GP-SPARK
High-performance AI silicon platform • Integrated security • Enterprise distribution service
- AI Acceleration
- Security Visibility
- SupremeRAID™ Integration
GP-SPARK + SupremeRAID™ AE
Combining futuristic AI compute with enterprise security to build distributor-grade enterprise solutions. A clean, bright and minimal brand identity that evokes the modern feel of high-performance technology.
Manuals & Drivers
Drivers
Not available yet
GP-SPARK - Core Technology Features
Extreme Performance & Low Latency
GP-Spark paired with SupremeRAID™ AE combines 100GbE, RoCEv2 and NVMe-oF high-speed transport to fully unlock NVMe SSD performance, delivering up to 7.5GB/s sequential read, 2.7GB/s sequential write and 1.9M random read IOPS, with low-latency access and RAID data protection.
0.58L Desktop Dimensions
The entire system fits within an ultra-compact 150mm × 150mm × 26mm form factor, completely condensing traditional enterprise high-performance storage into a desktop footprint.
Premium Vapor Chamber Cooling
Features a one-piece aluminum-magnesium alloy enclosure acting as a high-efficiency heatsink, paired with an intelligent silent fan to ensure a quiet office environment even under 100% full workload.
Plug-and-Play Cable Direct Attach
Supports direct connection to NVIDIA DGX Spark via DAC or AOC cables. The host automatically discovers storage drives for high-speed GDS access without formatting.
Enterprise-Grade RAID Data Protection
A single unit accommodates 4x M.2 NVMe SSDs. Paired with SupremeRAID™ AE it supports RAID 0/1/5/6/10 with no additional RAID card, providing reliable data protection for an external KV cache while reducing extra hardware, cabling and cooling requirements.
Green & Low Power Consumption
Upholding LUISUAN's core philosophy of green and low-carbon tech, the system delivers massive data throughput while capping maximum operational power consumption at just 100W.
GP-SPARK - Technical Specifications
| Item | Detailed Parameters & Descriptions |
|---|---|
| Node Type | Data Center Class NVMe-oF Storage Node |
| Hardware Offload Engine | Full-line ASIC/DPU Protocol Offload (RDMA RoCEv2 & NVMe-oF chip-hardened) |
| Maximum Random I/O (IOPS) | Up to 270M IOPS via a single 100 GbE port |
| Data Throughput | 12 GB/s (Single Host Direct-Attach) / Up to 24 GB/s (Dual-Host Aggregated) |
| End-to-End Latency | Ultra-low latency below 20µs |
| Storage Drive Support | Supports 4x M.2 NVMe SSDs (Up to 4TB per drive, highly scalable configuration) |
| Dimensions & Volume | 150mm × 150mm × 26mm, ultra-compact 0.58L desktop storage accelerator |
| Chassis Material & Cooling | One-piece Aluminum-Magnesium alloy cooling enclosure + active silent fan |
| Power Consumption | Energy-efficient low-carbon architecture, maximum power consumption capped at 100W |
| Network Connectivity | 100 GbE interface (Supports DAC / AOC Direct Attach Cable) |
| Host Acceleration Support | Supports NVIDIA™ GPUDirect™ Storage (GDS) direct access, bypassing host CPU |
| Supported Operating Systems | Ubuntu 18.04+, CentOS 7.5+, KylinV10 SP1+, Rocky Linux 9+ (Linux Kernel 4.13.0+ compatible) |
| Deployment Compatibility | Plug-and-play for NVIDIA DGX Spark, High-Frequency Trading, and High-Performance Workstations |