The Chinese company Innosilicon showed a graphics card of its own design based on the GPU Fantasy 2 presented last summer during the local annual China Chip Excellence Awards Ceremony. The accelerator is equipped with a passive cooling system.
The manufacturer did not say when the specified graphics card will go on sale. However there explained Portal Tom’s Hardware, it should have happened back in August of last year. Formally, the graphics card does not even have a name. It received the “Innovative Chinese Product” award at the event and is described in the accompanying documents as “a high-performance graphics accelerator based on the Fenghua 2 GPU with support for three monitors at 4K level”.
The Fantasy 2 GPU at the heart of the card uses the PowerVR architecture and has a low power consumption of 5-15W. According to Innosilicon, the Fantasy 2 chip delivers 1.5 TFLOPs of FP32 performance and 10 TOPS of Int8 processing power. In other words, this decision can hardly be called the ultimate dream of gamers.
For the Chinese GPU, the ability to work with LPDDR5X / 5/4X / 4 memory of 2, 4 and 8 GB is declared, at speeds of up to 10 Gb / s per contact and a total bandwidth of up to 102 GB / s The chip supports the PCIe 3.0 x8 interface, and solutions based on it can provide up to two HDMI 2.0 connectors, one DisplayPort and eDP1.4, one VGA and one LVDS. In addition, the chip supports the OpenGL, Vulkan and DirectX APIs as well as Windows, Linux and Android operating systems.
According to Innosilicon, the graphics card based on the Fantasy 2 GPU is ready for “everyday tasks”. It can be used for office applications, CAD and entertainment. The company claims that the novelty shows “good performance” in games. However, we are probably talking about 2D games.
The Fantasy 2 GPU-based accelerator will definitely not cope with 3D game projects. Graphics cards based on the Fantasy 1 GPU, which are not yet commercially available, would theoretically be better suited for this purpose. The Fantasy 1 series should consist of Type A models based on a single BXT-32-1024 GPU (again on PowerVR architecture) with a performance of 5 TFlops in FP32 operation and based on two BXT-32 -1024 with a performance of 10 TFlops in the same tasks.
The cards should get up to 16GB of GDDR6X memory with support for a 128-bit memory bus. Theoretically, the innovations should deliver performance at the level of the GeForce RTX 2080 or GeForce RTX 3060. Chinese graphics cards were unveiled in fall 2021, but even Chinese profile reviewers haven’t gotten to them yet.