At NVIDIA’s 2018 GPU Technology Conference keynote held in the Silicon Valley, the company CEO Jensen Huang announced world’s largest single chip graphics card – DGx2.
Alexnet is a a pioneering neural network, originally written with CUDA to run with GPU support, that won ImageNet Large Scale Visual Recognition Challenge 5 years. This has spawned thousands of AI networks. Jensen said,
“Alexnet five years ago took six days to train with 2 GTX 580s. That can now be done in 18 minutes on DGX-2.”
This new Graphics card could deliver 2 petaflops processing power with its 512 gigabyte RAM and 10 killowatts power input. Nvidia’s DGX 2 delivers 10 times more processing speed as compared to its predecessor – DGG 1 that was launched in September 2017.
DGX-2 weighs about 350 pounds or 158.7 Kg, this clearway means that no human can lift it. It replaces $3M of 300 dual-CPU servers consuming 180 kilowatts. Moreover it’s approximately 1/8th the cost, 1/60th of the space, 18th the power. This GPU has eight NICS from Melanox. It has two CPUs, 30 teraflops of storage on the system with 200 times moew bandwidth than the available highest speed NICs on the planet for processing data very fast. it’s 16 Volta equivalents connected by 12 new NVSwitches that creates the equivalent of a 512 gigabyte memory.Over 1,440 movies could be transferred over this switch per second for its 14 terabytes per second of aggregate bandwidth.
The fastest supercomputer in the world is 125 petaflops. And this can process data at a speed of 2 petaflops with 81,900 CUDA cores.
This gigantic GPU costs $399K, it can be used to build your personal supercomputer – Jensen added. But unfortunately this DGX-2 Graphics card can’t be used to play high-end games. Also Nvidia showcased its new Quadro GV100 which can be used at deep learning, CGI, data processing, etc.