Artificial Intelligence GPUs Latest news

Rambus introduces its new GDDR7 memory IP to further advance AI 2.0

Rambus should not be an unknown company to some. Among other things, it is known for its production of HBM, PCIe and CXL controller IP. Now it is continuing its portfolio and introducing its new GDDR7 memory IP, which is intended to further advance AI 2.0. First of all: What is the goal of AI 2.0? The aim is to obtain an interference model that can create new multimodal content from a user’s input prompts.

Image Source: Rambus

As AI inference continues to proliferate and move from the data center to the edge and endpoints, the need for more powerful processing engines with tailored high-performance memory solutions is increasing across the computing landscape. Here, most GPUs utilized GDDR6 memory. The problem here, however, is that the practical limit of standard data center signaling can only be reached at data rates of up to 24 gigabits per second (Gbit/s). A new generation is therefore needed to further promote progress and meet demand.

Image Source: Rambus

Rambus has set itself the task of working on a new GDDR7 memory IP with good values. This supports 40 Gbit/s operation and a data throughput of 160 GB/s. Compared to GDDR6, this is an increase of 67%. The Rambus GDDR7 controller is intended to represent a new generation of GDDR memory deployments for state-of-the-art AI accelerators, graphics and high-performance computing (HPC) applications.

Image Source: Rambus

For Soo-Kyoum Kim, Vice President of Memory Semiconductors at IDC, this is a major performance improvement over the previous version and should become an important tool.

“GDDR7 memory offers significant performance gains over GDDR6. The Rambus GDDR7 Controller IP solution will be a vital tool for anyone that wants to take advantage of the improved speed and latency features offered by GDDR7.”

GDDR7 offers the following key features:

  • Supports all GDDR7 interconnect features including PAM3 and NRZ signaling
  • Supports a wide range of GDDR7 device sizes and speeds
  • Optimized for high efficiency and low latency in a variety of traffic scenarios
  • Flexible AXI interface support
  • Low-power support (self-update, self-update in sleep mode, dynamic frequency scaling, etc.)
  • RAS (Reliability, Availability and Serviceability) features – such as end-to-end data path parity, parity protection for stored registers, etc.
  • Comprehensive support for memory testing
  • Integration support for third-party PHYs available
  • Validated with the latest GDDR7 VIP and memory models from manufacturers

It remains to be seen how GDDR7 will develop in the future, but it is certainly a technical advance and we can expect a lot more in the future.

Source: Rambus

 

Bisher keine Kommentare

Kommentar

Lade neue Kommentare

Redaktion

Artikel-Butler

1,832 Kommentare 8,883 Likes

Rambus sollte für einige kein unbekanntes Unternehmen sein. Es wird unter anderem bekannt für seine Herstellung von HBM-, PCIe- und CXL Controller-IP. Nun führt man sein Portfolio fort und stellt seine neuen GDDR7 Memory IP vor, der die KI 2.0 weiter voranbringen soll. Erstmal: Was ist das Ziel von KI 2.0? Man will als Ergebnis ein Interferenzmodell erhalten, dass aus den Eingabeaufforderungen eines Benutzers neue multimodale Inhalte erstellen können soll. KI-Inferenzen verbreitet sich immer weiter und sich vom Rechenzentrum zum Rand und zu den Endpunkten verlagert, steigt der Bedarf an leistungsstärkeren Verarbeitungs-Engines mit maßgeschneiderten Hochleistungsspeicherlösungen in der gesamten Computerlandschaft. Hierbei […] (read full article...)

Antwort 2 Likes

Danke für die Spende



Du fandest, der Beitrag war interessant und möchtest uns unterstützen? Klasse!

Hier erfährst Du, wie: Hier spenden.

Hier kannst Du per PayPal spenden.

About the author

Hoang Minh Le

Werbung

Werbung