PicOS N8560-32C — 32 x 100G Data Centre Switch
A PicOS data centre switch with 32 x 100G QSFP28 ports on Broadcom Trident 3. Breakout support for RoCE storage and management networks.
PicOS N8560-32C — 32 x 100G Data Centre Switch
A PicOS data centre switch with 32 x 100G QSFP28 ports on Broadcom Trident 3. Breakout support for RoCE storage and management networks.
- 32 x 100G/40G QSFP28 ports — up to 128 x 25G/10G with breakout
- 3.2 / 6.4 Tb/s switching capacity (unidirectional / bidirectional)
- Broadcom Trident 3-X7 silicon in a 1U chassis
- Open network operating system and automation with PicOS
The N8560-32C is a 1U data centre switch with 32 x 100G QSFP28 ports. In HPC deployments its typical place is not the compute spine but the two networks beside it: the storage plane and the management network.
Because breakout splits a single port into four 25G links, many lower-speed interfaces can be consolidated onto few switches. PicOS keeps hardware and network operating system separately selectable, which is decisive both for fitting your automation tooling and for long-term operating cost.
Specifications
| Specification | N8560-32C Trident 3-X7 |
|---|---|
| Switching | |
| Switch chip | Broadcom BCM56870 Trident 3-X7 |
| Port density | Native: 32 x 100G/40G — up to 128 x 25G/10G with breakout |
| Switching capacity | 3.2 / 6.4 Tb/s (unidirectional / bidirectional) |
| Physical | |
| Dimensions (H x W x D) | 44 x 442 x 560 mm (1U) |
Specifications follow the manufacturer datasheet. The exact configuration is confirmed in your quote.
Connectivity Solutions
A dedicated plane for storage
Separating parallel file system traffic from the compute fabric is the most practical cure for unpredictable latency in training jobs. A 100G storage plane is enough to saturate storage targets in most mid-sized clusters and leaves the InfiniBand spine to compute alone.
Management and IPMI network
A structure that scales to 128 x 25G/10G ports with breakout is used to consolidate node management interfaces onto a single switch. Keeping the management network physically separate from the compute network preserves access to nodes when the fabric has trouble.
Lossless RoCEv2 configuration
RoCE performance depends far more on congestion management than on raw speed. PFC priority classes, ECN thresholds and buffer settings are configured to your workload; a RoCE network left at defaults falls well short of expected performance.
Features
Open network operating system
With hardware and software decoupled, the switch is managed with the tooling your team already knows. Automation, version management and configuration templates are not tied to one vendor's closed ecosystem.
Flexible port planning through breakout
A single 100G port splits into four 25G links. That makes it possible to consolidate many lower-speed nodes onto few switch ports and reduce switch count per rack.
Airflow direction matters
Models in this family come in back-to-front and front-to-back airflow variants. An airflow direction that does not match your hot-aisle / cold-aisle layout makes the switch re-ingest its own exhaust; it is one of the items we verify before ordering.
Resources
Documents open on the manufacturer's or supplier's own site.
Request a quote for this product
Tell us the quantity and the cluster it will join; we send a compatible part list with lead times — usually the same business day.
- Authorised dealer supply, manufacturer warranty
- Compatibility check for switch, adapter and optics
- Lead time per line item
Related products
Let's Build Future Together.
