Nvidia announces tesla p40 and tesla p4 for artificial intelligence

Nvidia Tesla P40 and Tesla P4 offer tremendous advances in artificial intelligence

Many modern artificial intelligence (AI) services such as voice assistance, spam filters, and content recommendation services are experiencing tremendous growth in complexity, requiring 10 times greater computing power than just a year ago. In this situation, the current CPUs are not able to offer enough power, so the GPU is increasingly taking center stage.

We recommend that you read our virtual reality settings.

The new Nvidia Tesla P40 and Tesla P4 cards have been specifically designed to offer maximum performance in artificial intelligence scenarios such as voice, image or text recognition to respond as quickly as possible. These new cards are based on the Pascal architecture with 8-bit instructions (INT8) and are capable of offering 45x the performance of the most powerful CPUs and 4x the performance of the previous generation GPUs. The Tesla P4 has a starting consumption of only 50W to be 40 times more efficient in its work than a CPU, a server with only one of these cards can replace up to 13 CPU-based servers for video inference tasks, which means 8 times savings in total costs.

For its part, the Tesla P40 offers maximum performance in deep learning scenarios with its impressive 47 tera-operations per second (TOPS), a server with eight of these cards is capable of replacing up to 140 CPU-based servers, which translates saving over $ 650, 000 in server acquisition costs.

New TensorRT and NVIDIA DeepStream SDK software for maximum performance

Alongside the Tesla P40 and Tesla P4, the two new software NVIDIA TensorRT and NVIDIA DeepStream SDK have been released to accelerate artificial intelligence inference operations.

TensorRT is a library created to optimize deep learning models in order to offer an immediate response in the most complex network situations. Maximizes the efficiency and performance of deep learning with its 16-bit and 32-bit operations, plus 8-bit precision operations.

We recommend our post on the best graphics cards for gamers.

For its part, the NVIDIA DeepStream SDK offers the power of an entire server to simultaneously decode and analyze up to 93 streams of HD video in real time, a breakthrough compared to the 7 streams that a server with two CPUs can process. This represents a tremendous advance in the field of artificial intelligence by allowing video understanding operations for autonomous driving systems, interactive robots, content filters and deep learning, among others.

specs	Tesla P4	Tesla P40
Single Precision FLOPS *	5.5	12
INT8 TOPS * (Tera-Operations Per Second)	22	47
CUDA Cores	2, 560	3, 840
Memory	8GB	24GB
Memory bandwidth	192GB / s	346GB / s
Energy	50 Watt (or higher)	250 Watt