We are just hours away from Jen-Hsun entering the stage at GPU Tech Conference. More and more rumors are pointing towards imminent Pascal architecture introduction.
Pascal gets 70% boost in Deep-Learning
NVIDIA has shown a new slide at GTC mentioning Pascal. This time Pascal is compared to Tesla K40 and M40 in cuDNN (CUDA Deep Neural Network), a library designed for deep-learning.
According to the slide, NVIDIA has improved the source code, which resulted in better performance in Kepler, Maxwell and Pascal-based computing hardware. The performance of Tesla K40 (GK110) has doubled, while the deep-learning speed of Maxwell M40 has increased 6.25 times. Meanwhile Pascal running cuDNN v5 sees 10.5x boost over first iteration of cuDNN. That’s roughly 70% over Maxwell GM200.
Supermicro confirms high-end workstation Pascal GPU solution this year
ComputerBase took part in Supermicro presentation at GTC. Although Pascal was not directly named, it was obvious that ‘2016 – Next Gen GPU Innovation’ stands for high-end computing hardware from NVIDIA. ComputerBase also confirmed that those solution will use NVLink, which as we know are big part of Pascal introduction.
Source: ComputerBase, Hardware.fr
Related Posts
NVIDIA to enable DirectX Raytracing (DXR) for Pascal and Volta
ZOTAC accidentally reveals GeForce GTX 1070 with GDDR5X memory
GeForce GTX 1060 with GDDR5X memory confirmed to use GP104 GPU
