A New Architecture That Outperforms Traditional CPU + GPU Systems by a Wide Margin
SANTA CLARA, Calif. and SEOUL, South Korea, Dec. 8, 2025 /PRNewswire/ — DEEPX, a leader in ultra-low-power on-device AI semiconductors, announced on the 8th the launch of a next-generation AI video analytics platform engineered to break the limits of GPU-centric designs. The solution, which combines DEEPX accelerators and Ampere CPUs, reflects a fundamentally new compute architecture—one that delivers far higher density, lower power consumption, and dramatically superior cost efficiency for large-scale Vision AI deployments.
DEEPX– Announces Ultra-Efficient AI Video Analytics Solution Based on AmpereOne® Platform
The new system fuses DEEPX’s DX-H1 Quattro AI accelerator with Ampere’s high-efficiency CPUs, creating an optimized pipeline purpose-built for modern VMS requirements. As the global video management market shifts from storage-focused operations to real-time AI analytics, the industry is confronting the structural bottlenecks of GPU architectures: high heat, excessive power draw, and poor scalability. The DEEPX platform decisively eliminates these limitations, marking a clear departure from conventional GPU-only approaches.
AmpereOne® CPUs handle the rapid decoding of hundreds of high-resolution CCTV streams, while DEEPX’s DX-H1 Quattro is dedicated exclusively to high-density AI inference. This division of labor unlocks extraordinary throughput: a single server can analyze 200+ cameras simultaneously and sustain over 8,000 real-time AI inferences per second. With scalable performance beyond 400 TOPS, the platform is uniquely capable of powering mission-critical infrastructure such as smart cities, video surveillance logistics hubs, and industrial control systems.
The system’s power efficiency is equally disruptive. Whereas traditional CPU + GPU servers routinely exceed 1,200W to achieve similar workloads, the DEEPX platform maintains total system power between 400W and 750W, depending on configuration. This results in a step-change reduction in operational expenses for 24/7 monitoring centers—an increasingly urgent requirement as cities and enterprises scale AI deployments globally. The AmpereOne® architecture delivers exceptional decoding performance with minimal heat and exceptional density, enabling a single server to achieve what previously required entire racks of GPU hardware
Built to support the full spectrum of Vision AI workloads—including YOLO-based object detection, behavior analysis, CLIP-based semantic search, and real-time RTSP encoding—the solution offers unified analytics from edge deployments to hyperscale data centers. With compatibility across major server OEMs such as ASRock Rack, NEXT Computing, Gigabyte, System76, and Supermicro, and seamless integration with leading VMS platforms including Network Optix, ZoneMinder, or your own, customers’ deployment can begin immediately.
Lokwon Kim, CEO of DEEPX, stated, “The combination of the DX-H1 Quattro and AmpereOne® is redefining what’s possible in the global VMS industry. This is the platform that markets facing extreme cost and power pressures have been waiting for.” He added, “We intend to aggressively target multi-billion-dollar markets—including smart cities, video surveillance, transportation, and industrial automation.”
“Ampere processors are engineered to handle intensive analytics workloads and continuous data ingestion while maintaining exceptional power efficiency,” said Jeff Wittich, Chief Product Officer at Ampere. “That combination is crucial for customers deploying secure, reliable video infrastructure across everything from major enterprise deployments to edge-based systems.”
DEEPX plans to accelerate initial deployments across video surveillance, large-scale smart city programs, infrastructure monitoring for manufacturing and logistics, and real-time traffic and security control. Beginning in the first half of 2026, the company will introduce the platform to global enterprise customers through a series of international exhibitions, technical webinars, and roadshows.
