cuisine-spirit.com
baitronAvatar de baitron172 billets | Profil Recherche Google

ce blog tous
Derniers billets Connexion
Archives

oemie

28/08/2018

AlphaICs offers RAP chip selection, which is expected to create strong AI

AlphaICs offers RAP chip options that can be deployed specifically in computing devices, self-driving vehicles and data centers.

The breakthrough of artificial intelligence (AI) in machine learning has driven the demand for AI chips. Currently, AI chips are dominated by graphics processors (GPUs), and FPGAs are available for selection. In the future, AlphaICs will provide another option. It is called the Real AI Processor (RAP) chip.

According to ZDNet, GPUs have been around for a long time, originally designed to quickly draw graphics, especially in the video game industry, and then gradually deepened into the field of machine learning. The GPU basically uses parallel computing. Although the central processing unit (CPU) can also do it, the GPU can continue to evolve more than the general CPU.

GPUs don't just rely on hardware, they also have high requirements for software. The key to NVIDIA's dominance of the AI ​​chip market lies in the CUDA and cuDNN libraries. Although AMD is trying to catch up with hardware, the investment in AI chips is not enough.

FPGA is not a new technology, it has appeared since the 1980s, the main dynamic restructuring. Now the main manufacturer is Intel (Intel), Intel previously acquired one of the world's top two FPGA manufacturers Altera, want to enter the AI ​​chip market.

Intel's latest research found that if low-precision data types are used, FPGAs can rival GPUs. It is obvious that FPGAs still have a bright future in the field of machine learning, but there is still no machine learning library supporting FPGAs so far. A long way to go. Overall, GPUs seem to have a more mature ecosystem, whereas FPGAs are highly resilient.

As for how to choose a GPU or FPGA vendor, you may want to choose from the cloud remote and local deployment. NVIDIA GPUs are supported by AWS, Azure, and Google Cloud cloud services. As far as FPGAs are concerned, only AWS and Azure support. Speaking of cloud or local deployment, there are not many opportunities to use the infrastructure. If there are enough, it is worth buying and installing. Otherwise, it is better to resort to cloud services only for occasional use.

In addition to FPGAs and GPUs, California company AlphaICs offers another option, called RAP chips, specifically deployed in computing devices, self-driving cars, and data centers. Linley Group analyst Linley Gwennap admits that although the GPU is not suitable for AI, the current technology development is only to this extent.

GPUs are good at classifying, but not good for decision making. AlphaICs' RAP chips aggregate operational tensors and can digest real-world feedback and react. Former Intel executive Vinod Dham said that since CPU has limitations, GPU is only suitable for the game industry, can not make decisions for the ever-changing environment, so that now is full of weak AI, it seems that only RAP chips can make strong AI, RAP chip is expected Listed in mid-2019.

looking for obsolete electronic components such 2SD2017 and XC17V02PC20I, oemsemi.com have more than 300 thousand parts you can choose.