Intel has invested billions of dollars into its GPU compute project known as Larabee. Now the architecture is ready for market and it can barely compete with AMD’s VLIW architecture, which is the last generation of AMD graphics cards.
It is widely known that AMD’s VLIW architecture was quite handicapped when it came to GPU compute work and Nvidia’s 2090 accelerators, while being considerably better, were launched a year ago.
Intel’s performance per watt improvement measures just 0.7% when compared with a server using VLIW AMD cards and 9% when compared with a server using Nvidia’s Tesla 2090 cards.
So it appears even with all the money invested and all the advanced technology used, Intel has something which can barely outperform some last generation AMD graphics cards.
We are still surprised with the results given that Intel’s Xeon Phi coprocessors are manufactured using the 22nm process and have some of the best software optimisations and compiler support in the industry. Maybe Intel will be able to figure out what’s going wrong in time for their next generation because it seems likely that AMD and Nvidia’s latest GPU compete solutions will walk all over Xeon Phi.
