Did Kioxia just unveil the fastest SSD ever? GP series uses Storage Class Memory to feed the HBM GPU with millions of IOPS
Kioxia is targeting high-performance AI workloads
· TechRadarNews By Efosa Udinmwen published 22 March 2026
Share this article 0 Join the conversation Follow us Add us as a preferred source on Google Newsletter
Get the TechRadar Newsletter
Sign up for breaking news, reviews, opinion, top tech deals, and more.
Contact me with news and offers from other Future brands Receive email from us on behalf of our trusted partners or sponsors
By submitting your information you agree to the Terms & Conditions and Privacy Policy and are aged 16 or over.
You are now subscribed
Your newsletter sign-up was successful
An account already exists for this email address, please log in. Subscribe to our newsletter
- Kioxia GP Series SSD provides GPUs with faster memory access beyond HBM limits
- Storage Class Memory bridges the performance gap between DRAM and conventional NAND flash storage
- XL-FLASH prioritizes low latency and millions of random IOPS over sequential speed
Kioxia has introduced a new type of solid-state drive designed to function as a direct memory expansion for GPUs.
The new Kioxia GP Series, announced at Nvidia GTC 2026, is not a replacement for existing storage, but rather an additional tier in the memory hierarchy.
Its primary role is to provide a larger pool of fast-accessible data for GPUs, effectively acting as an overflow for the expensive and capacity-limited High Bandwidth Memory.
Article continues below
Memory-hungry AI models drive the change
The drive leverages Storage Class Memory (SCM), a category of technology that sits in the performance gap between traditional NAND flash and system DRAM.
This concept was popularized years ago by Intel’s now-discontinued Optane technology, which aimed to bridge the same divide, but failed.
Kioxia’s version, branded as XL-FLASH, prioritizes low latency and high input/output operations per second over raw sequential throughput, allowing finer-grained data access at just 512bytes.
This development is a direct response to a fundamental problem in current AI infrastructure, GPU memory is simply not big enough for the models it is asked to run.
Are you a pro? Subscribe to our newsletter
Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!
Contact me with news and offers from other Future brandsReceive email from us on behalf of our trusted partners or sponsors