Innosilicon presents the GPU of Fengua n ° 3 in China, but the allegations of Cuda compatibility will have calls from Nvidia his lawyers
- The Innosilicon of China launches the GPU of Fengua n ° 3 with the integration of the RISC-V processor
- Fengua n ° 3 claims Cuda compatibility, 112 GB HBM memory and tracing of the shelves
- New GPUs targets AI workloads, games, CAD, medical imaging and 8K screens
Innosilicon Technology introduced a new GPU in China which combines a RISC-V processor with a graphical processor affirming compatibility with the Cuda platform of Nvidia.
The Fengua n ° 3 (or Fantasy III as English translation on the chip), the last flagship product of the company, is equipped with more than 112 GB of HBM memory and targets a wide range of IT workloads, from AI to the game.
Unlike its predecessors, which were based on Powervr IP, Fengua n ° 3 relies on RISC-V architecture and is described by society as a fully cultivated design.
Cuda compatibility
During the launch, the representatives avoided giving detailed specifications, but underlined the care of the chip for several applications, including scientific IT, CAD and commercial entertainment.
Fengua n ° 3 was also presented as the first GPU capable of supporting the native dicom for medical imaging without specialized displays.
Cuda compatibility is easily the most striking claim, as technology is the owner of Nvidia, and little rivals have tried to reproduce compatibility.
If the claims are correct, Fengua n ° 3 could be attractive for researchers and developers who are already counting on the Nvidia ecosystem, but I cannot imagine that Nvidia does not take a form of action to protect its platform and intellectual property.
In the events, the card was shown during the execution of games such as Tomb Raider and Valurant, but no performance benchmark, resolution details or image frequencies has been provided.
For professionals, the card supports modern APIs, including DirectX 12, Vulkan 1.2 and OpenGL 4.6, as well as the material -based tracing. It is compatible with Windows, Android, UOS, Kylin and other operating systems.
The configuration of the GPU memory is well suited to AI applications and a single card is supposed to manage the language models from up to 72b parameters, while the systems with eight cards would have evolved at 685b.
The company also claims compatibility with several AI models developed by Chinese, pointing ambitions in the country’s national AI sector.
Innosilicon claimed the support of the 8K resolution in six monitors, as well as the YUV444 color format, a functionality intended for design and video professionals.
China has worked on self-sufficiency in semiconductors’ technology, and products like Fengua N ° 3 are another example, although the real performance of the card remains to be seen.
Via Ithome
You might also love
