 7 6|0 2 4 1 7 3 6 5|0 3 1 5 7 2 6 4";
static ncclResult_t parseChordalRing(struct ncclTopoSystem* system, char **str) {
static const char *ringBase = "0 1 2 3 5 4 7 6|0 2 4 1 7 3 6 5|0 3 1 5 7 2 6 4|0 6 7 4 5 3 2 1|0 5 6 3 7 1 4 2|0 4 6 2 7 5 1 3";
static char ringRemap[256];
int id[8], dist[8];
int i;
*str = 0;
int ngpus = system->nodes[GPU].count;
// single node CR8G only
if (ngpus != 8 || system->nodes[NET].count != 0)
if (ngpus != 8)
return ncclSuccess;
// validate chordal ring and calculate distance
for (i=0; i<ngpus; i++) {
struct ncclTopoNode* node = system->nodes[GPU].nodes+i;
if (node->paths[GPU] == NULL) continue;
int sum = ngpus*(ngpus-1)/2 - node->gpu.rank;
int sum = ngpus*(ngpus-1)/2 - node->gpu.dev;
int count = 0;
for (int n = 0; n<ngpus; n++) {
struct ncclTopoLink* link;
for (link = node->links; link->remNode; link++) {
if (link->remNode->gpu.rank == n) break;
if (link->remNode->gpu.dev == n) break;
}
if (!link->remNode) continue;
if (link->type != LINK_NVL) continue;
sum -= system->nodes[GPU].nodes[n].gpu.rank;
sum -= system->nodes[GPU].nodes[n].gpu.dev;
count ++;
}
if(count != ngpus-2 || sum < 0 || sum > ngpus-1) {
@@ -765,6 +766,7 @@ ncclResult_t parseChordalRing(struct ncclTopoSystem* system, char **str) {
}
ringRemap[i] = 0;
*str = ringRemap;
system->type = RCCL_TOPO_CR8G;
INFO(NCCL_GRAPH, "Use chordal ring: %s", ringRemap);
return ncclSuccess;
}
@@ -806,18 +808,34 @@ ncclResult_t ncclTopoCompute(ncclTopoSystem* system, struct ncclTopoGraph* graph
// Translate gpu numbers into ranks
graph->intra[i] = system->nodes[GPU].nodes[graph->intra[i]].gpu.rank;
}
// TODO : let user specify NICs
graph->inter[0] = graph->inter[1] = 0;
graph->speedIntra = graph->speedInter = system->maxWidth;
if (graph->pattern == NCCL_TOPO_PATTERN_RING) {
// Reverse the loop
for (int c=0; c<graph->nChannels; c++) {
for (int i=0; i<=ngpus/2; i++) {
int tmp = graph->intra[ngpus*c+i];
graph->intra[ngpus*c+i] = graph->intra[ngpus*c+(ngpus-i)%ngpus];
graph->intra[ngpus*c+ngpus-i] = tmp;
if (system->nodes[NET].count) {
int *intra, *used;
graph->nChannels = system->nodes[NET].count;
NCCLCHECK(ncclCalloc(&intra, ngpus));
NCCLCHECK(ncclCalloc(&used,system->nodes[NET].count));
for (int n = 0; n < system->nodes[NET].count; n++) {
graph->inter[n*2] = graph->inter[n*2+1] = n;
struct ncclTopoNode* net = system->nodes[NET].nodes+n;
struct ncclTopoLinkList* paths = net->paths[GPU];
// find the first unsed GPU that is closest to NIC
int f, m;
for (f = 0; f < ngpus; f++) {
int j = 0; for (j = 0; j < n; j++) if(used[j] == system->nodes[GPU].nodes[f].gpu.rank) break;
if(j >= n) break;
}
for (int i = 0; i < ngpus; i++) {
int j = 0; for (j = 0; j < n; j++) if(used[j] == system->nodes[GPU].nodes[i].gpu.rank) break;
if (j < n) continue;
if (paths[i].count < paths[f].count) f = i;
}
for (m = 0; m<ngpus; m++) if (graph->intra[n*ngpus+m] == system->nodes[GPU].nodes[f].gpu.rank) break;
used[n] = graph->intra[n*ngpus+m];
for (int i = 0; i < ngpus; i++) intra[i] = graph->intra[n*ngpus+((i+m)%ngpus)];
for (int i = 0; i < ngpus; i++) graph->intra[n*ngpus+i] = intra[i];
}
free(used);
free(intra);
}
if (graph->nChannels) return ncclSuccess;
}
src/graph/topo.cc
+6
Просмотреть файл
@@ -400,6 +400,12 @@ ncclResult_t ncclTopoAddCpu(struct ncclXmlNode* xmlCpu, struct ncclTopoSystem* s
NCCLCHECK(xmlGetAttrInt(xmlCpu, "modelid", &modelId));
cpu->cpu.model = (familyId == 6 && modelId >= 0x55) ? NCCL_TOPO_CPU_TYPE_SKL : NCCL_TOPO_CPU_INTEL_BDW;
}
if (cpu->cpu.vendor == NCCL_TOPO_CPU_VENDOR_AMD) {
int familyId, modelId;
NCCLCHECK(xmlGetAttrInt(xmlCpu, "familyid", &familyId));
NCCLCHECK(xmlGetAttrInt(xmlCpu, "modelid", &modelId));
cpu->cpu.model = (familyId == 143 && modelId >= 49) ? NCCL_TOPO_CPU_TYPE_ROME : NCCL_TOPO_CPU_TYPE_ZEN;
}
}
for (int s=0; s<xmlCpu->nSubs; s++) {
struct ncclXmlNode* node = xmlCpu->subs[s];
src/graph/topo.h
+3
Просмотреть файл
@@ -77,6 +77,8 @@ struct ncclTopoLinkList {
#define NCCL_TOPO_UNDEF (-1)
#define RCCL_TOPO_CR8G 1
struct ncclTopoNode {
int type;
int64_t id;
@@ -119,6 +121,7 @@ struct ncclTopoNodeSet {
struct ncclTopoSystem {
struct ncclTopoNodeSet nodes[NCCL_TOPO_NODE_TYPES];
float maxWidth;
int type;
};
ncclResult_t ncclTopoGetNode(struct ncclTopoSystem* system, struct ncclTopoNode** node, int type, uint64_t id);
src/include/graph.h
+2 -2
Просмотреть файл
@@ -42,6 +42,8 @@ ncclResult_t ncclTopoSetAffinity(struct ncclTopoSystem* system, int rank);
#define NCCL_TOPO_CPU_VENDOR_AMD 2
#define NCCL_TOPO_CPU_TYPE_BDW 1
#define NCCL_TOPO_CPU_TYPE_SKL 2
#define NCCL_TOPO_CPU_TYPE_ZEN 3
#define NCCL_TOPO_CPU_TYPE_ROME 4
ncclResult_t ncclTopoCpuType(struct ncclTopoSystem* system, int* arch, int* vendor, int* model);
#define NCCL_TOPO_MAX_NODES 256
@@ -101,6 +103,4 @@ ncclResult_t ncclTopoTuneModel(struct ncclComm* comm, int minCompCap, int maxCom
#include "info.h"
ncclResult_t ncclTopoGetAlgoTime(struct ncclInfo* info, int algorithm, int protocol, float* time);
ncclResult_t parseChordalRing(struct ncclTopoSystem* system, char **str);
#endif
tools/scripts/topo_val.sh
+1 -1
Просмотреть файл
@@ -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"
tools/topo_expl/models/topo_8p6l.xml
+10 -10
Просмотреть файл
@@ -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>
tools/topo_expl/models/topo_8p6l_1nic.xml
+107
Просмотреть файл
@@ -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>
tools/topo_expl/models/topo_8p6l_2nic.xml
+112
Просмотреть файл
@@ -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>
tools/topo_expl/models/topo_8p6l_3nic.xml
+117
Просмотреть файл
@@ -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>
tools/topo_expl/models/topo_8p6l_4nic.xml
+122
Просмотреть файл
@@ -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>
tools/topo_expl/models/topo_8p6l_5nic.xml
+127
Просмотреть файл
@@ -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>
tools/topo_expl/models/topo_8p6l_6nic.xml
+132
Просмотреть файл
@@ -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>
tools/topo_expl/topo_expl.cpp
+48
Просмотреть файл
@@ -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);