Chinese GPU Maker, Innosilicon, Unveils The Fantasy 2 GPU With A Whole 1.5 TFLOPs of Compute Power
- up to 10Gbps
- up to 102.4GB/s
— Löschzwerg (@Loeschzwerg_3DC) August 4, 2022
The previous generation 1 GPU made its debut at the end of 2021, offering up to 10 TFLOPs Compute power. For comparison, the Fantasy 2 GPU offers 1.5 TFLOPs so we can tell that this chip is already slower than its predecessor. But unlike the 1, the Fantasy 2 GPU isn't designed to be a fast card as it serves a completely different market segment.
Innosilicon's presentation stated that the Fantasy 2 GPU is designed to be used in low-power devices. The focus market for the new GPU is set for laptops, desktops, and embedded industrial systems.
The new Fantasy 2 offers 1.5 TFLOPs of FP32 or Single-Precision Compute. The power consumption remains low — between the 4W to 15W range, which is almost a five percent difference compared to the Fantasy 1. The chip will also offer 12.5 TOPs (INT8) processing power and offers a pixel fill rate of just 48 GPixel/s which is vastly lower than the 160 GPixel/s of the 1GPU.
As for the specifications revealed for the GPU, do not indicate that physical graphics cards with the Fantasy 2 GPU will be manufactured. The information suggests that fanless cooling options will be compatible with the Fantasy 2, an ideal solution for industrial embedded systems.
The Innosilicon Fantasy 2 GPU will be fitted with LPDDR5X/LPDDR5/LPDDR4X/LPDDR4 memory modules with transfer rates of up to 10 Gbps, offering up to 102.4 GB/s of bandwidth. Memory capacities include 2 GB, 4 GB and 8 GB flavors. The Fantasy 2 GPU will offer support for DirectX 11 API and will allow for that compatibility on all well-known operating systems. Since these GPUs are designed for the Chinese domestic market, we won't hear about them in other market regions so it remains to be seen how well the product is adopted by Chinese users.