The fastest GDDR6X / GDDR6 memory available today is the Micron MT61K512M32KPA-24 version on the NVIDIA GeForce RTX 4080, rated at 24 Gbps, but actually 22.4 Gbps; the second fastest is the 21 Gbps memory on the NVIDIA RTX 4090 through RTX 4070 cards. From the looks of things, GDDR6 will soon be upgraded to 24 Gbps.

Considering that NVIDIA has not yet launched the RTX 4080 and RTX 4090 “Ti” version, AMD’s RX 7900 has already used 20 Gbps memory, so the back of NVIDIA and AMD are likely to launch 24 Gbps upgrade products.

The fastest memory graphics cards ranked as follows:

• Nvidia RTX 4080: 22.4 Gbps (GDDR6X)
• Nvidia RTX 4090/4080/4070Ti / 4070: 21 Gbps (GDDR6X)
• AMD RX 7900 series: 20 Gbps (GDDR6)
• NVIDIA RTX 4060 Ti / Radeon RX 7600: 18 Gbps (GDDR6)
• Intel Razor A770: 17.5 Gbps (GDDR6)

The post Rambus expects 24 Gbps GDDR6 memory to be in volume production by the end of the year appeared first on TechGoing.

The extended license agreement, effective July 1, 2024, will maintain similar financial terms and provide SK Hynix with broad access to the complete Rambus patent portfolio through mid-2034. Additional terms and details are confidential.

Rambus and SK Hynix are both committed to advancing the industry with world-class products and technologies, and this extension is a testament to the continued value of our intellectual property and our continued commitment to the development of new technologies and technologies,” said Luc Seraphin, Rambus president and CEO. continued value of our intellectual property and our ongoing product partnership.”

Rambus was founded in March 1990 as a U.S. technology licensing company specializing in the invention and design of high-speed chip interfaces. in 1990, Mike Farmwald and Mark Horowitz founded Rambus, Inc. and began filing patent applications.

Rambus derives most of its revenue from patents and is therefore often in litigation. rambus’ products include DIMM chips, interface IP, etc.

The post Rambus Extends Full Patent License to SK Hynix Memory appeared first on TechGoing.

(Source: Rambus Press Releases)

Scott Houghton, general manager of Rambus Interface IP, said:

"Rapid advances in artificial intelligence (AI)/machine learning (ML) and data-intensive workloads are driving continued evolution in data center architectures, while also placing higher demands on performance levels.

The Rambus PCIe 6.0 interface subsystem meets the performance requirements of next-generation data centers with best-in-class latency, power, area and security."

In terms of specifications, the Rambus PCIe 6.0 interface subsystem provides data rates of up to 64 GT/s, complemented by full optimization to meet the needs of advanced heterogeneous computing architectures.

Within the subsystem, the PCIe controller has an Integrity and Data Encryption (IDE) engine dedicated to protecting PCIe links and the valuable data passing through them.

On the PHY side, it also provides full support for CXL 3.0, enabling caching, coherent memory sharing, and chip-level solutions for scaling and pooling.

Additional PCI Express Features:

"● Compatible with PCIe 6.0 (64 GT/s), 5.0 (32 GT/s), 4.0 (16 GT/s), 3.1/3.0 (8 GT/s) and PIPE 6.x (8 / 16 / 32 / 64 / 128-bit)

● PIPE 10b/20b/40b/80b bit width supporting SerDes architecture

● Support original PIPE 8b/16b/32b/64b/128b bit width

● Compliant with PCI-SIG's Single-Root I/O Virtualization (SR-IOV) specification

● Supports multiple virtual channels (VCs) in FLIT/non-FLIT mode

● Support endpoint (Endpoint), root port (Root-Port), dual-mode (Dual-mode), switch port (Switch Port) configuration

● Backwards support from PCIe 6.0 to PCIe 1.0 rates

● Supports Forward Error Correction (FEC) -- a lightweight algorithm with low latency

● Support L0p low power mode

● Up to 4-bit data path parity protection

● Supports clock gating and power gating

● RAS features -- including LTSSM timer overrides, ACK/NAK/Replay/UpdateFC timer overrides, unscrambled PIPE interface access, error injection on Rx and Tx paths, restoration of detailed state, etc., allowing for safe in-mission Deploy IP SoCs reliably."

Shane Rau, research vice president for computing semiconductors at IDC, said:

"PCI Express is ubiquitous in the data center, and CXL is growing in importance as companies pursue ever-escalating speeds and bandwidths to meet the higher performance levels of next-generation applications.

Additionally, as more chip companies emerge to support new data center architectures, access to IP solutions for high-performance interfaces will also be key to enabling this ecosystem."

Finally, the Rambus PCIe 6.0 interface subsystem has the following key features:

"● Supports PCIe 6.0 specification, including 64 GT/s data rate and PAM4 signaling.

● Improve link robustness by deploying low-latency forward error correction (FEC)

● Supports fixed FLIT size for high bandwidth efficiency

● Backward compatible with PCIe 5.0, 4.0 and 3.0/3.1"

The post Rambus Announces PCIe 6.0 Interface Subsystem to Enable Data Center and AI SoC solutions appeared first on TechGoing.