Interesting transfer in the technology industry. Yesterday Nvidia announced a new generation of graphics chips Ampere, designed for data centers within the system DGX. This is the first approach to the topic 7-nm production process, but without the participation of Intel.
Intel plans to upgrade to the “equivalent” of 7-nm process, only in 2021, and Nvidia needs a partner right now. Therefore, the company decided to go for a rather unusual collaboration, bringing to the aid of one of their biggest competitors, that is AMD.
Thus, Nvidia will provide the graphics chips for the system DGX, while processors will deliver exactly AMD – we are talking about a series of EPYC Rome. This is the first time the “green” decided to help “red” in their computers for the data center.
One system DGX A100 will include eight GPU from Nvidia-Tesla A100 Tensor Core. This should translate to quite a staggering 320 GB of available memory 40 GB on the chip. Processing power is also quite unimaginable: 5 petaflops. for artificial intelligence, and 10 for calculations.
One petaflops – a thousand teraflops. Thus, the recommended system will provide a capacity of nearly 417 Xbox Series X for problems with artificial intelligence and power 833 Microsoft devices if we need to do the math based on 8-bit numbers.
In the processor we are talking about two AMD chips EPYC 7742 Rome, based on the production process 7 nm from TSMC. One such processor has a… 64 cores and 128 threads with a base clock speed of 2.25 GHz and a clock frequency of 3.4 GHz.
System DGX is also a modest pool of 1 TB of DDR4 memory, and mass storage, of course, in the form of high-speed SSD capacity: 15 TB NVMe and two at of 1.92 TB. Now it’s time to announce the price. Attention, we recommend you to get: 199 thousand dollars. It’s not like the game equipment.