Energy Efficiency Boost in the AI-Infused POWER10 Processor

We present the novel micro-architectural features, supported by an innovative and novel pre-silicon methodology in the design of POWER10. The resulting projected energy efficiency boost over POWER9 is 2.6x at core level (for SPECint) and up to 3x at socket level. In addition, a new feature supportin...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Proceedings - International Symposium on Computer Architecture s. 29 - 42
Hlavní autoři: Thompto, Brian W., Nguyen, Dung Q., Moreira, Jose E., Bertran, Ramon, Jacobson, Hans, Eickemeyer, Richard J., Rao, Rahul M., Goulet, Michael, Byers, Marcy, Gonzalez, Christopher J., Swaminathan, Karthik, Dhanwada, Nagu R., Muller, Silvia M., Wagner, Andreas, Sadasivam, Satish Kumar, Montoye, Robert K., Starke, William J., Zoellin, Christian G., Floyd, Michael S., Stuecheli, Jeffrey, Chandramoorthy, Nandhini, Wellman, John-David, Buyuktosunoglu, Alper, Pflanz, Matthias, Sinharoy, Balaram, Bose, Pradip
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.06.2021
Témata:
AI acceleration
Computational modeling
Computer architecture
Design methodology
Energy efficiency
microprocessor design methodology
Power demand
POWER10
pre-silicon modeling
Sockets
Technological innovation
ISSN:2575-713X
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:We present the novel micro-architectural features, supported by an innovative and novel pre-silicon methodology in the design of POWER10. The resulting projected energy efficiency boost over POWER9 is 2.6x at core level (for SPECint) and up to 3x at socket level. In addition, a new feature supporting inline AI acceleration was added to the POWER ISA and incorporated into the POWER10 processor core design. The resulting boost in SIMD/AI socket performance is projected to be up to 10x for FP32 and 21x for INT8 models of ResNet-50 and BERT-Large. In this paper, we describe the novel methodology deployed and used not only to obtain these efficiency boosts for traditional workloads, but also to infuse AI/ML/HPC capability directly into the POWER10 core.
ISSN:2575-713X
DOI:10.1109/ISCA52012.2021.00012