AMD representatives made an surprising reveal immediately on the corporate’s Meet the Consultants webinar: AMD is working to carry CXL reminiscence expertise to its shopper CPUs over the subsequent three to 5 years. This might carry storage class gadgets, like SSDs, onto the reminiscence bus to enhance efficiency. This expertise allows improved efficiency, decrease latency, and reminiscence growth capabilities by bringing distant reminiscence gadgets into the identical pool with system DRAM. Consider it as enabling you to broaden your reminiscence by plugging in an SSD or extra reminiscence onto a tool that slots into your PCIe or m.2 port.
Not like Optane, which Intel is killing off resulting from poor adoption, CXL already enjoys broad trade assist by means of an open protocol. In truth, AMD and Intel, amongst many others, are working collectively on the brand new specification.
The Meet the Consultants present lined a various vary of subjects, together with AMD’s AM5 platform, DDR5 reminiscence, and PCIe 5.0 SSDs. The host then opened the ground to questions. In response to the query of why storage gadgets aren’t related to the reminiscence bus, AMD’s Senior Developer Supervisor Leah Schoeb defined that persistent recollections (like SSDs) and reminiscence at the moment talk with totally different protocols, stopping communication.
“[…]It isn’t that sooner or later, we can’t be bridging that communication. That is one thing that we’re taking a look at with applied sciences similar to CXL. So you may discover over the subsequent, you recognize, three to 5 years, you may see it first within the server space, however you may discover shifting down into the shopper [consumer] space, ways in which we will be sure that reminiscence and storage can talk on the identical bus by means of CXL.”
The host requested Phison’s Senior Supervisor of Technical Advertising, Chris Ramseyer, if the corporate had any extra so as to add to the subject.
“Nicely, to be sincere, I am on calls about this. A few of these are with Leah. I am unsure how a lot I can actually give out. We have not introduced something on this space. However I can say that there’s progress being made. And, once more, this will likely be one other ecosystem-type venture, the place it is not simply going to be Phison and never simply AMD placing this collectively. We’re all going to should work collectively to do that, and these collaborations have actually superior PCs over the previous few years. And we proceed to stay up for doing that to advance the expertise and the efficiency, the power effectivity, and so forth,” he commented.
As a reminder, the CXL spec is an open trade commonplace that gives a cache coherent interconnect between CPUs and accelerators, like GPUs, sensible I/O gadgets, like DPUs, and varied flavors of DDR4/DDR5 and protracted recollections. The interconnect permits the CPU to work on the identical reminiscence areas because the related gadgets, thus bettering efficiency and energy effectivity whereas decreasing software program complexity and information motion.
Nevertheless, the protocol requires devoted silicon in each the host CPU and the related system, like reminiscence, persistent reminiscence, GPU, or accelerator, to operate. That requires the function to be baked into the chip, and as with all new expertise, CXL will take a while to mature.
The primary CXL-capable processors are proper across the nook, although: Intel’s Sapphire Rapids and AMD’s EPYC Genoa will include early revisions of the specification constructed across the PCIe 5.0 interface. New revisions of the CXL spec nonetheless beneath improvement will assist PCIe 6.0 and extra refined capabilities, like reminiscence sharing and pooling. AMD will reveal its EPYC Genoa server chips in a stay stream on November 10, whereas Sapphire Rapids is anticipated to reach early subsequent yr, so CXL expertise is on the cusp of real-world use.
AMD’s disclosure immediately does not give us a particular date or chip era for CXL assist in shopper CPUs, however the three-to-five-year window talked about suggests we might see it nicely after PCIe 6.0 gadgets, which debut within the 2024 timeframe.