Raja Koduri Teases Ponte Vecchio Once more as GPU Rankings Design Win

Intel’s graphics guru Raja Koduri on Wednesday as soon as once more teased his Twitter fans with an animated symbol of ways the corporate’s Ponte Vecchio compute GPU can be made. He posted the teaser only a day after it used to be published that the Leibniz Supercomputing Centre will use Intel’s upcoming Sapphire Rapids CPUs in addition to Ponte Vecchio compute GPUs in line with the Xe-HPC structure beginning in 2022. 

See extra

Intel’s (*12*)Ponte Vecchio can be one of the business’s most complicated chips ever made  when it begins delivery in 2022. The compute GPU packs over 100 billion of transistors over 47 other tiles and guarantees to provide PetaFLOPS elegance AI efficiency. Construction a GPU of such complexity is an excessive problem, so it’s comprehensible that Koduri and different Intel engineers are extraordinarily proud. 

Hardwareluxx experiences that that the Leibniz Supercomputing Centre in Munich, Germany, will undertake Intel’s 4th Era Xeon Scalable ‘Sapphire Rapids’ CPUs and Ponte Vecchio compute GPUs for its segment two SuperMUC-NG supercomputer because of be delivered in 2022. The gadget could also be anticipated to make use of 1PB of Dispensed Asynchronous Object Garage (DAOS in line with Optane DC SSDs and Optane DC Continual Reminiscence.



(Symbol credit score: Intel)

There is not any phrase about efficiency anticipated from the brand new gadget, however the segment one SuperMUC-NG makes use of 12,960 Intel Xeon Platinum 8174 processors with 24 cores every. The gadget gives Linpack (Rmax) efficiency of 19,476.6 TFLOPS and Linpack Theoretical Height (Rpeak) efficiency of 26,873.9 TFLOPS. At the present, the supercomputer does no longer use any GPU accelerators for compute, however engineers from the Leibniz Supercomputing Centre reportedly use some Nvidia Tesla forums for checking out their methods.