Memory is performance, and DDR5 promises much faster memory at a lower power draw. Credit: Thinkstock This year, server vendors will begin shifting to a new form of memory, Double Data Rate version 5, or DDR5 for short. With its improved performance, it will be very appealing in certain use cases, like virtualization and artificial intelligence. We’ll get to that in a minute. The DDR spec has been developed by the Joint Electronic Device Engineering Council since 2001, and with each iteration the spec supports faster speed and lower power draw. This holds true for DDR5. With DDR4 memory chips have a clock speed ranging from DDR4-1600 up to DDR4-3200, or 1600Mhz to 3200Mhz, while DDR5 starts at DDR5-3200 and will eventually scale up to DDR5-6400. DDR5 memory also features a minor power reduction. DDR5’s operating voltage is 1.1 volts, down from DDR4’s 1.2 volts. That may not seem like much, but multiply that over hundreds or even thousands of DIMMs in a data center, and you start to add up to some real power savings. Chip density has also greatly increased. DDR4 memory maxed out at 16Gb chips, but DDR5-based memory chips will eventually reach up to 64Gb. That translates to memory sticks, or DIMMs (dual in-line memory module, sometimes referred to as RAM sticks) with much larger capacity. DDR4 DIMMs top out at 512GB vs. 2TB for DDR5 DIMMs. DIMM modules plug directly into computer motherboards and contain several RAM chips. Something else that the enterprise should appreciate is that DDR5 memory has internal error correction, so it’s correcting errors before the data is sent to the CPU, something DDR4 did not have. Generations of DDR memory are tied to generations of CPUs. For example, DDR4 memory shipped in 2016 and was first used with Intel’s Haswell-EP line of processors and AMD’s Ryzen CPUs. DDR5 memory could start to appear in servers later this year when CPUs that support it—Intel Sapphire Rapids generation of Xeon processors or AMD’s Genoa generation of Epyc processors—are expected to ship. Note: You cannot use the current DDR4 DIMMs in the new servers featuring these processors, nor can you put DDR5 DIMMs in older Xeon or Epyc servers. For starters, the pins on DDR5 DIMMs are physically different from those on DDR4 DIMMs, so you can even plug it into the memory slot. Secondly, the new CPUs are designed with the expectation of DDR5 features and electronics. The best use cases Memory–intensive applications will get the most out of hardware that complies with DDR5, while applications that hit the CPU cache will get less benefit. CPU-bound applications include databases, business intelligence, and anything else processor-intensive. The more complicated the code, or the more data accesses you have, the more often the DDR bus will be accessed. When that happens, then there will be very noticeable improvement in performance over DDR4 memory, says Jim Handy, principle analyst with Objective Analysis, which specializes in following the memory market. The two big beneficiaries of DDR5 performance will be virtualized systems and artificial intelligence (AI), he says. “Virtualized systems end up having an awful lot of cache misses, just because everything is moving in and out of the cache. For AI workloads, not an awful lot of stuff stays in the cache for very long, so [DDR5] would be really important for that,” he said. CPU processing starts at the cache, and if the CPU doesn’t find the data it needs, it looks to main memory. With AI, you might cache something and then immediately get rid of it then cache something else then immediately get rid of it then cache something else, and so on. And every time you move something else to the cache, you have to go to DRAM to get the previous data. Handy doesn’t expect to see DDR5 used in the solid-state drive market anytime soon if ever, because SSDs are really slow by comparison. “Anything that communicates over the NVMe bus is just infinitely slow compared to the memory channel. They could use DRAM from a couple of generations ago and get away with that,” he said. DDR5 memory for desktops is trickling out, but the major server vendors have yet to release new servers pending Intel and AMD releasing their chips yet. The DDR5 memory for desktops costs a lot more than DDR4 at the same memory capacity. Like everything else it will come down with time. Related content news High-bandwidth memory nearly sold out until 2026 While it might be tempting to blame Nvidia for the shortage of HBM, it’s not alone in driving high-performance computing and demand for the memory HPC requires. By Andy Patrizio May 13, 2024 3 mins CPUs and Processors High-Performance Computing Data Center news CHIPS Act to fund $285 million for semiconductor digital twins Plans call for building an institute to develop digital twins for semiconductor manufacturing and share resources among chip developers. By Andy Patrizio May 10, 2024 3 mins CPUs and Processors Data Center news HPE launches storage system for HPC and AI clusters The HPE Cray Storage Systems C500 is tuned to avoid I/O bottlenecks and offers a lower entry price than Cray systems designed for top supercomputers. By Andy Patrizio May 07, 2024 3 mins Supercomputers Enterprise Storage Data Center news Lenovo ships all-AMD AI systems New systems are designed to support generative AI and on-prem Azure. By Andy Patrizio Apr 30, 2024 3 mins CPUs and Processors Data Center PODCASTS VIDEOS RESOURCES EVENTS NEWSLETTERS Newsletter Promo Module Test Description for newsletter promo module. Please enter a valid email address Subscribe