AMD Instinct MI450X Compels NVIDIA to Modify Rubin AI Chip with Increased TGPs and Memory Bandwidth

The competitive landscape between NVIDIA and AMD in artificial intelligence (AI) architecture is heating up, as both companies are actively refining their next-generation designs to secure a competitive advantage.

The Intensifying Competition: NVIDIA’s Rubin vs. AMD’s MI450 Chips

Excitement surrounds the anticipated AI products from both NVIDIA and AMD, as significant upgrades are slated across various parameters including power efficiency, memory bandwidth, and manufacturing technologies. Recent analyses suggest that the competition between AMD’s Instinct MI450 AI series and NVIDIA’s Vera Rubin is poised to be more intense than in prior product cycles, leading to notable architectural adjustments.

Mi450X used to be 2300W TGP, to make it more competitive, AMD bumped it up to 2500W TGP. VR200 Rubin used to be 1800W TGP, and 2 months ago, it has been bumped up to 2300W TGP.2 months ago, in order for Rubin to maintain a lead over AMD’s MI450X, VR200 Rubin memory… https://t.co/FOGiXfQF7o

— SemiAnalysis (@SemiAnalysis_) September 26, 2025

According to insights from SemiAnalysis, AMD’s Forrest Norrod has expressed a strong sense of optimism regarding the MI450 lineup, likening it to a pivotal moment reminiscent of how the EPYC 7003 series transformed AMD’s server processor offerings. Norrod emphasized that they expect the MI450 to present a more formidable challenge to NVIDIA’s Vera Rubin, asserting that their upcoming technology stack will encourage widespread adoption among users.

NVIDIA innovation timeline featuring Blackwell 8S HBM3e, NVLink 5 Switch 1800 GB/s, and Feynman Next-Gen HBM in years 2025 to 2028. — Image Credits: NVIDIA

Reports indicate that both the MI450X and VR200 Rubin have undergone several modifications, particularly increasing their thermal design power (TDP) ratings and memory bandwidth to ensure competitive advantages. For example, the MI450X has seen its TDP increase by 200W, while NVIDIA’s Rubin has been adjusted upwards by 500W, now standing at 2300W. Moreover, Ruby’s memory bandwidth has reportedly surged to 20 TB/s, up from 13 TB/s, further illustrating the competitive market dynamics.

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, ~288GB per GPU
Memory Bandwidth (per GPU)	~19.6 TB/s	~20TB/s
Dense Compute (FP4)	~40 PFLOPS	~50 PFLOPS

As both companies prepare to launch their upcoming products, it is evident that the gap in technology between AMD and NVIDIA is diminishing. Both are poised to utilize similar advanced technologies, including HBM4 and TSMC’s N3P process node, as well as chiplet designs. Historically, AMD has struggled to maintain pace with NVIDIA’s rapid product cycles, but with the introduction of Vera Rubin, a more aggressive rivalry is anticipated.

While precise specifications for both lineups remain undisclosed, assurances from AMD officials, particularly Norrod, suggest that the MI450 will debut with highly competitive hardware. Concurrently, NVIDIA’s Vera Rubin has already begun to capture market interest, with companies like OpenAI already utilizing the latest advancements, signaling that the race for dominance in AI technology is firmly underway.

Source & Images