Nvidia Vera Rubin GPU Performance: 5x Faster Than Blackwell Coming 2026
Nvidia reveals Vera Rubin GPU performance metrics at CES 2026. Compare 5x speed gains, 10x power efficiency, and Blackwell's 2.8x software-driven boost.
Buy Blackwell now or wait for Rubin? Nvidia just made that decision a lot more interesting. At CES 2026, CEO Jensen Huang unveiled the raw power of the upcoming Vera Rubin architecture, promising a massive leap in AI compute efficiency while simultaneously supercharging current hardware through software magic.
Nvidia Vera Rubin GPU Performance vs. Blackwell
According to Huang, the Vera Rubin GPU is a beast. It's capable of 50 PFLOPs of NVFP4 inference and 35 PFLOPs of training performance. That's 5x and 3.5x the performance of its predecessor, Blackwell. But the real story isn't just speed; it's the economics. Rubin aims to deliver 10x more throughput per watt and reduce the cost per token to 1/10th of current levels.
| Feature | Blackwell | Vera Rubin |
|---|---|---|
| Inference (NVFP4) | Base | 5x Faster |
| Training (NVFP4) | Base | 3.5x Faster |
| Efficiency | Standard | 10x Throughput/Watt |
| Availability | Now Shipping | 2H 2026 |
Blackwell’s Software-Driven Evolution
Enterprises don't have to wait until late 2026 to see gains. Nvidia reported that Blackwell performance has already improved by 2.8x for inference in just three months. This was achieved through optimizations in the TensorRT-LLM engine, specifically tested on the DeepSeek-R1 model. Training performance on the GB200 NVL72 also saw a 1.4x boost within five months without a single hardware change.
Authors
Related Articles
Nvidia shipped roughly a billion RISC-V cores in 2024, then announced it would run CUDA on the open standard. We break down how royalty-free instruction sets and open software stacks are trying to route around CUDA's lock-in. Part 2 of the Semiconductor Sovereignty series.
US AI-chip export controls split into three layers in the first half of 2026 — January easing, a May crackdown on circumvention, and a pending bill. Nvidia erased China from its guidance and still posted a record $81.6 billion quarter. A look at the export policy that both shields and cages it.
AMD's MI325X matches or beats Nvidia on memory and bandwidth — yet Nvidia's 86-92% share holds. The real moat is CUDA, 20 years in the making. Part 1 of 4.
Snowflake's new $6 billion AWS contract is about more than cloud spending. It signals a shift in AI infrastructure—away from Nvidia GPUs and toward cheaper, homegrown chips for the agent era.
Nvidia shipped roughly a billion RISC-V cores in 2024, then announced it would run CUDA on the open standard. We break down how royalty-free instruction sets and open software stacks are trying to route around CUDA's lock-in. Part 2 of the Semiconductor Sovereignty series.
US AI-chip export controls split into three layers in the first half of 2026 — January easing, a May crackdown on circumvention, and a pending bill. Nvidia erased China from its guidance and still posted a record $81.6 billion quarter. A look at the export policy that both shields and cages it.
AMD's MI325X matches or beats Nvidia on memory and bandwidth — yet Nvidia's 86-92% share holds. The real moat is CUDA, 20 years in the making. Part 1 of 4.
Snowflake's new $6 billion AWS contract is about more than cloud spending. It signals a shift in AI infrastructure—away from Nvidia GPUs and toward cheaper, homegrown chips for the agent era.
Thoughts
Share your thoughts on this article
Sign in to join the conversation