Intel insists that its Gaudi accelerators are a cheaper and more cost-effective alternative to competing solutions

All the big companies that they make GPUs are focused on offer maximum performance regarding tasks of Artificial intelligence. Intel’s latest bet has been Gaudi 2a card that according to themselves It is the only economical alternative to face NVIDIA solutions. Intel wanted to share the results of the try MLPerf Training v4.0 where it offers solutions more economical to carry out Generative AI projects.

This test included a new measure fine tuning of the Llama 2 70B parameter model using low range adapters. With 8 Gaudi 2 accelerators a training time of 78.1 minutes. The results for GPT-3 training in Tiber Developer Cloud have been 66.9 minutes in an AI system 1,024 Gaudi accelerators.

This kit has been announced by Intel in the past COMPUTEX, composed of 8 Intel Gaudi 2 along with a universal motherboard for a price of $65,000which is equivalent to approximately one third of the cost from other competing platforms. If we choose a kit with Gaudi 3this would rise to $125,000which still places it in 2/3 parts of the cost from other kits.

Intel has estimated that only 10% of projects with generative AI were carried out last yeardue to high costs What does it mean to launch these projects? Intel wants to change this situationoffering solutions scalable at a good pricewhich can even be run from the cloud.

End of Article. Tell us something in the Comments!