When will NVIDIA release the next generation of 7nm ampere graphics cards? This is a matter of great concern to many players, and the March GTC conference is the most likely, but the first one is basically determined to be a data center-level ampere graphics card. The performance of the next generation NVIDIA GPU has been revealed in the next generation of super-computing Big Red 200 at Indiana University in the United States, which is said to improve performance by 70-75%.
According to data provided by Indiana University, the Big Red 200 Supercomputing initially planned a calculation of 5.9PFlops (5.9 trillion floating-point calculations per second), using the Tesla V100 acceleration card, but upgrading the next generation of NVIDIA GPU to 8PFlops. And fewer GPUs are required because the new architecture GPU (i.e. 7nm amp) delivers a 70-75% performance boost!
A 70% performance boost is already strong, but it may not be the true strength of 7nm ampere, and twitter user Dylan 522p’s calculation struck out even more surprising, as the 7nm amp GPU’s performance has improved by more than 100%.
According to his results, the FP64 of the 7nm amp GPU has up to 18TFLOPS, compared to the Tesla V100 acceleration card is 7-7.8TFLOPS and the latest Tesla V100s is 8.2TFLOPS.
In the Tesla series of acceleration cards, FP32:FP64 is 2:1, that is, 7nm ampere graphics FP32 performance up to 36TFLOPS, is currently my Tesla V100 series more than twice, the latter is the strongest is no more than 16TFLOPS.
From this information, NVIDIA’s next-generation GPU, 7nm Ampere, will experience a very large performance improvement, with a 100% increase of more than 50% from Pascal to Volta, which is only upgraded from 21TFLOPS to 30TFSLOP, It’s rare to multiply performance to GPUs after the 2000s.
Of course, how the game-level 7nm ampere ultimately is not a good idea, because NVIDIA in recent generations of GPUs Tesla and GeForce series are separate, RTX graphics with the 12nm Turing GPU full version of only 4608 CUDA core, Volta is 5120, And the game GPU FP64 performance castrated to 1:32, comparable only FP32 performance, TU102 core and GV100 core FP32 performance is similar, are around 14-16LOPTFS.
Overall, if the performance leakage on the Big Red 200 supercomputing is accurate, and NVIDIA does not have a big knife castration on the game GPU, then the next generation of 7nm ampere game card performance will also be significantly improved, along with the RTX tracking performance, Such performance advantages will obviously allow AMD’s Big Navi flagship card to catch up for a while.