SK hynix has developed its first LPDDR6 DRAM, which offers a 33% speed increase and 20% better power efficiency compared to LPDDR5X. The chips, built on a 10nm-class process, have a base speed of over 10.7 Gbps and a 16Gb capacity per chip.
While targeting smartphones and tablets, the new memory is also expected to be significant for datacenters, particularly in AI servers using SOCAMM modules like those in Nvidia's upcoming platforms. SK hynix follows Samsung in announcing LPDDR6, with future speeds anticipated to far exceed the current 10.7 Gbps.
The main topics covered are the specifications and benefits of SK hynix's new LPDDR6 memory, its target markets (mobile and datacenter/AI), and the competitive landscape of LPDDR6 development.
SK hynix introduces turbocharged LPDDR6, 33% faster and 20% more power efficient than LPDDR5X — 16Gb chips deliver 10.7 Gbps, uses 10nm node
LPDDR6 will likely be a big boon for the datacenter market, once this memory architecture debuts on SOCAMM modules.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
You are now subscribed
Your newsletter sign-up was successful
SK Hynix has announced the successful development of its first LPDDR6 DRAM, touting 33% greater speed and 20% better power efficiency compared to previous generation LPDDR5X memory. The memory manufacturer also announced that it has developed its LPDDR6 memory on its leading-edge 10nm-class (1c) process node that it announced in 2024.
The new modules have a base operating speed of "over" 10.7 Gbps — which exceeds the fastest outgoing LPDDR5X memory modules on the market today — and a memory capacity of 16Gb per chip. To help with power consumption, SK Hynix implemented a new sub-channel structure and DVFS, or Dynamic Voltage and Frequency Scaling, which it says reduces power consumption by more than 20% compared to previous generation LPDDR memory products. The sub-channel structure accomplishes power reductions by powering only the data paths that are in use, while DVFS reduces clock speed and voltage when memory demands are light.
SK Hynix has announced its LPDDR6 eight months after JEDEC finalized and published the LPDDR6 standard last July. However, SK Hynix is not the first manufacturer to make LPDDR6; Samsung already announced its first LPDDR6 product and showcased it at CES 2026, with speeds up to 10.7Gbps.
Article continues belowSK Hynix has confirmed that its 1c LPDDR6 will be used in smartphones and tablets, but we can also expect LPDDR6 to be a huge boon in the datacenter market. LPDDRX has been very popular in AI servers that take advantage of SOCAMM/SOCAMM2 memory modules that only support LPDDR memory. For instance, Nvidia's GB300 Grace Blackwell Ultra Superchip uses SOCAMM, and Nvidia's latest Vera Rubin Superchip uses SOCAMM2 memory modules. Late last year, SK Hynix stated that it expects post-Vera Rubin Nvidia AI chip designs to take advantage of LPDDR6.
This is just the beginning for LPDDR6; speeds well beyond 10.7Gbps are expected to become the norm as memory makers get to grips with optimizing and improving on their LPDDR6 designs. The JEDEC group believes that LPDDR6 will have data rates of up to 14,400 MT/s, which is well beyond what the fastest DDR5 overclocking record holds.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards.
-
bit_user I'd love to read a deep dive on LPDDR6 by Anton! Seems like some interesting details are buried in there.Reply
I've noted LPDDR memories becoming a lot more relevant to mainstream & server computing as the years go by, and only expect this to continue. LPDDR5 went from hardly being on my radar to being a subject of intense interest, as Apple used it in their M-series SoCs, Nvidia is using it in Grace and Vera server CPUs, and AMD's Ryzen AI Max (Strix Halo) and Nvidia's GB10/N1X are using it in client machines.
Meridiano
