After making the best GPU for gaming this year, Nvidia set out to make a GPU that’s good at everything else, with the Pascal series of GPUs.
Much like the its Quaddro range of specialized GPUs, that are mainly used for workstations, the new Tesla range of GPUs implement the Pascal architecture that also has its own special application.
Nvidia unveiled the latest additions to its Pascal architecture with new Tesla P4 and P40 GPU accelerators
Nowadays, Artificial Intelligence is taking over everything – from Facebook chat boxes to self-learning software developed by Google and IBM. Self-driving cars for one, are required to be aware of their surroundings and know the objects that are ahead of them. They require sensors and graphical interfaces that show detail, depth, size, height and a bunch of other factors to know the obstacle ahead and beside of them. And that’s where the Nvidia Pascal architecture comes in.
The Pascal will store such information in the servers that are connected to the cars to allow them to learn and adapt to the situation when needed. That just means that these autonomous cars will be super smart. But that’s just one application where the Pascal range of GPUs will work.
They will also be applied in supercomputers and other high-tech systems for correlation and calculation of data that us mortals will never really get to work on or see for ourselves. Nvidia is also pro-cloud, enabling all their data to be accessed through the cloud whenever required.
The new Nvidia Tesla GPUs come in two variants – the P4 and the P40. The P40 can output 12 teraflops of single-precision performance with 3,840 CUDA cores. It has a whopping 24GB of GDDR5 (which is a big upgrade from the already fast GDDR4) memory and takes in about 250 watts of power. All this comes with the higher P40 variant.
The P4 variant has 2,560 cores, 5.5 teraflops of single-precision performance and 8GB of GDDR5 memory. It also draws out 75 watts of power. Both of these variants are equipped with deep-learning features. Each of those cores inside the Tesla can process a big chunk of the information it takes and then string all the information from each core to interpret the data as a whole, which is good for deep-learning.
But can it run Crysis 3? Technically, somewhat-yes. But that’s not what it’s meant for. It belongs in servers developed by Dell, HP Enterprise, Lenovo, Quanta, Wistron and other server manufacturers.
COMMENTS