Toms Hardware Jan2024 illustration related to Tom's Hardware: Entry-level GPU RAID card enables mind-bending storage speeds 80 GB/s of throughput f...
Article
January 27, 2024

Tom's Hardware: Entry-level GPU RAID card enables mind-bending storage speeds — 80 GB/s of throughput from eight SSDs with SupremeRAID SR-1001

GPU-accelerated RAID offers excellent performance benefits over traditional software and hardware RAID solutions and has become more consumer-friendly.

By Christopher Harper | Jan 27, 2024

On January 25th, 2024, Graid Technology officially announced an entry-level version of their SupremeRAID series of GPUs to build a RAID array out of the best SSDs. The new SupremeRAID SR-1001 is the entry-level version of the higher-end SupremeRAID SR-1010. SupremeRAID features a unique application of RAID controlled through GPU technology and the high-performance gains that come with it, allowing server operators and other high-end users to leverage the best of their multi-NVMe drive setups.

While RAID is a mainstay in the PC space for multi-drive users, RAID has historically been somewhat problematic for SSD users since they can't always maximize their speeds while using RAID. This applies especially when using software RAID bound to the CPU since reduced CPU performance usually doesn't allow for the throughput of a traditional hardware RAID controller.

Graid Technology's solution to the issues with software RAID and traditional hardware RAID controllers is to approach the problem with GPU hardware. GPUs make notoriously good workhorses when you can use them for non-graphical issues, and if Graid's benchmarking holds, they've managed to utilize that.

SupremeRAID Storage GPU Performance vs Hardware/Software RAID

Header Cell - Column 04K Random Read/Write (IOPS)1M Sequential Read/Write (GB/s)Throughput (GB/s)Maximum SSDs SupportedSupremeRAID SR-10016M / 600K80 / 30808SupremeRAID SR-100016M / 820K220 / 922032SupremeRAID SR-101028M / 2M260 / 10026032Hardware RAID3.9M / 108K13.5 / 413.58Software RAID2M / 200K9 / 2932

Compared to its higher-end cousins, the latest SupremeRAID-1001 card is targeted squarely at prosumers, performance enthusiasts, and home server users. With support for only eight SSDs but a maximum throughput of up to 80 GB/s, SupremeRAID-1001 should provide enough leeway for a few modern NVMe Gen 3, 4, or 5 SSDs.

Of course, proper server or data center operators should probably be looking toward the higher-end SupremeRAID solutions. Both support 32 SSDs, though if you're tapping out that capacity or using a lot of NVMe Gen 4/5 drives, you'll almost certainly want to opt for the highest-end SupremeRAID SR-1010 to minimize any possible performance loss.

For most users, though, it seems the SupremeRAID-1001 will be the best solution— provided the pricing is correct. For pricing inquiries, contact TD SYNNEX: levelup.tdsynnex.com/sponsors/graidtechnology

Learn More

News & Resources

Don't miss our joint speaker session with Supermicro, where we'll walk through scaling approaches across every tier, from single-server to rack-scale to monolithic deployments, and show how dense NVMe GPU platforms paired with SupremeRAID™ turn SSDs into a high-performance, protected KV cache tier. If storage is becoming your AI infrastructure bottleneck, this is the session to catch. See you at FMS 2026!
Graid Technology has achieved an industry-first benchmark: 100 million IOPS on a protected RAID 5 volume for GPU-initiated I/O, built from 32 KIOXIA XD8 NVMe SSDs. The result shows that resilient NVMe storage can now operate at the scale that current and future AI deployments require, which is especially significant for environments that require both extreme performance and enterprise-class data protection. In benchmark testing with a WholeGraph training workload, [...]
Every NVMe array pays a hidden storage tax — either 12–18% of line rate lost to a hardware RAID controller, or 18–28% of host CPU consumed by software RAID. With enterprise NVMe pricing up ~257% since Q2 2025, that tax now hits a much bigger check. SupremeRAID™ eliminates both halves by running RAID I/O on an NVIDIA GPU: full line-rate throughput, CPU cores returned to your applications, enterprise-grade protection on one card.