We’ve talked a lot about this GPU and its two-die variant, which could even compete with a full RTX 3060 (in theory). What is leaked today leaves so many unknowns or perhaps more than just before knowing the existence of the revealed video, since the Fenghua Type-A does not seem to be what was promised.
GFXBench is the first test for this GPU
Recall that there are two aspects that are nothing more than a monodie GPU and a multi die one, where both include the same features per chip and PowerVR architecture. Type A will have 5 TFLOPS on FP32 and should supposedly match the RTX 2060, while the version Type-B with two dies it achieves twice the performance and should be right on the same level as the RTX 3060.
Both have support for PCIe 4.0 and some really incredible consumptions if we take into account that they are manufactured to 12 nm, since it is ensured that version A would only consume 50 watts, while the double die would reach the 100 watts only, so if the performance materializes we would be talking about the best GPUs in energy efficiency per FPS.
In addition, today we know that although they will arrive in several variants of VRAM (4GB, 8GB, and 16GB respectively with the 32 GB version for the dual die) they will do it with GDDR6X at 19 Gbps with a bandwidth of 304 GBps, curiously 10% below what an RTX 2060 achieves, which is strange given that current GPUs are tremendously dependent on this parameter
Fenghua 1 in benchmark
Innosilicon graphics card. pic.twitter.com/lyMpQAo5Km
– Stewie (@Stewrandall) December 23, 2021
What is seen in this video only leaves more doubts about the performance, although it is also true that there could be several explanations that could somehow justify what was seen. First of all, it must be said that the benchmark does not have an FPS counter and there is no software running behind that overlaps any value.
The only thing on the screen is a sense of fluidity that leaves a lot to be desired. Keep in mind that this benchmark is designed for mobile phones and that a GPU such as Type-A (version used for the test) should achieve values above 300 FPS Without too many problems, since the RTX 2060 with which it is compared by TFLOPS is giving similar values to the figures described.
The feeling we have when watching the video is that the smoothness is not what it should for such an extremely high framerate, but here more doubts are raised. Is it really monitor problem? Is it an effect created by the difference in FPS the camera can capture? We do not know and it is not clear at all because final or partial data or values are not offered.
Until then we will not be able to place this new GPU on the world stage. What we can say is that many companies are already receiving engineering samples to evaluate not only the performance, but the potential of these GPUs, so it is a palpable reality that before it was simply data on paper.
