Become a TechRadar Insider
- Longsys WM8500 offers as much as a 2:1 in-Memory lossless compression
- The 5nm chip currently supports up to 128TB in single-drive capacity versus mainstream consumer SSDs that are capped at 8TB
- This is made possible by using the DRAM-less SPU as an active player in the storage stack, leveraging both its High Level Cache (HLC) and intelligent Storage Agent (iSA) to deliver a compression ratio that is industry-leading
Longsys, the world's second-largest independent memory firm and the force behind one of the most well-known consumer brands in the West, Lexar, as well as one of the most important B2B storage players, FORESEE, may have a solution to rising SSD and DRAM costs: A chip that compresses data on the fly extraordinarily well.
It has come up with a 5nm chip which handles on-the-fly compression for large SSDs, allowing them to essentially double their capacity beyond the 128GB single-drive capacity it currently supports.
While this isn't as close to what hardware-based data compression on tape drives looks like (with ratios of up to 2.5:1), it is still an impressive feat for an industry reeling from ever-increasing NAND flash costs, even as many datacenters continue to use hard drives to keep costs low.
How does the Longsys WM8500 essentially double storage?
The Longsys WM8500 is what the storage giant calls an SPU, or Storage Processing Unit, built on a 5nm process and fundamentally different from technologies such as Samsung's SmartSSD.
Unlike Samsung's approach, which leverages a general-purpose FPGA or an ARM-based CPU inside the SSD to manage computational tasks on the drive, the SPU is an ASIC (Application-Specific Integrated Circuit) designed for a specific purpose: compression and storage management.
The 5nm chip also offers a cost advantage that most high-end consumer and enterprise-grade controllers do not have, apart from its compression capabilities: it is a completely DRAM-less design that allows it to command a lower price, even as Longsys's claim of offering 'virtually' twice the storage to its AI consumers kicks in.
It must be noted that the 2:1 figure is an 'up-to' ratio, and while it might be easier for an ASIC to compress data and context windows for certain AI models, others might make it considerably harder, especially if obfuscation or encryption is in play.
In an ideal scenario, however, the WM8500 SPU, coupled with its High Level Cache (HLC) implementation and its Intelligence Storage Agent (iSA), together make up what the manufacturer calls a "closed-loop software-hardware collaborative technology system" that focuses on AI customers.
The HLC cache claims a 40% reduction in DRAM requirements, making it a cost-effective alternative to HDD storage, even as competitors prepare to release 256TB enterprise SSDs later this year. AI data centers continue to demand large amounts of storage and memory alike to meet their ever-increasing needs.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
