Accelerating Datacentre performance through Memory Chips to efficiently manage the Big Data Age

Projektinformationen

ENERGIA/PIM

ID Finanzhilfevereinbarung: 190141232

DOI

10.3030/190141232

Projekt abgeschlossen

EK-Unterschriftsdatum 22 April 2022

Startdatum 1 März 2022

Enddatum 29 Februar 2024

Finanziert unter

• The European Innovation Council (EIC)

Gesamtkosten

€ 3 566 042,50

EU-Beitrag

€ 2 496 229,75

Koordiniert durch

SAS UPMEM
France

Periodic Reporting for period 2 - ENERGIA/PIM (Accelerating Datacentre performance through Memory Chips to efficiently manage the Big Data Age)

Berichtszeitraum: 2022-12-01 bis 2024-02-29

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

UPMEM has developed a ground-breaking semiconductor-based intelligent memory product for servers, which provides by far the best performing compute solution for big data and Artificial Intelligence applications. Through creating Processing-In-Memory (PIM) technology, UPMEM bypasses structural bottlenecks inherent within existing server architecture, which remain the root cause of Central Processing Unit (CPU) slowdown and low usage rates. By performing calculations in memory where the data resides, PIM eliminates the need to continually move data between the CPU and memory, a highly inefficient process that absorbs 80% of the energy consumption within respective computing nodes. The power of UPMEM’s technology comes from leveraging existing industry protocols and programming languages and designing PIM modules that fit into standardized server memory slots. When compared to those using conventional memory, PIM equipped servers have proven to be up to 20x faster, 12x more efficient and 10x cheaper.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

UPMEM team had conducted the design works of its new Processing In Memory chip and architecture to provide the main building block of its ultra efficient semiconductor based accelerator for big data and AI.
The team has designed: new enhanced chip interface, new low power processor pipeline, new repair and protection features, new scheme to boot as standard memory, new security scheme to make PIM elements secure enclaves...
The team has designed the associated control software for boot, run and secure modes based on the architecture. As well as a first enhanced version of the module to support those new chips.
The team also worked with prime server partners (Intel, QCT, Supermicro) to prepare the server integration of the Processing In Memory solution on the latest server platforms.
UPMEM team has continued to develop new use cases with international commercial and research labs with its first generation scheme to nurture the specification of the new Processing In Memory chip, architecture and software.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

Not applicable.