Live DB — analyst tables
Test
Amazon 24576 Trainium1 EFAv2 Nitro-v4 100G NIC Hyperscaler Ethernet Cluster Configuration

AWS-Trainium1-Default

Amazon · Trainium1 · EFAv2 Nitro-v4 100G NIC

Rack type
Cluster / location
Default Inferentia2
Buyer profile
Hyperscaler
Networking type
Ethernet
Number of XPUs
24,576
NIC used
EFAv2 Nitro-v4 100G NIC
NICs / XPU
0.5
Scale-out BW / XPU
50G
XPU / server
16

Scale-up network

Assigned blockDefault
Trainium1
Scale-UpXPU: Trainium10 linesshared by 1 SKU

Scale-up Units are based on the scale-up world size (16 XPUs).

XPU World Size

2
WS
Scale-Up Network LayerPart / Model
Units#
Unit BWGbit/s uni-d
Lane Count#
Lane BWGbit/s uni-d
XPUs per Node·····
Scale-Up World Size·16···

Switches

1
SW
Scale-Up Network LayerPart / Model
Units#
Unit BWGbit/s uni-d
Lane Count#
Lane BWGbit/s uni-d
per ClusterNo Scale-Up Switch····

Connections & cabling

1
CN
Scale-Up Network LayerPart / Model
Units#
BW / PortGbit/s uni-d
Unit DP#
DP / Port#
Total DPs#
BW / ConnectorCable / Model
Cable Units#
·No Connectors on the Switch Side········

Scale-Up Parts Summary

0
ItemQty/ClusterQty/ServerAttach/XPU

Backend network

6 rows
Assigned blockDefault
Trainium1 - EFAv2 - 3L - ETH - TH2 - AWS
BackendXPU: Trainium16 linesshared by 1 SKU

Backend Units are based on the cluster size (12,288 Number of NICs).

Switches

6
SW
 Network Layer
Switch and Server SpecAggregate Port UtilizationNetwork Bandwidth Aggregate
SwitchPart Name / Model
Units#
Unit BWGbit/s
Port BWGbit/s
Downlink Ports / SW#
Uplink Ports / SW#
Max Ports#
Downlink Ports#
Uplink Ports#
Total Ports#
Downlink BWTbit/s
Uplink BWTbit/s
OversubRatio
Number of NICs·12,288100100012,28812,288012,28812,28801,228.8·
XPUs per Instance·16····8······
NICs per InstanceEFAv2 Nitro-v4 100G NIC8100100··8······
ToRWhiteBox Tomahawk 2 6.4T3846,40010032326412,28812,28824,5761,228.81,228.81
LeafWhiteBox Tomahawk 3 12.8T19212,800100646412812,28812,28824,5761,228.81,228.81
SpineWhiteBox Tomahawk 3 12.8T9612,800100128012812,288012,2881,228.80·

Transceivers

8
CN
 Network Layer
Transceiver / CablePart Name / Model
Total Units#
Units per SW/GPU#
Unit BWGbit/s
Lanes#
BW per LaneGbit/s
Agg. BWTbit/s
XPU NodesNone - AECs are Used ->12,288·1004251,228.8
CPU NodesNone - AECs are Used ->768·····
Leaf
DownlinkGeneric 100G SR4 QSFP2812,288641004251,228.8
UplinkGeneric 100G SR4 QSFP2812,288#ERR100#ERR250
Spine
DownlinkGeneric 100G SR4 QSFP2812,288#ERR100#ERR250
Uplink·······
Core
Downlink·······
Uplink·······
Super-Core
ToR
DownlinkNone - AECs are Used ->12,288321004251,228.8
UplinkGeneric 100G SR4 QSFP2812,288321004251,228.8

Cables

8
CB
 Network Layer
Cable/ FiberType
Total Units#
Units per Transceiver#
Unit BWGbit/s
Lane Count#
SerDes BWGbit/s
XPU NodesGeneric 100G AEC12,2881100425
CPU Nodes······
Leaf
DownlinkAlready counted above·····
UplinkGeneric 100G Multimode12,2881100425
Spine
Downlink······
Uplink······
Core
Downlink······
Uplink······
Super-Core
ToR
DownlinkAlready counted above·····
UplinkGeneric 100G Multimode12,2881100425

Backend Parts Summary

6
ItemQty/ClusterQty/ServerAttach/XPU
Generic 100G AEC12,2888.000.50
EFAv2 Nitro-v4 100G NIC12,2888.000.50
Generic 100G SR4 QSFP2849,15232.002.00
Generic 100G Multimode24,57616.001.00
WhiteBox Tomahawk 2 6.4T3840.250.02
WhiteBox Tomahawk 3 12.8T2880.190.01

Frontend network

No converged frontend + storage fabric data for this cluster.

Frontend Parts Summary

0
ItemQty/ClusterQty/ServerAttach/XPU