Specifications
Model
DX-H1 V-NPU
Vendor
DEEPX
AI Performance
50 TOPS (INT8)
Architecture
Dual hardware video codecs plus dual NPUs on one card
Video Decoding
H.264 / H.265 - 64 channels at 1080p 30 fps
Video Encoding
H.264 / H.265 - 32 channels at 1080p 30 fps
Host Interface
PCIe Gen3 x16 physical slot with x8 lanes
Memory
24 GB LPDDR5 at 6000 MT/s
Power Consumption
30 W
Thermal Solution
Heatsink - fanless
Form Factor
Low-profile PCIe card
Dimensions
68.1 x 164.3 x 18.01 mm (with heatsink)
OS Support
Windows 11 / 10
Linux Support
Debian-based Linux - Ubuntu 24.04 / 22.04 / 20.04 LTS
Additional Platforms
Yocto Project and Docker
AI Frameworks
Ultralytics, PyTorch, TensorFlow, ONNX, Keras
System Support
x86 and ARM-based architectures
Density
Up to 4 cards per 2U server - 256 video channels
TCO
82% lower hardware cost and 85% lower electricity cost (1,000-channel system, 5-year basis)
Video Search
Supports natural-language video search via on-card AI models
Overview
The DEEPX DX-H1 V-NPU combines AI inference and the full video pipeline on a single low-profile PCIe card. Its dual hardware codecs decode H.264/H.265 at 64 channels of 1080p30 and encode at 32 channels, while dual on-card NPUs deliver 50 TOPS of INT8 inference - removing the separate decoder cards and CPU-side video handling that a conventional GPU-based video analytics stack requires.
The design is aimed at the TCO problem in video surveillance and smart-city deployments. DEEPX quotes 82% lower hardware cost and 85% lower electricity cost on a 1,000-channel system measured over five years. Because there is no fan and the card is low profile, up to four DX-H1 V-NPU cards fit in a single 2U server for 256 channels of processing, and event-only transmission to the VMS cuts bandwidth and storage costs.
Key Benefits
All-in-one architecture. Dual codecs and dual NPUs on one card eliminate separate video decoders and minimise host CPU overhead, keeping 24/7 operation stable. Rack density. Fanless low-profile cards allow up to four per 2U server - 256 channels per server. Dramatic TCO reduction. 82% lower hardware cost and 85% lower electricity cost on a 1,000-channel five-year basis. Bandwidth savings. Local event analysis means only critical events travel to the VMS, and natural-language video search turns days of manual review into seconds.
Applications
Video surveillance and VMS deployments, smart city camera networks, retail and campus analytics, industrial safety monitoring, natural-language video search, and high-density channel-count edge servers.
Request a Quote — DEEPX DX-H1 V-NPU — 50 TOPS AI INFERENCE WITH DUAL VIDEO CODEC PCIE CARD
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →