Innosilicon Fantasy One GPU: up to 32GB of ultra-fast GDDR6X memory

Innosilicon has just held its "Fantasy One GPU Product Press Conference" where it unveiled the new Fantasy One GPU family, and a few interesting new graphics cards.

15

Starting with the Innosilicon Fantasy One GPU itself, which is based on a multi-chip (chiplet) design and flagship dual-GPU graphics card in the form of the new "Type-B" dual-GPU solution uses 2 x Fantasy One GPUs and are connected using Innosilicon's just-announced Innolink interface.

Innosilicon's dual-GPU card has up to 10 TFLOPs of compute performance, and can handle 32 simultaneous 1080p 60FPS streams, or up to 64 streams at 720p 30FPS. The dual-GPU beast packs 32GB of GDDR6X memory, but they're super limited by the 128-bit memory bus on each of the respective Fantasy One GPUs.

• Read more: Innosilicon's new Fenhua No.1 high-end GPU: GDDR6X, PCIe 4.0, HDMI 2.1

15

15

Pictures: This is the Innosilicon "Type-B" card, featuring the dual-GPU "Fantasy One" chips (above) while (below) we have the "Type-A" single GPU card.

15

15

Under that, Innosilicon has the "Type-A" which is a consumer/workstation graphics card, using a single Fantasy One GPU (still a multi-chip or chiplet) with up to 5 TFLOPs of single-precision compute performance, up to 16GB of GDDR6X memory on the same tiny 128-bit memory bus.

The GDDR6X used by Innosilicon on its new Type-A and Type-B graphics cards is clocked at 19Gbps, something worth noting as NVIDIA and Micron have been using GDDR6X exclusively until now. Innosilicon has reportedly put considerable time and research into PAM4 signaling, and was able to drive the GDDR6X modules to 19Gbps... but you're still limited by that puny 128-bit memory bus.

Innosilicon has used a bunch of the most popular API supported on the Fantasy One GPU, including OpenGL, OpenGL ES, Open CL, Vulkan, and DirectX. The inclusion of DirectX means that this could work on Microsoft's Windows operating system, but there is no confirmation of that from Innosilicon (or Microsoft) just yet.

The company did tease that it is working on next-gen Fantasy Two and Fantasy Three (or is that Fantasy 2, and Fantasy 3) for next year on the new 5nm process technology.

While the Chinese company was forthright in detailing (at least most of it) with the Fantasy 1 (or is that Fantasy One, hah) where we don't know what node process the GPU is made on, but know about what the two next-gen GPU node processes will be. Righty-o.