Coffee Lake based edge AI computer supports Nvidia Tesla T4

Neousys’ rugged “Nuvo-7166GC” edge AI inference platform features 8th or 9th Gen CPUs with 2x PCIe x16 slots for an Nvidia Tesla T4 GPU plus mini-PCIe, 2x M.2, 6x GbE with optional PoE+, and MezIO modules.

Like last year’s Nuvo-8208GC, the Nuvo-7166GC is a powerhouse edge computer featuring Intel’s 8th or 9th Gen Coffee Lake CPUs and PCIe expansion. Described as a ruggedized edge AI inference platform, the Nuvo-7166GC offers up to 65 TFLOPS of FP16 compute and 130 TOPs in INT8 inference AI acceleration for voice, video, image and recommendation services. It’s especially suited for “deep learning-based medical imaging to provide diagnostic insights,” says Neousys.



Nuvo-7166GC, front and back
(click images to enlarge)

The Nuvo-7166GC is equipped with 2x PCIe Gen 3 x16 (8x lanes), one of which supports a 70W Nvidia Tesla T4 GPU card. By comparison, the earlier Nuvo-8208GC provides dual x16 slots for two more powerful, 250W Nvidia cards, as well as 2x PCIe Gen 3 x8 and a single-lane PCIe Gen 3.1 x4.

Like that larger system, the 240 x 225 x 110.5mm Nuvo-7166GC provides Neousys’ patented Cassette module technology for optimized cooling. In this case, it’s designed specifically for cooling the Tesla T4 with an air tunnel design that guides intake air through the card’s passive heat sink. An additional fan cools a second PCIe card, such as a sensor or image acquisition card. The design supports -25 to 60°C operation with 35W Coffee Lake CPUs and -25 to 50°C with 65W parts, both with 100% CPU and GPU loading, claims Neousys.



Nuvo-7166GC (left) and Cassette cooling system
(click images to enlarge)

Other ruggedization features include shock resistance per MIL-STD-810G, Method 516.6, and vibration per MIL-STD-810G, Method 514.6. There’s also 10%-90%, non-condensing humidity tolerance and EMC protection per EN55032 and EN55024. There’s an 8-35VDC terminal block input and a remote power terminal block with an LED output, as well as an optional 280W…

Source Link