The Apollo 11 conspiracy, which claimed that the landing on the Moon by astronauts was a hoax, had been doing the rounds for a long time. In September 2014, a team from NVIDIA proved that the incident did happen and debunked the conspiracy, thanks to their new Maxwell graphics processing solution. A demo team rebuilt the entire scene, complete with the light bouncing off the astronaut Aldrin. They said it was their voxel global illumination (VXGI) technology that helped solve some of the most complex lighting and graphics challenges in visual computing.
What is Maxwell? What is so special about it that it could put down the views of many conspirators with just one image? Read on to know.
GPU computing is the use of a graphics processing unit (GPU) and a central processing unit (CPU) together, to perform computational tasks. The CPU allocates computational tasks to the the GPU, which considers any form of data to be images or other graphic forms and processes them at speeds hundreds of times faster than that of the traditional CPUs.
The Maxwell architecture
The Maxwell microarchitecture for implementing GPUs is an architectural innovation from NVIDIA. A pioneer in this space, NVIDIA has created a series of GPU architectures in the past, starting from Tesla, to Fermi and Kepler. The latest is Maxwell, using their 28nm process from the Kepler series.
NVIDIA released the Maxwell architecture in later models of graphics cards like GeForce 700, 800 series, and in the 900, Quadro Kxxx series. The first generation of Maxwell-based products were released in February 2014, followed by a second generation in September 2014. The latest release was earlier this year, in January 2015.
The idea behind Maxwell
The Maxwell architecture, for the first time, features an ARM CPU of its own. Enhanced graphics capabilities, simplified programming and improved energy-efficiency are the main pluses of this series. It incorporates a new age streaming multiprocessor (SMM), which makes the latest Maxwell architecture twice as efficient as the earlier Kepler architecture and saves almost 40 per cent of the power.
New datapath organisation and a new instruction scheduler are responsible for this. The design of the SMM is quadrant-based with four 32-core processing blocks, with a dedicated warp scheduler for each. This can dispatch two instructions per clock. Each SMM contains eight texture units, one polymorph engine (geometry processing for graphics), a dedicated register file and shared memory.