DESCRIPTION

In the new (AI) artificial intelligence era, Pascal GPU is definitely the computational engine providing a fantastic user experience through acceleration of the current deep learning applications. Tesla P40 has been specifically designed to provide maximum through put relative to blasting data volumes.

AI Inferencing enhanced by NVIDIA through Pascal-based Tesla P40:

Pascal based Tesla P40 GPU accelerators are the latest announcement by NVIDIA. With this new card, it is able to boost AI / Neural Network inferencing by 45x over CPUs and in relation to past generation GPUs to around 4x increase. Powerful software tools back the GPU accelerators and are able to significantly increase the general efficacy quite well as needed.

The fact that Modern AI services for instance email spam filters, movie and product recommendation engines and voice-activated assistance are promptly getting more complex, even mandating for up to 10x more compute as opposed to what neutral networks mandated a year ago necessitates the development of better and efficient GPUs which is what Tesla P40 GPU has brought in. The available CPU-based technology is not well equipped as required to deliver real-time responsiveness as necessitated by present AI services and thus the resulting poor user experience.

The NVIDIA Tesla P40:

The newest Tesla card (Tesla p40) by NVIDIA has developed a significant deep learning plat form. This platform is divided into two aspects mainly which include the following: Training GPUs and Inferencing GPUs. The new Tesla P40 accelerators fall under the category of interfacing GPUs.

Tesla P40 is purposely designed for inference purposes. To recognize images, speech or text as a response to user or device queries, the Tesla P40 applies deep neutral networks. The Pascal architecture on which it is based, the GPUs feature specified commands centered on 8-bit (INT8) operations ensuring a 45x faster response in comparison to CPUs and an advancement of up to 4x over GPU solutions. The Tesla P40 accelerator in comparison to the now outdated Tesla M40 comes equipped with TensorRT support and Deep Stream SDK as well.

Specifications of the Tesla P40 (“Pascal GP102”):

The Tesla P40 accelerator is extremely power and is based on the latest GP102 GPU. This card comprises 3840 CUDA cores and 24 GB of GDDR5 worth of memory. The clock speeds of the Tesla P40 are up held at 1303 MHz base and 1531 MHz as boosters. Its memory provides 346 GB/s band width along a 384-bit interface particularly it clocks 7.2 GHz effective. Similar to its predecessor the NVIDIA Tesla M40, it is characterised by passive form factor. The Tesla P40 packs 12 FP32 TFLOPs and 47 INT8 compute performance TFLOPs which come in a TDP package of 250W.

SPECIFICATION

Processing:

NVIDIA® CUDA® Cores: 3840
Peak Single Precision Floating Point Performance: 12 Tflops (GPU Boost Clocks)
integer Operations: 47 TOPS (GPU Boost Clocks)

 

Memory:

Capacity: 24 GB GDDR5  
Memory Interface: 384-bit
Memory Bandwidth: 346 GB/s

Technologies:

Bus Type: PCI-Express 3.0 x16
Thermal Solution: Passive

 

Power:

250 W
1 × 8-pin CPU power connectors