Nvidia is hoping to attract machine learning developers with the Jetson TX1, an ARM-based development board powered by the top-end Tegra X1 SoC. The company claims that in certain deep learning tasks that rely on dynamic input and computations—autonomous drones, facial recognition and behavioural analysis, and computer vision—the Jetson TX1 will beat out an Intel Core i7 6700K Skylake CPU in performance.
The TX1 is almost identical to the SoC used in the Shield Android TV, which features an 8-core 64-bit ARM CPU—made up of four A57 cores with 2MB L2 cache, and four A53 cores with 512KB L2 cache—and a Maxwell-based 256-core GPU that's claimed to provide up to 1 teraflop of processing power. The key difference is the RAM, which has been bumped up to 4GB of LPDD4 with 25.6GB/s of bandwidth.
Unlike its predecessor the TK1, the TX1 is split into two parts: a credit card-sized module that contains the SoC itself (with a 400-pin interface), and a separate carrier board for I/O. The idea is that developers who wish to focus purely on software and/or peripheral development can buy a complete kit, while those wanting to use in the TX1 in more integrated solutions can just buy the compute module.
Walang komento:
Mag-post ng isang Komento