Google announced that the Google GCP Cloud Platform introduces the new NVIDIA Tesla P4 GPU computing power, optimized for graphics-intensive applications and machine learning applications.
Prior to this, the Google Cloud Platform already had top-level computing cards such as Tesla K80 (Kepler Architecture), Tesla P100 (Pascal Architecture), and Tesla V100 (Volt Architecture) to meet high-performance computing and machine learning needs. The Tesla P4 claims to offer a better price/performance ratio.
Based on Pascal architecture, Tesla P4 has 2560 stream processors (GTX 1080 level), peak single-precision computing performance of 5.5TFlops (5.5 trillion floating points per second), INT8 integer computing performance of 22TFlops, and latency is 40 times shorter than traditional GPUs. And there is an 8GB GDDR5 memory.
The Google Cloud Platform now supports the NVIDIA GRID virtualization platform based on the Tesla P4/P100 computing card. Users can remotely use one or more workstation-level GPUs to perform 3D modeling rendering, computer-aided design, visual search, and voice interaction. Video editing and transcoding work.
Google Tesla P4 supports 1/2/4 GPUs per system, can be virtualized to 1-96, and system memory can be matched with up to 624GB.
In terms of price, Tesla P4 has an exclusive cost of $0.60 per hour, $432 for a month of continuous use, or about $3,000; non-exclusive expenses are $0.21 per hour, only $150 per month, but may be used by others at any time. interrupt.
The exclusive costs of the Tesla K80, P100, and V100 are 0.45, 1.46, and 2.48 dollars per hour, respectively.