Nvidia is gaining upper hand as GPUs are becoming increasingly important in AI. However, it has surely occupied the driving seat after it successfully released Volta, the latest GPU venture that will shake up the world of AI altogether. In a grand announcement, Nvidia charted out the plan that will take them to the pinnacle of AI in front of 7,000 people. Without these GPUs, it is impossible to train neural networks to perform machine learning.
Not without competition
However, not everyone is rosy in the GPU front. Not only are the old competitors turning the innovation screw tighter, but also the news ones are joining the effort. In fact, AMD, the longstanding rival of Nvidia, has also released their new GPU Vega. The start-ups working on new GPU architecture have already been absorbed by Intel, who has also absorbed Altera, the renowned data accelerator designer.
All about Volta
Volta has around 5,000 CUDA cores and takes 815 sq. mm area. Powered with a 16 MB cache and 16 GB of HBM2 memory courtesy of Samsung, its speed can go up 900 GB per second. The processor, the Tesla V100 has a linking speed of 300 GB per second. There is an inbuilt proprietary NVLink designed by Nvidia. The new GPU gives a massive 50 percent boost in performance over the erstwhile Pascal cheap.
A shift in GPU making
With Volta, Nvidia is clearly setting machine learning to be their top priority. While it has not compared the performance with TPU ASIC, Google’s GPU venture, it surely is top-notch by all means. There can be a graphics-only version later this year since they are waiting to see how AMD fares with their highly anticipated Vega. Some of the biggest names like Amazon, Google, Facebook and Microsoft have come forth in taking Volta to be the preferred choice.