DGX SuperPOD Architecture#

The DGX SuperPOD architecture is a combination of DGX systems, Ethernet networking, InfiniBand Networking, management nodes, and storage. Figure 3 the rack layout of a single SU. With DGX SuperPOD B300, we introduce the MGX-based, DC busbar powered design for the best datacenter density and efficiency.

In our reference design, for DGX B300s are within a single rack. The rack-level power consumption per rack exceeds 50 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.

For legacy datacenter without the possibility for DC busbar, it is still possible to build SuperPOD with traditional PDU and AC powered EIA racks.

Figure 3 shows 64 x NVIDIA DGX B300 Busbar systems in MGX racks each with DC Power shelfs and MGX rack stiffeners.

_images/image4.jpeg

Figure 3 DGX B300 in Racks#

Figure 4 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.

_images/image5.png

Figure 4 Management rack configuration#

This reference architecture is focused on 4 SU units with 256 DGX nodes. DGX SuperPOD can scale to much larger configurations up to and beyond 64 SU with 2000+ DGX B300 nodes. See Table 3 for more information.

Table 3 Larger DGX SuperPOD component counts#

SU Count

Node Count

GPU Count

Cable Count

Leaf

Spine

Node-Leaf

Leaf-Spine

1

64

512

16

8

512

512

2

128

1024

32

16

1024

1024

4

256

2048

64

32

2048

2048

8

512

4096

128

64

4096

4096

16

1024

8192

256

128

8192

8192

Contact NVIDIA for information regarding DGX SuperPOD solutions of four scalable units or more.