Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/chinas-xiangdixian-unveils-5nm-fuxi-gpu-at-iccad-expo.24145/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2030770
            [XFI] => 1060170
        )

    [wordpress] => /var/www/html
)

China's Xiangdixian unveils 5nm Fuxi GPU at ICCAD Expo

Fred Chen

Moderator

Chinese GPU Maker Demos Its Next-Gen Fuxi A0 GPU Based on Imagination DXD GPU IP: Features 5nm Core, 12 GB VRAM, Ray Tracing & Super-Res Support​

Hassan Mujtaba•Nov 30, 2025 at 03:20am EST

The Chinese Fuxi A0 GPU is based on a 5nm Imagination DXD Graphics IP, offering Ray Tracing & Super Resolution support.

China's Fuxi A0 GPU Is Ready For Mass Production: 5nm, Ray Tracing & Super-Resolution Ready With Imagination DXD GPU IP.​

China-based GPU company, Xiang Dixian, has unveiled its brand new product, the Fuxi A0 graphics card. The graphics card was put up on display and was even demoed during the recent ICCAD Expo held in Chengdu, China. The most interesting part about this graphics card is that it leverages a next-gen DXD GPU IP from Imagination Technologies.

Imagination Technologies made a brilliant appearance at the 31st ICCAD-Expo 2025 exhibition held in Chengdu recently. As an important partner, Imagination Technologies showcased its new generation of GPU graphics cards independently developed based on the Imagination DXD architecture.

As the only mass-produced IMG DXD architecture product currently available, this graphics card boasts more than double the overall rendering performance of its predecessor. It can not only run large-scale 3D graphics applications smoothly, but also demonstrated various complex 3D digital twin demos on-site, while simultaneously showcasing next-generation graphics processing technologies such as ray tracing and super-resolution.


via Xiang Dixian

Imagination Technologies has various graphics IPs such as DXTP, DXS, DXD, and DXT. The closest one to a standard desktop-tier architecture is the DXD GPU IP, which has already been utilized by Chinese graphics manufacturers such as Innosilicon. Now, Xiang Dixian has showcased its own take on the Imagination GPU IP, leveraging a brand new architecture based on a 5nm process technology.

As per previous reports, the Fuxi chips are expected to produce up to 160 TFLOPs of compute power and offer 12 GB of memory. From the pictures, we can see that the graphics card features a dual-slot and dual-fan design. The PCB seems to offer 4 Physical DRAM sites, though we cannot confirm from the one picture that is available.

A graphics card with two cooling fans, displaying a 'FUXI_AO 4PHY' label on the circuit board.
Based on initial specifications, the Fuxi GPU series will come in two flavors, the A0 and B0. The Fuxi A0 was demoed at the expo, and the company claims it offers performance rivaling that of overseas graphics cards while also supporting features such as ray tracing and super resolution (upscaling). The A0 variant will focus on rendering and gaming, while the B0 variant will combine the GPU with an NPU that offers FP8 capabilities.

The chipmaker also stated that the A0 model is capable of running Black Myth: Wukong at a very good frame rate. There was a demo being shown at the booth, which featured a ray-traced scene, and the GPU managed around 35 FPS on average.


Chinese GPU Maker Demos Its Next-Gen Fuxi A0 GPU Based on Imagination DXD GPU IP: Features 5nm Core, 12 GB VRAM, Ray Tracing & Super-Res Support 2
Xiang Dixian has a long history of producing graphics cards based on previous Imagination Technologies IPs, such as the XDX 121, XDX 151 for desktops, XDX R1900 for workstations, and XDX X1900 for servers. So with this new offering, the company is going to offer an even faster and modern graphics card with the latest Imagination Technologies GPU IP. We expect to see a launch sometime next year as the GPUs are expected to enter mass production soon.


 
Wow, that's pretty good:

RTX Pro 6000 is rated for 120 TFLOPS at FP32: https://www.nvidia.com/en-us/data-center/rtx-pro-6000-blackwell-server-edition/

Though for AI use, the Fuxi A0 may still be a bit limited with 12GB of memory. I can't determine if the HBM2 memory is stacked or not, but at 12GB probably not as much bandwdith as the 512-bit GDDR7 on the RTX Pro card.

Pro 6000 uses a 750mm2 die on a (presumably) better process. The Fuxi card OTOH also doesn't look very large - indicating <250W maximum power. I suspect the 160 TFLOPS number is bogus, but I'm open to being surprised here too.
 
Single 8-pin connector suggest TDP under 75+150W. RTX card is up to 600W. :D


Imagination claims 2304 Flops/clk, which would suggest 72 unit with 32 ALU each. That would mean 2.304 TFlops. So they would need 70 chiplets/superblocks or something to achieve that performance.

It is bit tricky but i think we could estimate size. Mobile GPUs are typically around 20mm2. So we can assume that Imagnation is targeting similar size. That would make 1,400 mm2 (20*70). Probably much more since there must be more robust uncore.

SoC Manufacturing Process Total SoC Die Size Estimated GPU Die Area Performance (GFLOPS / FP32)

MediaTek Dimensity 9400 3 nm (TSMC) 125-130 mm² (est.) ~22 mm² (est.) 4952.1

Snapdragon 8 Elite (Gen 4) 3 nm (TSMC) 125-130 mm² (est.) ~23 mm² (est.) 3379.2

Apple A18 Pro 3 nm (TSMC) 105 mm² ~21 mm² (est.) 2289

Google Tensor G5 3 nm (TSMC) 121 mm² Not publicly available 1689.6

Snapdragon 8 Gen 2 4 nm (TSMC) ~118 mm² (est.) Not publicly available 2089 (Standard) / 2208.8 (for Galaxy)

Apple A16 Bionic 4 nm (TSMC) ~110 mm² (est.) Not publicly available 1789.4

Exynos 2200 4 nm (Samsung) ~104.7 mm² (est.) Not publicly available 1382.4 (est.)

Snapdragon 888 5 nm (Samsung) ~124 mm² (est.) ~18 mm² (est.) 1152 (est.)


Xiang Di Xian, which had developed earlier GPUs such as the XDX 121, XDX 151, XDX R1900, and XDX X1900, continues expanding its graphics portfolio with the Fuxi A0.

However, in September 2024, we and multiple outlets reported that Xiangdi Xian/Xiang Di Xian faced a cash crunch, terminated approximately 400 staff, faced asset freezes and lawsuits, and discussed restructuring, while seeking new investors.

Since then, we have not found any credible filing or announcement indicating fresh financing, a completed restructuring, or a workforce rebuild. What has appeared is product-side noise: Reports that the firm demonstrated the Fuxi A0 GPU (at ICCAD 2025). Those demo reports don’t include funding or balance-sheet details.

Bottom line: As of the latest reporting, the company has not publicly shown that it fixed the September-era financial problems; at best, it’s still operating in some capacity (trade show demo), while its capital status remains unclear. Bring in the smoke and mirrors.
 
Last edited:
Back
Top