x
Thank you! Your submission has been received and we will follow up shortly.
Oops! Something went wrong while submitting the form.
Graid Technology Inc. logo in front of a shiny blue decorative background.
Article
March 14, 2023

Blocks & Files Features New SupremeRAID™ Software Release

Our GPU-powered RAID card has an upcoming software release designed to accelerate sequential writes to 90GBps or more in RAID-5 configurations. Read about it in Blocks and Files!

March 13, 2023 (Blocks & Files) -- GPU-powered RAID card supplier Graid Technology has a coming software release designed to accelerate sequential writes to 90GBps or more in RAID-5 configurations.

It ascribes this 4-9x speed boost for its SupremeRAID™️ product to intelligent parity data reduction technology. Graid claims this will enable users in AI/ML, video processing, and similar workloads to write data to NVMe drives at record speeds.

CEO Leander Yu said in a statement: “This new enhancement offered by Graid Technology will give users of SupremeRAID™️ a clear competitive advantage in the marketplace. The ability to write unstructured data sets to high-speed NVMe drives at such an accelerated rate will save our customers tremendous amounts of time and drive a lower TCO on their infrastructure purchases.”

RAID-5 aggregates a set of drives into a single logical unit and stripes blocks of data across them as well as using distributed parity to protect against a single drive failure. The group size is determined by the number of drives needed to provide a desired capacity level plus one spare drive. A feature of RAID-5 is that reads are fast while writes are slower due to parity calculations. Graid Technology appears to think it has cracked that nut.

Currently, there are two SupremeRAID™️ products. The SR-1000 uses PCIe gen 3, delivers 16 million IOPS, and 110GBps sequential read but only 11GBps sequential write throughput, with up to 32 SSDs. A 4-9x write speed boost would take that 11GBps number past 90GBps. The SR-1010 uses PCIe 4 and goes faster on the IOPS front, putting out 19 million IOPS, but has the same 110GBps sequential read throughput.

The new software release will be available for the SR-1000 and SR-1010 products in the summer. As well as improving performance by removing IO bottlenecks, it is designed to lower host CPU utilization too, the company claims. It will be available to all Graid Technology customers, regardless of channel, as a free upgrade upon general release.

Bootnote

The >90GBps sequential write number was based on a host running Linux RAID-5 with Intel Xeon Gold 6338 CPU, 32 cores at 2.0 GHz x 2, and with 20 Phison EPW5970 NVMe drives.

Read it on Blocks & Files

About Graid Technology

Chosen by CRN as one of the Ten Hottest Data Storage Startups of 2021 and a 2022 Emerging Vendor in the Storage & Disaster Recovery category, Graid Technology Inc. has developed the world’s first NVMe and NVMeoF RAID card to unlock the full potential of enterprise SSD performance. We’re headquartered in Silicon Valley, with an R&D center in Taiwan, and are led by a dedicated team of experts with decades of experience in the SDS, ASIC and storage industries. Graid Technology Inc. is redefining performance standards for enterprise data protection: a single SupremeRAID™ card delivers 19 million IOPS and 110GB/s of throughput. For more information on Graid Technology Inc., connect with us on Twitter or LinkedIn.

Additional Resource

Learn More

News & Resources

Visit Graid Technology at SC26 Booth 1326 in Chicago, November 17–19, to see GPU-accelerated I/O for AI inference and CPU-native data protection with VROC™ by Graid Technology. Stop by to enter our raffle for a pair of Meta glasses.
Storage Review built a shared storage host that gives eight NVIDIA DGX Spark systems access to one parity-protected Solidigm QLC pool, with SupremeRAID™ handling RAID 5 on a GPU. Here is what the testing showed and why it matters for AI labs and enterprise desktop AI fleets.
Graid Technology is featured in TechArena’s report The Integrated Machine. Field CTO Randy Kreiser discusses KV cache, inference performance, and why storage has to be designed into the rack from the start.