Innosilicon Launches Fantasy 3 GPU Featuring 112 GB VRAM, DX12, Hardware Ray Tracing, and CUDA Compatibility

Innosilicon, a prominent GPU manufacturer based in China, has unveiled its latest innovation: the Fantasy 3 GPU. This powerful graphics processing unit boasts a staggering 112 GB of VRAM, specifically designed for demanding artificial intelligence applications, while also being compatible with DX12, HW-RT, and CUDA technologies.

Innosilicon’s Fantasy 3 GPU: A Versatile Powerhouse for AI, Gaming, and Content Creation

The newly launched Fantasy 3 GPU represents the third evolution in the Fantasy series, following the earlier releases of Fantasy 1 and Fantasy 2 that catered primarily to the domestic Chinese market. With this latest iteration, Innosilicon aims to penetrate various sectors, including artificial intelligence, gaming, and content creation.

Innosilicon's Fantasy 3 GPU Launch — Image Credit: Innosilicon

At a recent showcase event, Innosilicon demonstrated the Fantasy 3 as a comprehensive GPU solution that will function seamlessly with an RISC-V CPU. The GPU’s architecture supports various standards, including NVIDIA’s CUDA, DX12, Vulkan 1.2, and OpenGL 4.6, making it a robust option for AI training and large-scale scientific workloads. Its innovative OpenCore architecture enables enhanced performance across gaming platforms and cloud computing environments.

While the details surrounding the Fantasy 3 GPU remain somewhat limited, its 112 GB VRAM is already a significant attraction, particularly for video editing and content creation, given its support for YUV444 color space. The GPU is designed for compatibility across multiple platforms including Windows, Linux, and Android. Innosilicon claims that the Fantasy 3 will efficiently run local 32B/72B Long Language Models (LLMs) and, when deployed in a server with an 8-card configuration, can manage LLMs as large as 671B and 586B, including Qwen 2.5 and Qwen 3 models.

Innosilicon's Fantasy 3 GPU Presentation — Image Credit: Innosilicon

In addition to the Fantasy 3 GPU, Innosilicon showcased its next-generation DDR5 and MRDIMM DDR5 memory solutions designed for server applications, along with a sophisticated 120-channel PCIe Gen5/4 server switch chip. The company eagerly anticipates the upcoming launch of the Fantasy 3 GPUs and also provided live demonstrations of a game utilizing DX12 and HW-RT support running effectively on the new hardware. Notably, the company alluded to the possibility of releasing variants of the Fantasy 3 with over 112 GB of VRAM, contingent on market demand.

Innosilicon’s intellectual property portfolio boasts a diverse array of high-speed interface subsystems. Their key offerings for the next-generation AI sector include LPDDR6/5X Combo memory, GDDR7/6X, MR DDR5, UCIe Chiplet, UALINK, and state-of-the-art PCIe 6.0/5.0 solutions, among others.

“With one of the industry’s most advanced memory interface solutions, Innosilicon empowers customers to overcome the memory wall and accelerate leadership along the high-bandwidth AI innovation curve. It has supported over 300 top-tier global customers, contributed to more than 10 billion SoCs, and delivered silicon-proven IP across advanced process nodes at leading foundry TSMC, ranging from 28nm down to 3nm technologies including 28nm, 22nm, 16/12nm, 7nm (including N6), 5nm (including N4), and 3nm processes.”

“To tackle key memory bottlenecks in AI workloads, Innosilicon introduced its cutting-edge LPDDR6/5X Combo PHY + Controller IP, fully compatible with TSMC’s N6 and N3 process technologies. At the TSMC North America OIP Ecosystem Forum, the company also will present its latest paper, ‘Shaping the Future of Memory Interfaces: LPDDR6/5X Combo Subsystem for AI, Security, Automotive, and Beyond, ’ highlighting its leadership in next-generation interface solutions.”

“The Innosilicon dual-protocol LPDDR6/5X Combo achieves peak performance of 14.4Gbps at core voltage in LPDDR6 mode. This high-speed capability suffices for numerous application scenarios with stringent data transmission speed requirements.”

Moreover, Innosilicon plans to further exhibit its high-speed interface IP and advanced SoC solutions at the upcoming 2025 TSMC OIP Ecosystem Forum. For more details on these developments, you can check out this article.

Source & Images