Nvidia’s long-rumored and long-awaited Ampere architecture is here. Nvidia has announced A100 based on the latest Ampere architecture. But don’t get excited right now because this GPU is not for you. This GPU is made for the data center. A100 is the replacement for data center centric Nvidia V100.
A100 is based on 7nm architecture AKA Ampere architecture. For instance, all current-gen Nvidia GPU is based on 12nm architecture. Ampere architecture has allowed Nvidia to pack more transistors per square millimeter. Nvidia has squeezed 54 billion transistors in an 826 square millimeters die. For comparison Nvidia’s current gaming king RTX 2080Ti has 754 square millimeters die and it packs 18.6 billion transistors.
Apart from that A100 has 6,912 FP32 CUDA cores, 3,456 FP64 CUDA cores, and 422 Tensor cores. While the last-gen V100 has 5,120 CUDA cores and 640 Tensor cores. According to Nvidia A100 is 20x faster than V100 at AI inference and training.
Nvidia claims that a system that costs $11 million today and requires 25 racks of servers and 630 KW of power can be replaced by an Ampere system that fits in a single rack, costs $1 million, and takes 28 KW of power to operate.
A100 will have 48GB of Samsung made HBM2 memory which is capable of 1.5 terabyte transfer per sec. The A100 is being sold packaged in the DGX A100, a system with 8 A100s, a pair of 64-core AMD server chips, 1TB of RAM, and 15TB of NVME storage, for a wapping $200,000. So up until now, you have understood I said earlier that this GPU is not for you.
But what this GPU and architecture mean for consumer-grade GPU means Gaming GPUs. Well, it is clear that Ampere architecture will able to pack more transistors in a single GPU means better performance. But one thing to remember that we may see 7nm based other architecture for Geforce GPUs. As we have seen last-gen V100 is based on Volta architecture while Geforce GPU is based on Turing architecture. Although Nvidia has brought tenser core from Volta architecture to Turing architecture. We know that Ampere cards will support PCIe 4.0 so you can also expect that Geforce cards will have that support too. And Nvidia CEO Jensen Huang said already said that consumer-oriented Ampere GPUs will be biased towards graphics and less towards compute.