[News] GTC 2025: NVIDIA Unveils More on Rubin GPUs, Announces Feynman for 2028 in Roadmap Update

According to a TechNews report citing Tom’s Hardware, NVIDIA revealed updates to its data center roadmap for 2026 and 2027 at GTC 2025. The announcement outlined the planned configurations for the upcoming Rubin and Rubin Ultra. Here are some key details we currently know about them.

First, there’s a slight name change. As the report indicates, NVIDIA plans to adjust the naming for its upcoming Rubin product line. CEO Jensen Huang revealed that the Blackwell B200 GPU actually consists of two dies, which alters the NVLink topology. So, even though the company currently refers to the solution as Blackwell B200 NVL72, Huang believes that calling it NV144L would be more appropriate.

The report highlights that Rubin will have an HBM upgrade, transitioning from HBM3/HBM3e to HBM4, while Rubin Ultra will adopt HBM4e. The memory capacity will remain at 288GB per GPU, the same as the B300, but bandwidth will increase from 8 TB/s to 13 TB/s. Moreover, NVLink speed will double.

The other half of the Rubin family is the Vera CPU, which will succeed the current Grace CPUs. As noted in the report, Vera will be a relatively compact processor featuring 88 custom ARM cores and 176 threads. It will also include a 1.8 TB/s NVLink core-to-core interface for seamless integration with Rubin GPUs.

Regarding the Rubin Ultra, it is anticipated to arrive in the latter half of 2027, bringing a substantial boost in GPU capabilities. According to the report, the full rack configuration will be upgraded to a new layout, NVL576, allowing for up to 576 GPUs per rack.

TrendForce’s latest findings on the AI server supply chain have revealed that NVIDIA is expected to launch the GB300 chip ahead of schedule in 2Q25.

TrendForce highlights recent supply chain developments, noting that GB300-related suppliers are set to initiate design planning in Q2 2025. The GB300 chip and Compute Tray are expected to enter production by May, with ODM manufacturers designing early engineering samples. By 3Q25, as rack system configurations, power specifications, and SOCAMM designs are finalized and move into mass production, GB300 full-rack systems are expected to gradually scale up shipments.

NVIDIA Unveils Next-Gen Feynman GPU in Updated Roadmap

In addition, TechNews highlights that NVIDIA’s undated roadmap reveals the successor to the Rubin series will be named after theoretical physicist Richard Feynman and is scheduled for launch in 2028.

According to a report from Wccftech, while NVIDIA has not yet disclosed detailed specifications of the Feynman GPU, the company confirmed that it will utilize next-generation HBM memory, potentially indicating HBM4e or HBM5. Furthermore, NVIDIA’s Feynman GPUs will be paired with the same Vera CPUs featured in the Rubin and Rubin Ultra platforms.

As for a potential Feynman Ultra variant, there is no official confirmation at this time, but the Wccftech report suggests that an upgraded chip could emerge closer to 2030.

Read more

• [News] NVIDIA Teams up with TSMC for Silicon Photonics Switches, Expected in 2025-26
• NVIDIA GB300 To Feature Enhanced Specifications, Full Rack Shipments Expected to Gradually Scale in 3Q25, Says TrendForce
• [News] SK hynix Reportedly Locks in Exclusive 12H HBM3E Deal for NVIDIA’s GB300

(Photo credit: NVIDIA)