The Chinese company Innosilicon, as promised, presented a new generation of its own fantasy graphics processor. As it turned out, the chip Fantasy No.2 turned out to be even weaker in its characteristics than the solutions of the previous generation Fantasy No.1, which the manufacturer presented last fall.
According to Innosilicon, the Fantasy No.2 processor delivers 1.5 teraflops of FP32 performance. For comparison: One of the chips of the Fantasy No.1 series delivers a performance of up to 5 Tflops. And a graphics card based on two such GPUs offers performance of up to 10 Tflops. The manufacturer emphasizes that the advantage of Fantasy No.2 is not in performance, but in power consumption. The TDP of Fantasy No.2 GPUs can range from 4W to 15W. Fantasy No.1 chips have a power consumption range of 20-50W.
The Chinese developer of graphics solutions only announced the technical characteristics of the Fantasy No.2 chip without announcing a graphics card based on it. According to Innosilicon, Fantasy No.2 will be able to work as part of solutions that do not require active cooling, which can be useful for embedded systems, for example.
GPU supports LPDDR5X/5/4X/4 memory 2, 4 and 8GB, which offers speeds of up to 10 Gb/s per contact and aggregate throughput of up to 102 Gb/s. The chip claims to support the PCIe 3.0 x8 interface. Solutions based on this can offer up to two HDMI 2.0 ports, a DisplayPort and eDP1.4, a VGA and an LVDS.
The company also lists 12.5 TOPS of inference performance (INT8) and a fill rate of 48 Gpix/s for the chip, which is significantly lower than the Fantasy No.1 chips.
The Fantasy No.2 GPU is designed for use in desktop and mobile graphics cards as well as in embedded systems.