Event: SC26, the International Conference for High Performance Computing, Networking, Storage and Analysis
Dates: November 15-20, 2026
Venue: McCormick Place
City: Chicago, Illinois
SC is the yearly gathering of the supercomputing crowd. National labs, universities, cloud operators and the chip and system vendors who sell to them all show up. It used to be mostly about scientific computing. These days AI training clusters take up a big share of the floor, and the problems are the same ones: moving data fast enough and keeping expensive accelerators busy.
That’s the pitch Marvell (NASDAQ: MRVL) is bringing to Chicago. Its argument is simple. Big AI and HPC systems are stalling on connectivity and on memory, so a lot of compute sits idle. Marvell sells into both problems, switches and interconnects on one side, CXL memory expansion on the other. It also says it wants to scale memory and interconnect together, as one system.
What Marvell is showing
The booth is #4003. Demos are grouped around three ways of growing a cluster. Scale-up means optical links that let memory and compute stretch across several racks, so one scale-up domain gets bigger. Scale-out covers switching, pluggable optics and electrical interconnects. Scale-across means coherent pluggables for links between data centers, frontend and backend.
On the floor:
- Photonic Fabric memory module
- CXL memory, including the Structera add-in card, near-memory compute and expansion/compression
- Partner demos with Structera X expansion controllers running DDR4 and DDR5
- Bravera SSD controllers
- OCTEON DPU-based storage
- Multipath reliable connection (MRC) on Teralynx 10 switch silicon
- A data center simulator on Teralynx T100
- Alaska A AEC DSP-based active electrical cables at 1.6T and 800G
- A virtual rack with Teralynx T100 ODM switch trays and CXL gear
The CXL and Photonic Fabric pieces are worth a look if you follow memory pooling. Reusing DDR4 behind a CXL controller is one of the cheaper ways operators are stretching memory capacity right now, and the partner demos should show how far along that is.
Panel
Preet Virk, SVP and GM of Marvell’s Photonic Fabric business unit, sits on the Birds of a Feather panel “Photonics for AI and HPC Clusters: Breaking the Bandwidth and Power Barrier.” Wednesday, November 18, 12:15 to 1:15 p.m., Room W184bcd.
Why go
If you design or buy cluster infrastructure, SC is where the interconnect and memory roadmaps get argued over in person. Optics versus copper, CXL in practice, how much power the network eats. The exhibit floor is huge, so pick your booths before you get there.
Registration and the full program are on the SC26 website.