Slide, the new deep learning algorithm developed by intel

A few days ago, Intel Labs and Rice University announced SLIDE, an innovative deep learning algorithm that leverages today's hardware much more efficiently.

SLIDE, The new deep learning algorithm developed by Intel

With SLIDE, the CPU efficiency for deep learning training of traditional AI models is greatly improved. The research work exemplified that a set of platforms with 44 Xeon cores and a set of platforms worth USD 100, 000 backed by 8 NVIDIA Volta V100 acceleration cards performed the same training task in 3.5 times less time than with any other algorithm.

SLIDE does not need a GPU because it has a fundamentally different approach to deep learning. The standard "back propagation" training technique for deep neural networks requires matrix multiplication, an ideal workload for GPUs. With SLIDE, Shrivastava, Chen, and Medini turned neural network training into a search problem that could be solved with hash tables.

This radically reduces SLIDE's workload compared to backpropagation training. A high-end GPU platform like those offered by Amazon, Google and others for cloud-based deep learning services has eight Tesla V100s and costs around $ 100, 000, Shrivastava said.

Visit our guide on the best processors on the market

Interestingly, Intel also stated that its platform has not been fully optimized since DLBoost was not enabled. However, the exact model of the Xeon CPUs used to complete the 44 cores has not been announced. Speculating, they could be two Xeon Platinum 6238 of 22 cores and 44 threads, although they could also be unpublished models that have not been announced.

It seems that the future of computing goes through AI and deep learning, a segment well used by Intel and Nvidia. We will keep you informed.

Mydriverswccftech Font