It seems that NVIDIA and AMD are racing to create a superior AI architecture, with both firms revising their next-gen designs to gain an edge.
The Race For High-End AI Architectures Will Be a Lot More Aggressive With NVIDIA's Rubin & AMD's MI450 Chips
There's a lot of optimism around the future AI products coming from both NVIDIA and AMD since massive upgrades are planned in several departments, including power consumption, memory bandwidth, process node utilization, and much more. However, based on a few reports and an X post by SemiAnalysis, it seems like the competition between AMD's Instinct MI450 AI lineup and NVIDIA's Vera Rubin is expected to be the highest relative to previous products, which is why, over time, there have been plenty of changes within the architectures.
SemiAnalysis quotes remarks from AMD's Forrest Norrod about how he expressed optimism about the MI450 lineup. He claimed that the Instinct MI450 AI lineup will be the firm's 'Milan moment', which refers to how the EPYC offerings changed with the debut of EPYC 7003 series server processors. More importantly, Norrod clearly declared that MI450 will be a much more competitive offering than NVIDIA's Vera Rubin and that there would be no hesitation in adopting AMD's tech stack over Team Green with the next-gen lineup.
Now, SemiAnalysis claims that both the MI450X and VR200 Rubin designs have been revised over time, seeing increments with TGP ratings and memory bandwidth. The adjustments were mainly targeted to debut with a superior product over the other. For instance, the MI450X saw a 200W TGP rise from initial figures, and in response, Rubin saw a bump of 500W as well, reaching up to 2300W. Similarly, Rubin's memory bandwidth reaches up to 20 TB/s per GPU, marking an increase from 13 TB/s per GPU, and SemiAnalysis claims that these revisions are associated with a "competitive market".
| Specifications (rumored) | AMD Instinct MI450 (MI400 family) | NVIDIA Vera Rubin VR200 (R200) |
|---|---|---|
| Launch window | 2026 (MI400 family launch). | 2H 2026 for NVL144 platform. |
| Memory type & capacity | HBM4, up to 432 GB per GPU. | HBM4, ~288 GB per GPU. |
| Memory bandwidth (per GPU) | ~19.6 TB/s. | ~20 TB/s. |
| Dense compute (FP4) | ~40 PFLOPS. | ~50 PFLOPS. |
There's no doubt that the technological gap between AMD and NVIDIA products will narrow with upcoming products since both firms are expected to employ the same technologies, whether HBM4, TSMC's N3P node, or a chiplet-based design. With past products, AMD had a considerable gap, mainly because it couldn't keep up with NVIDIA's product cycle, but with Vera Rubin, the competition will be a lot more aggressive.
For now, neither lineups have its exact specifications surfaced, but based on statements by AMD officials, especially Norrod, it is evident that the MI450 will debut with hardware which is the company is confident will see market adoption. Similarly, on NVIDIA's front, Vera Rubin has already started to see adoption, such as from OpenAI, which shows that the race is definitely on.
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
