New technologies based on the next-generation CXL interface are transforming the way data centers manage memory, storage, and more.
With increasing numbers of high-power computing applications such as AI, machine learning, cloud computing, and video transcoding, the data centers need to process big data more than ever.
To help address the unprecedented crunch that data servers and IT systems now need to process, Samsung has adopted the CXL (Compute Express Link) standard. Products like Samsung’s CXL Memory Expander provide data centers with the ability to hyperscale while improving performance, cost, and efficiency.
How? Read on to learn more about the most innovative new technologies in memory, and how Samsung is pioneering it.
What is CXL?
CXL is the advanced next-generation interface and Samsung's most innovative new memory technology. It is an interconnect based on an open, industry-backed standard that allows different types of network devices to work more efficiently when processing high-performance computations. How does it do that? CXL lets network devices bypass the bottleneck of a host processor and communicate with each other directly and quickly.
Imagine trying to juggle a dozen balls while also calculating taxes in your head. You'd probably drop the balls, right? Or at least make a calculation mistake. But now imagine that your eyes can see your hands directly without being processed by your brain first. In such a scenario, you might easily be able to juggle balls and do tax calculation at the same time.
In a nutshell, that's what CXL does for data servers. By leveraging the unique strengths of the CXL standard, Samsung has improved the performance and efficiency of hyperscale applications. CXL has reduced the software stack and overhead and enabled greater capacity and memory bandwidth.
CXL Memory Expander
Last year, Samsung unveiled the first 512GB CXL DRAM module called the CXL Memory Expander. This module uses CXL technology to enable extremely high memory capacity in existing IT systems.
In standard computer architecture, a central processing unit (CPU) or host processor can access only DRAM slotted into a dual in-line memory module (DIMM). In other words, the upper limits of how much memory a CPU can access are defined by the maximum capacity of DIMM.
For massive data centers like the ones running ChatGPT, that upper limit is much higher. Hyperscale applications require an extraordinary amount of memory. But there’s still a limit, and once servers reach it, they traditionally have only one option to build a new system with a higher upper-memory limit.
What makes CXL Memory Expander so innovative is that it enables extremely high memory capacity in existing IT systems. Instead of being bottlenecked by the memory capacity in DIMM, CXL Memory Expander allows networks to expand to potentially tens of terabytes of memory, all while greatly reducing latency.
CXL Memory Expander also helps businesses to save their server operating costs and improve memory efficiency. The secret sauce is memory pooling, which enables host processors to sip only the amount of memory they need from the memory pool. But that’s not all. Advanced security features limit which hosts can have access to memory pooling, and hot plug functionality enables admins to swap out CXL DRAM on “hot” servers without shutting them down first.
The Future of CXL
CXL Memory Expander is just the first product in Samsung Semiconductor’s roadmap of CXL innovations, which also includes the Scalable Memory Development Kit (SMDK). With these products, data centers will be able to take advantage of unprecedented versatility and cost efficiency in terms of how they choose to manage and expand their memory.
But the best is yet to come. Over the next few years, we expect that the server requirements of AI, ML, VMs, and other HPC applications will continue to explode. We can’t wait to show you the CXL memory innovations we have planned to help your business manage the challenges of the hyperscale era.