inference/research
verifiedhigh · 0.95updated 2026-10-04

AMD Infinity Fabric scale-up link (CDNA4: 153.6 GB/s per direction, 20% faster)

153.6 GB/s

Fields

id
interconnect/amd-infinity-fabric-link-cdna4
Kind
xgmi
Version
CDNA4 (MI350X/MI355X)
Bandwidth (GB/s, per link, unidirectional)
153.6
Bandwidth basis
Per the CDNA4 whitepaper: 'The Infinity Fabric links in the AMD CDNA 4 architecture run up to 20% faster generationally at 38.4Gbps for a total link bandwidth of 76.8GB/s IN EACH DIRECTION' and endnote MI350-007: 'up to 153.6 GB/s peak total aggregate theoretical transport data GPU peer-to-peer (P2P) bandwidth per AMD Infinity Fabric link'. The 153.6 = 2 x 76.8 identity confirms: 76.8 GB/s per direction, 153.6 GB/s bidirectional per link. Total aggregate P2P: 1075.2 GB/s (7 x 153.6). Total peak aggregate theoretical transport rate: 1203.2 GB/s.
Link count
7
Topology
Fully connected 8-GPU mesh (each GPU connects to 7 others). No switch. 7 links per GPU for GPU-to-GPU communication. Identical topology to CDNA3. The MI355X product page lists 'Infinity Fabric Links: 7' and 'Peak Infinity Fabric Link Bandwidth: 153 GB/s' (rounded). Scale-out link: 128 GB/s (unchanged from CDNA3).
Scale-up
yes
Scale-out
yes
Switching
None - direct point-to-point links between GPUs
notes
CDNA4 increases Infinity Fabric link speed by 20%: from 128 GB/s per direction (CDNA3) to 153.6 GB/s per direction (CDNA4). The links are 16 bits wide and fully bidirectional, running at 38.4 Gbps. Per-link bidirectional bandwidth: 153.6 GB/s (76.8 GB/s each direction). Total aggregate P2P: 1075.2 GB/s (up from 896 GB/s on CDNA3). Total peak aggregate theoretical transport rate: 1203.2 GB/s (up from 1024 GB/s). The 7-link full mesh topology is unchanged. The scale-out link remains at 128 GB/s (unchanged). This 20% improvement directly translates to 20% higher AllReduce bandwidth for tensor parallelism. The CDNA4 whitepaper states the links are 'fully bi-directional' and the per-direction figure is 76.8 GB/s.
way
each

Notes

CDNA4 increases Infinity Fabric link speed by 20%: from 128 GB/s per direction (CDNA3) to 153.6 GB/s per direction (CDNA4). The links are 16 bits wide and fully bidirectional, running at 38.4 Gbps. Per-link bidirectional bandwidth: 153.6 GB/s (76.8 GB/s each direction). Total aggregate P2P: 1075.2 GB/s (up from 896 GB/s on CDNA3). Total peak aggregate theoretical transport rate: 1203.2 GB/s (up from 1024 GB/s). The 7-link full mesh topology is unchanged. The scale-out link remains at 128 GB/s (unchanged). This 20% improvement directly translates to 20% higher AllReduce bandwidth for tensor parallelism. The CDNA4 whitepaper states the links are 'fully bi-directional' and the per-direction figure is 76.8 GB/s.

Cross-references

This record references no other records — it is a leaf. It is cited by 1.