Our new NVMe RAID controllers from Graid Technology
New GPU-accelerated NVMe RAID controllers for modern storage requirements
Graid Technology Inc.'s SupremeRAID product line offers an architecture for accelerating and enhancing the reliability of RAID storage by utilizing GPUs instead of traditional CPU-based logic. The basic principle: RAID calculations are offloaded to the fast GPU.
This reduces CPU load, increases parallelism, and leverages the enormous computing power of modern graphics processors for parity and striping operations. This solution addresses bottlenecks in traditional RAID controllers, especially in NVMe SSD arrays with high IOPS and throughput requirements.
The platform supports common RAID levels (0, 1, 5, 6, 10), works with leading NVMe SSDs, and is compatible with Linux and Windows servers as well as Intel, AMD, and ARM platforms. It covers deployment scenarios ranging from edge environments to medium-sized enterprise workloads to compute-intensive data centers.
SupremeRAID Core
Efficient solution for edge and smaller data centers
SupremeRAID Core is the most cost-effective GPU-accelerated RAID solution in the portfolio and is designed for edge environments, branch offices, or smaller enterprise deployments. With support for up to 12 NVMe SSDs per card, Core offers performance figures such as ~10 million IOPS and 140 GB/s throughput.
The strengths of this model lie in its easy integration, low power consumption, and single-slot PCIe Gen4 solution, which enables high performance without battery support or external modules. Core is suitable for workloads that require performance and redundancy but do not need high-end data center-level throughput.
SupremeRAID Pro
Scalable enterprise controller
The SupremeRAID Pro model is designed for traditional enterprise data centers that require high scalability and robustness without directly targeting the maximum throughput of the Ultra variant. Pro supports up to 32 NVMe SSDs, delivers high peak values (around 22 million IOPS, ~280 GB/s throughput) and uses a mid-range GPU (e.g., NVIDIA RTX A1000 with 8 GB).
Nevertheless, the Pro variant efficiently covers typical enterprise workloads such as virtualization, database transactions, and large server file systems. Here, too, RAID processing is offloaded from the CPU, which reduces the application load and ensures better overall system response times.
SupremeRAID Ultra
High-end performance for demanding enterprise workloads
SupremeRAID Ultra is the top model in the series and is designed for maximum performance and scalability in data-intensive production environments. It enables NVMe RAID configurations with up to 32 directly connected NVMe SSDs per card and delivers extremely high performance metrics: up to 30 million IOPS and 280 GB/s throughput, making it ideal for AI/ML training, large databases, in-memory workloads, and real-time analytics. GPU-based acceleration frees up CPU resources and provides consistently high bandwidth, even during rebuild operations.
Technically, Ultra supports broad OS and platform compatibility as well as the same RAID levels as the other models, but is designed for maximum scalability and performance. The GPU used (e.g., NVIDIA RTX 2000 with 16 GB) is the performance core and distinguishes this model from the smaller variants.
Possible application scenarios for our Graid systems
AI/ML workloads – training large models, GPU clusters, real-time inference
With large AI models, storage quickly becomes a bottleneck, as AI workloads require extremely high I/O rates. Traditional RAID controllers slow things down because they tie up CPU resources. SupremeRAID shifts RAID calculations to the GPU and, according to the manufacturer, achieves over 95% of raw NVMe performance with full RAID protection. This lowers latencies (e.g., via GPUDirect Storage), reduces GPU idle time, and shortens training and inference times.
Enterprise databases & virtualization – Relational databases, private cloud (vSphere, Hyper-V), real-time analytics
OLTP/OLAP databases and VM environments require low latency and high parallelism. Traditional RAID controllers often limit performance due to CPU load. SupremeRAID can deliver significantly higher IOPS and throughput, thereby reducing the load on the CPU for applications.
HPC & Big Data – Parallel data transformations, HPC clusters, analytics engines
Big data analytics, HPC simulations, and scientific applications require high and consistent performance. GPU-based RAID architectures eliminate CPU bottlenecks and enable near-linear scaling with the number of NVMe SSDs – a limitation of traditional RAID systems.
Edge & Real-Time Data Acquisition – Industrial Edge Analytics, Mobile/Remote Systems, Real-Time Video Analysis
GPU RAID also increases efficiency in the edge environment: sensor, video, and streaming data can be stored with high performance and RAID protection, while CPUs remain free for application and network tasks.
Availability
You can order the SupremeRAID Core (RRP: €1,050) and Pro (RRP: €2,500) from our shop now. The SupremeRAID Core model will follow shortly. (Special conditions apply for registered specialist dealers and resellers.) You can download high-resolution images here.
