- Kioxia GP Sequence SSD gives GPUs with quicker reminiscence entry past HBM limits
- Storage Class Reminiscence bridges the efficiency hole between DRAM and standard NAND flash storage
- XL-FLASH prioritizes low latency and hundreds of thousands of random IOPS over sequential velocity
Kioxia has launched a brand new kind of solid-state drive designed to perform as a direct reminiscence enlargement for GPUs.
The brand new Kioxia GP Series, introduced at Nvidia GTC 2026, is just not a alternative for present storage, however slightly an extra tier within the reminiscence hierarchy.
Its major function is to offer a bigger pool of fast-accessible knowledge for GPUs, successfully performing as an overflow for the costly and capacity-limited Excessive Bandwidth Reminiscence.
Article continues under
Reminiscence-hungry AI fashions drive the change
The drive leverages Storage Class Reminiscence (SCM), a class of expertise that sits within the efficiency hole between conventional NAND flash and system DRAM.
This idea was popularized years in the past by Intel’s now-discontinued Optane technology, which aimed to bridge the identical divide, however failed.
Kioxia’s model, branded as XL-FLASH, prioritizes low latency and excessive enter/output operations per second over uncooked sequential throughput, permitting finer-grained knowledge entry at simply 512bytes.
This improvement is a direct response to a basic downside in present AI infrastructure, GPU reminiscence is solely not large enough for the fashions it’s requested to run.
As AI fashions scale towards trillions of parameters and context home windows develop to hundreds of thousands of tokens, the demand for reminiscence to retailer gadgets just like the KV cache has outpaced the bodily limits of HBM.
Nvidia’s Storage-Subsequent initiative, which this SSD helps, was created to handle this precise bottleneck by encouraging storage distributors to construct drives that GPUs can hyperlink to straight.
“Kioxia totally helps the NVIDIA Storage-Subsequent initiative and can ship purpose-built SSDs to successfully deal with the necessity for GPU-accessible reminiscence,” stated Makoto Hamada, Senior Director of the SSD Division, Kioxia Company.
Whereas the GP Sequence goals for hundreds of thousands of IOPS to feed knowledge to GPUs, the broader business is chasing much more formidable efficiency targets.
The idea of reaching 100 million IOPS is a known industry pipedream, one which will require inventing totally new courses of reminiscence.
Different firms are additionally focusing on the area of interest market that Optane left behind.
For instance, the InnoGrit N3X SSD makes use of Kioxia’s XL-Flash in SLC mode to ship excessive endurance.
This drive can reportedly stand up to as much as 50 full drive writes per day for 5 consecutive years.
Kioxia itself has beforehand signaled its intent to push towards even larger efficiency figures.
The corporate is aiming for 10 million IOPS using SLC NAND, which means that the GP Sequence is only one step in an extended race to eradicate storage latency for processors.
It stays to be seen whether or not the Kioxia GP Sequence SSD will succeed the place earlier SCM makes an attempt stumbled.
Follow TechRadar on Google News and add us as a preferred source to get our knowledgeable information, critiques, and opinion in your feeds. Be sure that to click on the Comply with button!
And naturally it’s also possible to follow TechRadar on TikTok for information, critiques, unboxings in video type, and get common updates from us on WhatsApp too.
