China’s InnoSilicon continues to push forward with its attempts to provide the country with a proud “domestic desktop GPU.” The company launched Fenghua No. 1 last November, but is still learning about the first Innosilicon branded GPUs and isn’t on the list of the best graphics cards. The November desktop GPU still looks like it’s in stealth mode, but with the release of Fenghua No. 2 on August 3, such utility doesn’t stop the company’s PR machines. Hmm.
The central idea behind Fenghua No. 2 seems to be to double the potential performance by freeing up the GPU core. However, this is not done with larger monolithic GPU dies with twice as many cores. Rather, it is intended to hit another identical GPU on the PCB.
AMD and Nvidia GPUs have been testing this path for many years using SLI and CrossFire, but as a gaming solution, it has become unpopular due to inherent issues and complexity. Games typically required support for CrossFire and SLI drivers, and specific game code, yet multi-GPU setups were prone to microstutter issues. Nvidia has almost completely abandoned SLI with the latest generation Ampere GPUs and has NVLink connectors that only require the RTX 3090 and 3090 Ti.
Anyway, let’s go back to Fenghua No.2 GPU. Baseline can be set as Fenghua No. 1, the new dual GPU card accurately doubles performance, 5 to 10 TFLOP FP32, 160 to 320 GPixels / s, AI (INT8) performance from 25 to 50 More and more TOPS. All codec output capabilities are also doubled, but such scaling sounds more theoretical than practical or measured.
The previous report on Innosilicon graphics cards included GeForce RTX 3060 performance level tips for single-die desktop cards. Currently, Fenghua No. 2 is expected to double its performance. Add twice as much salt.
The main specifications of Fenghua No. 1 include using a 12nm manufacturing process for the GPU. It is believed to be based on Imagination Technologies’ PowerVR technology. In addition, Innosilicon equips the No. 1 card with 4GB, 8GB, or 16GB of GDDR6 or GDDR6X VRAM, and speeds of up to 19Gbps with bandwidths up to 304 GB / s. Graphics cards are likely to support a variety of operating systems, APIs and codecs, and single GPU models are touted as consuming only 50W. (I need more salt.)
One of the areas where Innosilicon is undoubtedly superior to Western graphics tech companies is to prevent Chinese leaks such as performance and benchmarks from forums, websites and social media video platforms. It’s frustrating to see these launches and claims without the hardware falling into the hands of third-party testers for a thorough evaluation.
We hope that next Wednesday’s presentation will include some substantive information and benchmarks that deserve to be reported. I would like to get some of these graphics cards so that I can add them to the GPU benchmark hierarchy. Given the features and specifications, we can only speculate that the actual performance probably doesn’t match the claim.
