Aitech expands adds rugged GPGPU AI supercomputer
Based on Nvidia’s Xavier AGX , the A178 Thunder expands the general purpose graphics processing unit (GPGPU) systems available from Aitech.
According to Dan Mor, GPGPU product line manager of Aitech: “GPGPU uses a parallel structure, with multiple small cores that process multiple tasks simultaneously. As artificial intelligence (AI) continues to grow and system size continues to shrink, computing systems will be expected to perform in increasingly remote, harsh environments.”
The A178 Thunder has twice as many CUDA cores as Jetson TX2-based systems as well as the addition of new Tensor cores. It is claimed to have some of the most powerful processing capabilities in a small form factor (SFF) system. The system features 512 CUDA cores and 64 of NVIDIA’s new Tensor cores.
Tensor cores in the Xavier system on module (SoM) amplify the matrix processing of large data sets for AI, by enabling higher levels of computation with lower power consumption. When the SoM is coupled with Aitech’s rugged computing expertise, the A178 Thunder handles up to 22 Tera operations per second (TOPS) to provide local processing of high volumes of data where it is needed, closest to system sensors.
The A178 Thunder incorporates two dedicated Nvidia deep learning accelerator (NVDLA) engines that provide an open-source inference design method. The system also features pre-installed Linux OS, which includes the bootloader, Linux kernel, NVIDIA drivers, an Aitech BSP and flash programming utilities.
The A178 Thunder weighs less than 2.4kg/5.5lbs and measures 215 x 81 x 172mm (8.5 x 3.2 x 6.8-inch). It is SWaP-optimised says Aitech and suitable for a number of military and defence applications, such as distributed computing and signal processing in next generation autonomous vehicles as well as surveillance, targeting and EW systems.
Seven pre-set power modes are standard on the A178 Thunder and users can create custom pre-sets with specific clocks and online cores. It also provides a wide input voltage range of 11 to 36V DC with steady state operation. Input power is protected by an EMI/RFI input filter as well as reverse polarity protection.
I/O options include Gigabit Ethernet, 10GBase-T, USB 3.0, HDMI output, UART Serial and CANbus as well as discrete and display ports. Video is simultaneously captured on up to four channels via SDI (SD/HD) or over eight channels at full frame rates using Composite (RS-170A [NTSC]/PAL) interfaces. There is also a high speed, internal NVME SSD with a storage capacity of up to 1Tbyte.
An evaluation board – the EV178 – is also available.
The A178 Thunder will be on display at NVIDIA’s GPGPU Technology Conference (GTC 2019) in Washington DC, USA (5 to 6 November), Stand 411.