DGX SuperPOD Architecture
The DGX SuperPOD architecture is a combination of DGX systems, InfiniBand and Ethernet networking, management nodes, and storage. Figure 2 shows the rack layout of a single SU. In this example, power consumption per rack exceeds 25 kW. The rack layout can be adjusted to meet local data center requirements, such as maximum power per rack and rack layout between DGX systems and supporting equipment to meet local needs for power and cooling distribution.
Figure 3 shows an example management rack configuration with networking switches, management servers, storage arrays, and UFM appliances. Sizes and quantities will vary depending upon models used.
This reference architecture is focused on 4 SU units with 128 DGX nodes. DGX SuperPOD can scale to much larger configurations up to and beyond 64 SU with 2000+ DGX H200 nodes. See Table 3 for more information.
Table 3. Larger SuperPOD component counts
SU Count |
Node Count |
GPU Count |
InfiniBand Switch Counts |
Cable Counts |
||||
---|---|---|---|---|---|---|---|---|
Leaf |
Spine |
Core |
Node-Leaf |
Leaf-Spine |
Spine-Core |
|||
4 |
128 |
1024 |
32 |
16 |
– |
1024 |
1024 |
– |
8 |
256 |
2048 |
64 |
32 |
– |
2048 |
2048 |
– |
16 |
512 |
4096 |
128 |
128 |
64 |
4096 |
4096 |
4096 |
32 |
1024 |
8192 |
256 |
256 |
128 |
8192 |
8192 |
8192 |
64 |
2048 |
16384 |
512 |
512 |
256 |
16384 |
16384 |
16384 |
Contact NVIDIA for more information regarding DGX SuperPOD solutions beyond four scalable units.