Fenghua, this is what Xindong / Innosilicon called their fastest GPU to date, which translated into our language would be Fantasy 1, or Fantasy 1 as the English like to call it. This card aims very high on the market so as not to have TOP functionalities and also sets itself rather interesting objectives if one takes into account its options and its characteristics: could it correspond to an RTX 3060?
The company has been pretty concise in terms of architecture and chips, so discerning whether or not it’s faster than the competition is the least risky on our part given the lack of data available. We know that it will have two different chips and that these mark two rather curious different performance options.
Fantasy 1 Type A and Fantasy 1 Type B
Imagination Technologies is undoubtedly the linchpin of these two GPUs since they both integrate the PowerVR architecture. Under this premise, we better understand the two types of chips that Xindong / Innosilicon arranged:
- Fantasy 1 Type A: corresponds to a single die GPU which has a performance of 5 TFLOPS in FP32 and 25 TOPS INT8 for artificial intelligence and ML.
- Fantasy 1 Type B: it’s a dual-matrix GPU in true AMD and NVIDIA style that achieves performance FP32 of 10 TFLOPS and 50 TOPS in INT8.
To give us an idea of where the two GPUs are going in the gaming market, the Fantasy 1 Type A would give similar performance to the current RTX 2060, while the Type B would overtake an RTX 3060. We agree on that. . TFLOPS in FP32 are just one more metric for years that shouldn’t be taken as a benchmark, but there is the data and performance that needs to be seen.
In terms of VRAM configurations, Fantasy 1 Type A will be equipped with three different options: 4 GB, 8 GB and 16 GB, which can be GDDR6 or GDDR6X. Instead, Fantasy 1 Type B would go so far as 32 GB obviously, where its speed and capacity will be necessary to be able to host the data of the two GPUs under PCIe 4.0.
A ridiculous consumption for the power acquired
| name | Fantasy 1 Tipo A | Fantasy 1 Type B |
| FP32 | 5 TFLOPS FP32 (1 GPU) | 10 TFLOPS FP32 (2 GPU) |
| INT8 | 25 TOP | 50 TOP |
| Pixel rate | 160 GPixel / s | 320 GPixel / s |
| Video | 4x4Kp60, 16x1080p60, 32x720p30 | 8x4Kp60, 32x1080p60, 64x720p30 |
| Go VRAM | 16 | 32 |
The most curious thing about the two GPUs is their consumption, because made with a 12nm lithographic process, it is not clear how they are able to achieve 50W and 100W respectively for this FP32 performance.
That is to say that these are cards much more powerful than those of their rivals, it would be necessary to add much more, to the point that they are impressive if these data are reflected after successive reviews. Likewise, neither the Fantasy 1 Type A nor the Fantasy 1 Type B are currently commercially available and there is a possibility that they will not do so until they are properly debugged in terms of drivers, so it is assumed they would hit data centers given their capabilities, virtualization, and IT.
Of course and as expected, there isn’t a single mention or rumor of pricing coming in, so we’ll have to quietly wait for someone to take the glove and at least confirm performance and consumption to see if these are. real options present and future against NVIDIA, AMD and Intel.
