The NVIDIA Turing architecture brings a new ray-tracing revolution in GPU history, and after two generations of product iterations, the next-generation Ampere architecture is also in the works, expected to be announced as early as march at the GTC graphics conference. Indiana University will create a new supercomputer, “Big Red 200”, on the 200th anniversary of its founding, with nVIDIA GPUs based on next-generation architectures deployed!
It is reported that this super-calculation will be built in two phases, the first phase has been on line, including 672 dual-way nodes, the processor is AMD’s latest flagship Dragon 7742, 64 core 128 threads, the entire system a total of 86016 core, 172032 threads.
Phase II will include NVIDIA’s next-generation GPU.
In fact, Indiana University originally planned to adopt NVIDIA’s existing high-performance computing GPU Tesla V100, Volta architecture, but they are now ready to wait and fall, it is clear that NVIDIA’s new architecture has a deadly appeal, and upgrading the GPU is the reason for the project into a two-phase construction.
According to data provided by Indiana University, the Big Red 200 Supercomputing initially planned a calculation performance of 5.9PFlops (5.9 trillion floating-point calculations per second), but upgraded the next generation of NVIDIA GPUs up to 8PFlops, with fewer GPUs required. Because the new architecture GPU delivers 70-75% performance improvement!
Of course, the upgrade is said here to compare the target of the Tesla V100, and certainly refers to the professional computing performance changes, but such a magnitude is still amazing, to the game card how to have about 50%, not to mention the previous rumors that the performance of the Ampere architecture will be 50% higher, energy efficiency is twice as good as Turing.
AMD will also launch the next generation of RDNA 2 GPU architecture this year, but don’t know if it will be able to fight Ampere?