Bach­e­lor’s The­sis / Mas­ter’s The­sis / Stu­dent Re­search Pro­ject

Ab­stract

This stu­dent pro­ject’s goal is to per­form bench­marks on the Nvidia NVDLA Ma­chine Learn­ing Ac­cel­er­a­tor and use those mea­sure­ments to cre­ate a sta­tis­ti­cal per­for­mance es­ti­ma­tor using the Per­for­mance Rep­re­sen­ta­tives (PR) ap­proach. This model should then be com­pared to ex­ist­ing an­a­lyt­i­cal mod­els, like AMAIX, and other per­for­mance es­ti­ma­tion ap­proaches, like ACADL/AIDG.

Ref­er­ences

• A. L.-F. Jung, J. Stein­metz, J. Gietz, K. Lübeck, und O. Bring­mann, „It’s all about PR – Smart Bench­mark­ing AI Ac­cel­er­a­tors using Per­for­mance Rep­re­sen­ta­tives“. arXiv, 12. Juni 2024.
• Mika Markus Müller, Alexan­der Richard Man­fred Borst, Kon­stan­tin Lübeck, Alexan­der Louis-Fer­di­nand Jung, Felix Wedlich, Oliver Bring­mann - Using the Ab­stract Com­puter Ar­chi­tec­ture De­scrip­tion Lan­guage to Model AI Hard­ware Ac­cel­er­a­tors
• Kon­stan­tin Lübeck, Alexan­der Louis-Fer­di­nand Jung, Felix Wedlich, Oliver Bring­mann - Work-in-Progress: Ul­tra-fast yet Ac­cu­rate Per­for­mance Pre­dic­tion for Deep Neural Net­work Ac­cel­er­a­tors
• N. Zurstraßen, L. Jünger, T. Kogel, H. Ked­ing, und R. Le­u­pers, „AMAIX In-Depth: A Generic An­a­lyt­i­cal Model for Deep Learn­ing Ac­cel­er­a­tors“, Int J Par­al­lel Prog, Bd. 50, Nr. 2, S. 295–318, Apr. 2022.

Re­quire­ments

• Python
• Linux
• Suc­cess­fully at­teded the lec­ture “Grund­la­gen der Rech­ner­ar­chitek­tur” and/or “Par­al­lele Rech­ner­ar­chitek­turen” and/or “Mod­el­lierung und Analyse Einge­bet­teter Sys­teme” (op­tional)

Con­tact

Jung, Alexan­der

Lübeck, Kon­stan­tin

Bring­mann, Oliver