One more post about AMD's efforts to get into deep learning. Some photos were published that show Radeon Technologies Group head Raja Koduri holding a Vega-based cube consisting of four Vega prototype boards. These pictures were taken last week at AMD's secret press meeting in California.
The device was more of a mockup and didn't have any Vega GPUs installed. VideoCardz reports there's very little information about this cube, other than that it could provide up to 100 teraflops of FP16 computing power for AI tasks. It's unknown if AMD has any plans to actually launch it, it wouldn't be the first time the company shows off something impressive but fails to deliver (think Project Quantum).
The ‘Koduri Cube’ (not the actual name) would deliver 100 TFLOPS of half-precision computing power (4x 25 TFLOPs). This deep-learning device has a special interface, which was not documented or officially introduced yet. We suspect this could be AMD’s response to NVIDIA’s NVLink, but unlike Tesla P100, it would be installed vertically.