MetaX MetaXLink (multi-GPU interconnect)

Fields

id

interconnect/metaxlink

Kind

other

Version

MetaXLink, MXC500 / MXN100 generation (MACA 2.0)

Bandwidth (GB/s, per link, unidirectional)

896

Bandwidth basis

PER 8-GPU DOMAIN, AGGREGATE, DIRECTION NOT STATED. MetaX's own product page for the C550 (primary vendor source, cn-metax-c550) states: 'MetaXLink interconnect enables an 8-card full-interconnect topology with bandwidth up to 896 GB/s.' This is the FIRST-PARTY published figure for MetaXLink. No direction convention is stated - the 896 GB/s could be per-direction or aggregate. The figure is recorded exactly as published.

Link count

8

Topology

8-card full-interconnect topology. MetaX's C550 product page states 'MetaXLink interconnect enables an 8-card full-interconnect topology.' The C550 3D Mesh Supernode supports up to 8 servers and 64 cards with 3D Mesh topology. No switch part is named.

Switching

NOT PUBLISHED. No MetaX switch part is named in any source this repo holds.

notes

UPDATED 2026-10-03 with the first-party published MetaXLink bandwidth figure. MetaX's C550 product page (cn-metax-c550) states 'MetaXLink interconnect enables an 8-card full-interconnect topology with bandwidth up to 896 GB/s.' This closes the previous documented negative where no bandwidth was available. The 896 GB/s figure is for the 8-GPU full-interconnect domain. No direction convention is stated by the vendor. MetaX's product taxonomy: MXN = inference, MXC = training/HPC, MXG = graphics. The C550 is an MXC (training) part. MetaX does not publish a separate interconnect specification for the MXN (inference) series - the same MetaXLink technology is used across the product line. The MXMACA software stack is compatible with mainstream GPU ecosystems. MetaX IPO'd on the SSE on 17 Dec 2025 (ticker 688802).

way

unknown

Notes

UPDATED 2026-10-03 with the first-party published MetaXLink bandwidth figure. MetaX's C550 product page (cn-metax-c550) states 'MetaXLink interconnect enables an 8-card full-interconnect topology with bandwidth up to 896 GB/s.' This closes the previous documented negative where no bandwidth was available. The 896 GB/s figure is for the 8-GPU full-interconnect domain. No direction convention is stated by the vendor. MetaX's product taxonomy: MXN = inference, MXC = training/HPC, MXG = graphics. The C550 is an MXC (training) part. MetaX does not publish a separate interconnect specification for the MXN (inference) series - the same MetaXLink technology is used across the product line. The MXMACA software stack is compatible with mainstream GPU ecosystems. MetaX IPO'd on the SSE on 17 Dec 2025 (ticker 688802).

Cross-references

This record references no other records — it is a leaf. It is cited by 3.