NVIDIA officially unveiled its Rubin platform, which will be the heart of the next generation of data centers. The platform, which is five times more powerful than its predecessor (Blackwell), is expected to revolutionize artificial intelligence processing.
The Rubin platform is made up of six different chips, all of which are currently being tested in NVIDIA labs. These chips include:
Rubin GPU with 336 billion transistors
Vera CPU with 227 billion transistors
NVLink 6 switch for communications
CX9 and BF4 for networking
Spectrum-X 102.4T CPO for silicon photonics These chips are used in NVIDIA’s DGX, HGX, and MGX systems. At the heart of each data center are the incredibly powerful “Vera Rubin Superchip,” which includes two Rubin GPUs, a Vera CPU, and massive amounts of memory in HBM4 and LPDDR5x configurations.
Key features of the Rubin platform:
6th generation NVLink with 3.6TB/s bandwidth
Vera CPU with Olympus dedicated cores
50 Petaflops NVFP4 Rubin processor
3rd generation Confidential Computing
2nd generation RAS engine with non-stop health check The Rubin GPU is specifically designed for heavy AI workloads. It delivers 50 Petaflops NVFP4 analysis and 35 Petaflops NVFP4 training, which is five times faster and 3.5 times faster than the Blackwell model, respectively. The HBM4 memory used in this chip also has a bandwidth of up to 22TB/s, which is 2.8 times faster than the previous model. NVIDIA has developed its own ARM architecture called Olympus for the Vera CPU. The chip has 88 cores and 176 threads, and its memory is combined with 1.8TB/s of converged memory and 1.5TB of system memory. These specifications allow the Rubin platform to deliver 2x better performance in data processing and compression.
NVLink 6 switches power the Rubin platform’s network. These switches have a bandwidth of 3.6TB/s per CPU and a total bandwidth of 28.8TB/s. The Rubin platform uses a liquid-cooled design that is designed for better performance and longer life.
With these features, NVIDIA Rubin becomes one of the most powerful artificial intelligence platforms that will have widespread applications in data centers and massive processing.
