KTNF

Graid Technology’s SupremeRAID™ solution breaks the limits of traditional RAID by leveraging the immense computing power of modern GPUs. As the world’s first GPU-based NVMe RAID solution, SupremeRAID™ delivers exceptional performance with throughput up to 260GB/s and 28M IOPS, enabling ultra-fast data access for AI, HPC, and large-scale workloads. At the same time, it ensures uncompromised data protection through advanced support for RAID levels 0, 1, 5, 6, and 10, along with a patented distributed journaling for write operations to RAID systems that prevents data loss during power failure or system crashes, and a proactive bad block recovery function to extend SSD lifespan. The result: dramatically enhanced storage performance and reliability, with maximized ROI for high-performance infrastructure.

Unlike CPU-based software RAID and ASIC-based hardware RAID, which bottlenecks storage performance and reduces the return on investment (ROI) for NVMe SSD spend, SupremeRAID™ employs unique out-of-path technology, with data transfers occurring directly between the NVMe SSDs and the host memory. This direct connection allows data to be transferred at the full PCIe bandwidth of every NVMe SSD, delivering maximum storage performance, comprehensive data protection, unmatched configuration flexibility, attractive cost justification, and compelling value.


For example, using 16 NVMe SSDs on PCIe Gen5, each capable of 14 GB/s, the theoretical total bandwidth is 224 GB/s. Traditional RAID solutions, however, are limited by lane bottlenecks and typically deliver only about 25% of that performance, or roughly 56 GB/s. In contrast, SupremeRAID™ unleashes the full potential of PCIe Gen5, achieving 100% performance utilization and delivering the full 224 GB/s throughput.

SupremeRAID™, a GPU-accelerated RAID engine powered by patented innovations in data transmission and protection, offloads RAID computations to the GPU — freeing up CPU resources for core applications. This unique architecture boosts storage performance without increasing system complexity and effectively eliminates PCIe throughput bottlenecks, achieving throughput up to 260 GB/s read and 100 GB/s write of bandwidth.
To overcome IOPS limitations, SupremeRAID™ leverages the massive parallel processing power of modern GPUs to handle each I/O request directly. This design only requires minimal CPU involvement and delivers sustained performance of up to 28 million IOPS, making it ideal for high-throughput, latency-sensitive workloads.

1. Ensuring Data Integrity with Consistency Checks
The SupremeRAID™ is designed to provide high reliability and data integrity levels. A key feature that enables this is the consistency check function. The consistency check function allows administrators to ensure that the data stored on the SupremeRAID™ system is intact and uncorrupted. These checks can be performed on a regular schedule or manually initiated as needed. While running the consistency check, the system compares the data on each disk to identify any discrepancies or errors. Depending on the settings chosen by the administrator, the consistency check function can either automatically fix any errors that are found or stop the check and alert the administrator to any detected errors. This feature provides administrators with flexibility and control over how the system responds to errors.

2. Dual-Controller Architecture for Auto-Failover and High Availability (HA)
This feature enables the SupremeRAID™ system to automatically fail over to another SupremeRAID™ card when one SupremeRAID™ card experiences an issue without any interruption in service. This increased reliability and availability ensures that the system remains operational even in the event of a single card failure. SupremeRAID™ supports dual-controller configurations in two modes: dual-active and active-passive. This enhances our RAID solution with comprehensive protection and security. Additionally, the high availability (HA) functionality remains unaffected by the root complex. Whether within the same root complex or across different root complexes, we have implemented failover mechanisms to ensure high availability.


3. Distributed journaling for write operations to RAID systems
SupremeRAID™ incorporates a patent-awarded distributed journaling mechanism specifically designed to safeguard data during abnormal shutdowns in double-failure scenarios. This system ensures data integrity by logging data in a dedicated journaling space before writing it to the storage area, any incomplete I/O operations are replayed upon service restart to maintain data consistency. This journaling feature is automatically enabled in degraded mode to uphold data integrity. Additionally, users still have the flexibility to bypass journaling space reservation when creating a drive group.