18 July, 2024
 |
CXL product image Photo/ Samsung Electronics
Samsung is known for producing chips at scale, optimizing for cost-effectiveness and high volume. However, when it comes to pioneering new technologies, they have not seen significant success. The CXL technology, which was not originally developed by Samsung, is starting to be adopted by leading companies in Silicon Valley.
Creating a platform technology like CXL, which enables all devices within a server to share DRAM memory, requires significant effort. It involves convincing system software developers across various components, such as CPUs, GPUs, network devices, storage, and more, to integrate this technology. Achieving this level of collaboration is challenging even for companies like Intel, as the market does not support a monopolistic approach to such a comprehensive platform technology.
“In CXL 3.0, the multi-level connection of switches enables
near-infinite memory expansion in theory. The market will open up in earnest
when CPUs (central processing units) equipped with CXL 2.0 technology are
released in the second half of this year.” (Jangseok Choi, Senior Vice
President, New Business Planning Team, Memory Division, Samsung Electronics)
 |
Jang Seok Choi, head of the new business planning team
at Samsung Electronics' memory division Photo/ Samsung Electronics
CXL: The Future of Faster Computing
CXL, short for Compute Express Link, is a new technology that connects memory chips to CPUs (Central Processing Units) or GPUs (Graphics Processing Units) to make data processing much faster. Big companies like Samsung and SK Hynix are developing special CXL memory chips to get ready for this new market. Nvidia has a similar technology called NVLink, which could be replaced by CXL.
Why is CXL Important?
There are other emerging technologies like PIM,
neuromorphic chips, and silicon photonics. These are being considered as the
next big thing for AI chips, but CXL is expected to be available sooner because
many companies are already working on it. PIM and neuromorphic chips need more
applications to be useful, and silicon photonics is still in the research
phase.
What Makes CXL Special?
CXL 3.0 will allow almost unlimited memory expansion
by connecting multiple switches. This means that once CPUs with CXL 2.0 come
out later this year, the technology will really take off.
As AI becomes more popular, the semiconductor industry
is focusing on new technologies like CXL. This technology allows different
parts of a computer, like memory and processors, to work together more
efficiently, much like everyone speaking the same language.
How Does CXL Work?
Currently, server memory is tied to one CPU, which
isn't very efficient. Different parts of the system (memory, storage,
accelerators) all speak different "languages." CXL solves this by
letting everything communicate smoothly, eliminating the need for multiple
translations.
Future of CXL:
Samsung is leading the way by developing the first CXL
memory products. They expect the market to grow rapidly, especially by 2028.
CXL allows for easy memory expansion in data centers, reducing costs and
improving efficiency. Samsung's new memory pooling feature means multiple CPUs
can share memory, making it easier to manage and reducing wasted resources.
Conclusion:
CXL is set to revolutionize data processing in AI and other high-demand areas by making computer components work together more efficiently and allowing for much larger memory capacities, however, Samsung appears to lack the influence necessary to set industry standards in the semiconductor field. Additionally, substantial support from system software developers is essential to implement the CXL standard effectively, ensuring it meets all the memory requirements of server systems.