The market believed Rubin CPX was dead @mingchikuo says it is back for 1Q27, and the spec is the story The 2025 design was Nvidia's own HBM dodge: 128 GB of GDDR7 to make prefill cheap The revived chip carries 168 GB of HBM4 at Rubin's full 2,300 W, with NVLink cut to 1.5 TB/s at most Nvidia kept the memory and cut the fabric Kuo: over 50% of AI inference workload is prefill and KV cache build At the recommended 1:1 pairing, every Rubin now adds another 168 GB of HBM4 Even prefill pays the tax

- Lightmatter officially introduced Passage L20 CPX on September 17 and joined the Open CPX MSA, describing the product as the first bidirectional Open CPX optical engine for AI infrastructure. - Passage L20 CPX uses 1311 nm and 1331 nm wavelengths to carry transmit and receive traffic over the same single-mode fiber, cutting fiber and connector requirements by roughly 50% versus conventional UniDi architectures. - The module is designed for 6.4 Tbps Tx plus 6.4 Tbps Rx, or 12.8 Tbps aggregate bandwidth, using 32 optical lanes per direction at a 212.5 Gbps PAM4 line rate with more than 500 meters of SMF reach. - In Lightmatter's 512-GPU scale-up pod model, the architecture reduces fiber count from roughly 131,072 to 65,536 while eliminating approximately 16,000 connectors and more than 200 miles of fiber. - Lightmatter estimates that BiDi could reduce total scale-up interconnect network spending by roughly 15%, although this figure comes from internal modeling and will vary by topology and deployment configuration. - Importantly, this is not yet a volume-production product. Passage L20 CPX evaluation kits are expected to ship to customers in Q1 2027, and the disclosed specifications remain preliminary. - The announcement follows Lightmatter's June entry into NVIDIA's NVLink Fusion ecosystem and its August launch of an OCP CPO architecture effort targeting AI systems scaling from 72 to more than 1,024 nodes.

Samtech displayed their Co-Packaged Copper (CPC) set-up in the CPX form factor at Taiwan OCP (See the open CPX MSA here: Bandwidth exits the 6.4T pluggable optical engine via a copper connector interface. There are 128 connector pins per module corresponding to 64 differential pairs (DPs) or 32 lanes of 200G. We expect the CPX form factor to present a compelling opportunity for many copper interconnect companies such as TE, Amphenol and Molex even as we see bigger optics TAM in scale-up networking.

Just as the market had come to believe that Rubin CPX had been dropped from Nvidia’s product roadmap, my latest industry checks indicate that Nvidia has revived the program, with production expected to begin in 1Q27. Compared with the previous design, the revived Rubin CPX delivers stronger prefill performance and features major changes to both its GPU specifications and rack architecture, underscoring the high priority Nvidia places on prefill solutions. Key changes: 1. Rubin CPX GPU specs CPX delivers near-Rubin compute performance and matches Rubin’s maximum power rating of 2,300 W per GPU. CPX moves to 168 GB of HBM4, vs. 288 GB on Rubin and 128 GB of GDDR7 on the previous CPX design. 2. Rack design The new CPX uses a standalone MGX ETL rack rather than sharing a rack with Rubin, as in the previous design. Customers can opt for 64, 128, 192, or 256 CPX GPUs depending on their needs. Within a CPX rack, each group of 64 CPX GPUs forms a rack module comprising eight compute trays (eight CPX GPUs per tray) and one switch tray. 3. Scale-up and scale-out NVLink is used only for scale-up among the eight CPX GPUs within each tray, with 1–1.5 TB/s of NVLink bandwidth per CPX (vs. 3.6 TB/s per Rubin). Inter-tray scale-out within each rack module runs over Spectrum-6 Ethernet using all-copper L1 links. Across rack modules, scale-out is handled by each module’s Spectrum-6 switch over OSFP optical links. 4. How it works CPX must be paired with Vera Rubin NVL72, and Nvidia recommends a 1:1 ratio of CPX to Rubin GPUs. CPX handles prefill and builds the KV cache, which is then transferred to Rubin over Ethernet RDMA for decode. 5. Product positioning: Best performance per dollar for long-context prefill Over 50% of today’s AI inference workload comes from processing input context and building the corresponding KV cache. CPX therefore offers a more flexible, lower-cost way to handle prefill. Each eight-CPX tray has approximately 1.34 TB of HBM4, sufficient for most long-context prefill workloads and associated KV cache requirements.