CXL™ Consortium member company GigaIO participated in a recent Q&A session to discuss GigaIO’s participation in the Consortium, CXL’s impact in the memory and storage industry and future use cases for CXL technology. Continue reading for the full Q&A session with GigaIO below.
Can you share a brief introduction of GigaIO?
GigaIO is democratizing access to Artificial Intelligence (AI) and High-Performance Computing (HPC) by transforming previously static datacenter infrastructure into infinitely reconfigurable architectures with our universal dynamic fabric, FabreX™. We are pioneering the use of PCI Express® (PCIe®) technology (and soon CXL™) as a routable network throughout the rack, breaking the server sheet metal barrier to place previously stranded server components like accelerators and storage into pools of resources that our software-defined platform can reconfigure and compose on the fly. Think of it as “software-defined hardware” which helps deliver next-generation agile infrastructure, where fast development and delivery mandate rapid and continuous integration. The main benefit is faster time-to-insight to increase innovation, thus driving higher revenue and/or enabling more research.
Committed to a fundamentally open approach, on the hardware side we work with any PCI end point device, any storage, persistent memory, any type of accelerator and any network adapter; on the software side, whether on Windows or Linux, we are building an ecosystem of orchestration software vendors based on open Redfish APIs and away from proprietary solutions that lead to vendor lock-in.
Why did GigaIO decide to join the CXL™ Consortium?
GigaIO was one of the earliest Contributor members of the CXL Consortium. As a pioneer and leader in the composable disaggregated infrastructure space, we are committed to full and complete disaggregation; and while there are certain things we can do today with memory using PCIe technology, getting the full benefits of disaggregated memory will require CXL. Today, we can have a server write directly to the memory of another server using DMA (Direct Memory Access) over PCIe interface – with I/O performance, and of course without memory coherency. CXL improves on both limitations, enabling us to build a memory pool for the entire rack that all servers in the rack will be able to share and access at PCIe performance.
What expertise does GigaIO bring to the consortium?
Our R&D team has decades of PCI and PCIe technology experience along with deep knowledge of high-performance interconnects from team members previously with Cray, Sun Microsystems, Intel, Broadcom, Emulex and QLogic. As 451 Research analyst John Abbott noted in his GigaIO Coverage Initiation in January 2021: “GigaIO’s technologists have a history of developing PCIe fabrics that goes back many years. They have seen what works and what doesn’t. This experience gives them a significant advantage this time around.”
What is the biggest advantage of CXL Consortium membership? How does GigaIO participate?
We believe it is important to support the work of the consortium through dues-paying membership because making this interconnect universal for memory will require the commitment of the entire computing and networking ecosystem. It is encouraging to note that this is the first time in history that all the major microprocessor vendors are working on the same caching standard. As the company pioneering the capability of using PCIe as a routable network, GigaIO is very excited about the growth path with CXL as a data plane, which is essential to our roadmap. We participate in the consortium by contributing to several technical workgroups to help shape the direction of the effort by sharing our technical expertise.
What use cases will be ideal targets for CXL technology? Which market segments will benefit from CXL?
Shared memory and dynamically pooled memory across all servers in a rack will drive much higher utilization of memory. This higher performance should be especially beneficial for large in-memory databases for example.
CXL will bring benefits across multiple segments, including enterprise computing, but will be especially valuable for data intensive areas such as AI, HPC, High Performance Data Analytics, and scientific computing.
What does GigaIO see as CXL’s impact within your industry?
CXL will revolutionize the datacenter architecture at the cluster-level interconnect level when combined with FabreX to treat the entire rack as the unit of compute, instead of the server, with completely disaggregated, yet coherent memory. We expect that CXL with FabreX will become the de facto rack-level interconnect, making it possible to disaggregate all rack resources including what so far has been the Holy Grail: memory. Check out our white paper “The Future of Composability with CXL” and our blog “The Holy Grail of memory pooling” to learn more about our perspective on CXL’s impact on the industry.
Keep up with CXL Blog to see future member spotlight posts from other CXL Consortium member companies. Are you a member interested in participating in the Member Spotlight series? Send a message to press@computeexpresslink.org for details.